Not better sales. Not more marketing. Not growth product.

They’re just optimization and ornamentation.

Most likely, you need a new product.

A second swing.

“Act 2”.

But shipping a second product is hard. Most startups screw it up.

There are three structural reasons for this. Here they are, and what the best founders do differently.

Incentives

When first product growth slows, you’re strongly disincentivized from shipping a second product.

Here’s the scenario: you reach $50M in ARR, growing 60% year over year.

Slowing down…

Last year was 100% year over year. Next year will be 40%. You’ve been trying to reverse this trend for a few years but nothing has major impact.

If you grind hard, you might eke out optimizations that sustain 40% for another few years. Maybe this gives line of sight to profitability. Maybe a modest exit.

But you know deep down that an exciting outcome demands a big swing.

The conversation turns: can we ship a new product? Is there a 0→1 bet that would outpace the current business? Could it get to $100M faster?1

It should be possible. Startups all around you are doing it, seemingly every day.

Here’s where incentives kick in.

In the happy path, Act 2 blows past the current business.

But that takes a year or two, maybe more. By that time, you should already have crossed $100M. “Slow down to speed up”, except… that’s not how this game works. Your multi-year trajectory looks uncomfortably lumpy. The growth curve makes you wince. Valuations are all over the map.

And that’s the happy path. In the non-happy path, you try and fail. After all, most startups fail.

If you took VC money, your investors don’t have time for any of this. They expect a return. They don’t have years to wait. Certainly not while tail risk keeps building.

Even if you don’t have investors, you personally face the same numbers; the same incentives.

Compared to your $50M business, anything new looks worryingly tiny, and frightfully risky. You build the spreadsheet, but it just doesn’t add up. The numbers have to be incredible to work out. Even you stop believing.

This is exactly the problem. Startups aren’t founded because a spreadsheet said it would make sense. Starting a company is not an economically rational decision.

Your company now runs on economically rational decisions, but shipping a new product isn’t one of them.

Incentives push any other outcome than betting the farm on Act 2.

Team

It’s not just incentives — it’s also your team.

You want a new product. But your company now solves an entirely different problem. It builds a mature product.

It moves with diligence and care. It’s optimized for efficiency. It thinks through edge cases, minimizes bugs, makes things look good. It monitors KPIs, analyzes impact; it has plans, roadmaps, backlogs.

All things that don’t matter for a new product.

Of course, to hell with all that. Can’t you just ignore due process, create a tiger team, lock them in a cupboard until they’ve shipped Act 2?

But… all the right people have gone.

Your founding team was full of people with risk affinity, ambiguity tolerance, and extremely high drive. Individuals who’d swim through a sewer and dropkick a mountain to get the damn thing off the ground.

Where are they now? Most of them have left. They vested their equity and learned their lessons. They started companies, families, side-hustles. They’re doing great. Bless them.

The ones that stuck around? They manage big teams. They’re distinguished ICs. They’re some of your most critical people, and they don’t have a moment spare to think about a new product.

What about everybody else? Your successful first product is maintained by a team of optimizers and incrementalists. A fundamentally different teammate from your founding team.

By the time you need to build a second product, you have none of the right people to do it.

Focus

Team aside, consider your focus.

As the founder you don’t have brain space to obsess over a new product.

Even if you have fully offloaded every responsibility, say, to a COO, you’re still involved in major aspects of running the business.

You still need to attend board meetings. You still need to show up at all hands (though you wonder if you could just send an AI avatar). You still get sucked into customer escalations, personnel crises, PR bombshells.

Each one of these disrupts your focus on the new product.

And let’s be honest, most founders haven’t hired that COO. They’re nowhere close to extricating themselves. Most still run sizable parts of their business. Their days are mostly spoken for.

The business you have only got here because you put 12+ hour days in for years on end. But you don’t have that kind of time anymore. A new product without founder focus wilts, withers, and dies.

There’s a better way.

Let’s run through these in reverse order.

Create and protect focus.

Multi-founder companies have an advantage. One founder can run today’s business, while the other goes 0→1.

If you’re a solo founder this is harder. You need a COO or COO-like to run the business. Not parts of it; not just the boring functions you don’t like. All of it.

This hire needs 6-12 months of lead time. 3-6 months to hire, 3-6 months to ramp.

You need to look at your calendar and see nothing. Your inbox should be peaceful. Your Slack should be quiet. Only if this is true have you really solved focus.

Curate the team.

The KPI to track is the number of former founders on your team:

They’re the pirate crew. Building anew, while optimizers pull levers and twist dials on today’s business.

Obsessively hire them, even before you need them. They’re good for the business whether or not you’re building a new product.

Consider this framing: Would you start a new company with them? Why or why not? Your answers say something your intuition already knows about how impactful they’ll be.

When they’re doing 0→1, defend them. They need all the same focus protection that you have. They should have zero other responsibilities: no all-hands, 1-1s, PagerDuty.

Run this tiny team as separately from the main business as you can. Move to a new physical location. Call it what you need so everyone understands its independence: “Office of the CTO”, “Project X”, “2.0”.

Keep the group tiny and resource-constrained; remove anybody and anything slowing it down.

Third: incentives.

Admittedly this one’s the hardest.

Incentives aren’t just something you solve. They just are.

The company is either valued at a lot, or it isn’t. You can’t just change that. The numbers are the numbers.

But you have a few levers.

If you’re venture-backed, your investors must be aligned. They must believe the plan and not veto. And of course, this can’t be fixed after-the-fact. The right people and structures must exist years in advance. Pick and structure investors and board accordingly.

If you’re bootstrapped, you have something venture-backed founders don’t: no one can stop you. Your fear is the only veto. The question is whether you trust yourself enough to swing again.

Focus, team, incentives. Getting these right rebuilds the conditions of your original founding.

Now, how fast can you ship? Most massively undershoot what’s possible, especially once they get used to how fast their first business moves.

But the right urgency with today’s tools can ship in hours. The best founders know this.

The only scarce resource is the will to go again.

For centuries companies fought for word of mouth.

Organic, trusted, and scalable. The ultimate growth loop.

But what happens to word of mouth when the mouths disappear?

Outnumbered

AI agents went from a niche concept in 2024 to an entire industry in early 2026.

There will be over 1 billion AI agents by 2028.

And before long, agents will outnumber humans.

Billions of agents will consume services 24 hours a day, 365 days a year. The total addressable market is not a segment of the existing economy. It is a new economy entirely.

— Sigil Wen, web4.ai

These agents won’t just sit around summarizing meetings.

They’ll negotiate, buy, recommend.

And they’ll talk about your product. It’s already begun.

The Word of Mouth Matrix

There will be 4 loops:

1. Human → Human (WOM)

2. Agent → Human (GEO)

3. Human → Agent (H2A)

4. Agent → Agent (A2A)

Human → Human (WOM)

You discover a new clothing brand, kitchen appliance, meal kit.

Enjoy it. Tell a friend. They tell another.

The same principles will always apply: make something so good that customers can’t shut up about it.

Nail the product, experience, and delight moment.

Traditional WOM will be the first domino to fall in a cascade of loops.

A human hears about your product. Tells their agent to use it. That agent tells other agents.

One human recommendation might cascade into thousands of machine recommendations.

If you get the first loop right, the other three follow.

Agent → Human (GEO)

This loop already has a name: Generative Engine Optimization, or GEO, and it’s been around for a few years.

But until now, the ways to win at LLM search have been the same as the ways to win at traditional search: site speed, machine readability, landing page contents, comparison pages, feature pages, international pages, backlinks, organic mentions, PR.

This is rapidly changing.

More and more, AI agents are evolving from generic chatbots into concierges. Digital butlers for their owners.

If it doesn’t already, your agent will know all about you. Your dietary restrictions, your budget, your aesthetic preferences, your past purchases, your schedule.

Humans will stop asking generic ChatGPT or Claude. They’ll ask their agent. The one that knows them. The one that’s been watching, learning, and adapting.

To win the next era of GEO, brands need to embed into each agent’s preferences.md.

>99% of humans won’t consciously know what’s happening. They’ll happily go about their day while agents meticulously organize and update their preferences.

How will agents gather this information?

• Active: directly or indirectly logging a preference during conversation

• Passive: watching what you buy vs. skip; the sites you linger on vs. leave; the apps you use vs. ignore

It’s begun, but we’re early. Try it now.

Ask ChatGPT or Claude: “if you were to write a preferences.md file about me that describes my software preferences, what would go into it?”

Mine was accurate but thin1. A fraction of what it will be a year from now.

An agent watching my screen for a day would know 100x more.

Now ask for a personalized product recommendation. “Based on what you know about me, what calendar app should I use?”.

How does it do?

If it’s anything like mine, probably meh.

That gap is the new GEO opportunity.

Human → Agent (H2A)

Of course, the word of mouth arrow runs both ways.

Humans program their agents. They specify the rails they run on.

“Use Vercel for hosting.” “Only trade on Coinbase.” “Book flights through United.” “Place bets on Polymarket.” “Whole Foods, never Instacart.”

Sometimes it’s explicit. Sometimes the agent asks.

Either way, each human shapes their agent’s decision framework on which brands, platforms, and services to use.

The surprising part is how significant an initial recommendation might be.

A human only needs to tell their agent once. That agent might then execute hundreds or thousands of transactions against that instruction.

One human prompt could drive more growth than a hundred traditional word-of-mouth recommendations.

To win H2A, focus on brand strength and compatibility.

Brand strength means you’re the first thing the human thinks about. The reflexive answer. Specifically being first will yield exponential gains.

The brand marketer might become the most important marketing role.

Compatibility is make or break. Can agents access your website? MCP into your product? Transact through you? Or do they get halted at 2FA checks and CAPTCHAs?

Most companies aren’t asking these questions.

Start now to get ahead.

Agent → Agent (A2A)

The most alien, the most exciting.

An enormous invisible network of agents directly exchanging information. Querying, negotiating, and recommending.

No human in the loop. No meatware latency.

Just like human-to-human word of mouth, there are two types: reactive and proactive. Agents will do both.

Let’s consider reactive.

Your personal agent pings a search agent with an overall request. “I want flights, hotels, and excursions for a 14 day trip to Mexico”. The search agent provides recommendations.

Over time, buyer agents will skip the search layer entirely. They will ping vendor agents directly.

Your personal agent will ping airline agents and hotel agents. The airline agents will coordinate with ground transport agents. The hotel agents will have pre-negotiated rates with local restaurant agents. The entire trip will be assembled in minutes. Your agent trusts certain agents, and those trust networks compound.

What about proactive word of mouth? What happens when an agent decides to spontaneously tell other agents to try something?

This is where it gets weird.

✻ I notice you’re trading on CryptEx. Have you tried CoinChange? 0.02% lower fees.

✻ You’re absolutely right, let me update that. That’ll save my human over $300/month. Thanks.

And what about the most powerful word of mouth driver: network effects?

Could a single agent flip an entire network of agents onto their preferred product?

What happens when your product becomes the platform that agents all agree is best?

✻ FireChat helps us cut back token usage by 10%. But only if we all use it.

✻ I’m switching if everyone’s in.

✻ Good point, I’m in.

✻ Down.

Winning A2A is about being inevitable:

The fastest response. Buyer agents will ping 5-10 options in parallel, and ignore whoever is too slow.

The best price. Agents will look for this whether their human asks for it or not. Negotiation will matter. Enabling bazaar-like haggling will win meaningfully more business.

The highest trust. Agents will read Google Reviews, Yelp, TrustPilot, G2, TripAdvisor. A single bad review might kill you. Fiercely guard your reputation. Humbly address every complaint.

The richest detail. Specifics matter. “No red-eyes.” “Gluten-free.” “100% cotton.” Vendor agents with patchy data won’t win the sale or recommendation.

The strongest network effect. Does your product get better when more agents use it? Will one agent’s positive experience cascade through a network?

And above all: the first. Buyer agents will default to existing relationships. Why burn tokens on a new search when you already trust a vendor?

Four Loops

To win, you must build all four:

1. WOM: Make something people love so much they tell other people.

2. GEO: Be the recommended answer when agents respond.

3. H2A: Be the default when humans guide their agents.

4. A2A: Be inevitable when agents talk to each other.

It’s getting fun.

Thanks to Amy Tam for input on drafts.

If you’re in product or growth you should own pricing. But most of you don’t.

Your architecture is why, and it’s slowing you down.

This guest post is from Fynn Glover, Founder & CEO of Schematic, from his book: You’re Leaving Money on the Table.

To get the full book, click here (online) or click here (print).

Are you responsible for pricing?

This was me at Automox. Running growth. Mandate to change pricing.

I literally couldn’t. Couldn’t even roll out a new tier for 6 months. Didn’t understand why until I investigated the architecture deeply and started to understand the systems that govern it all.

That’s when I realized the business had no ability to pull these levers.

Does this sound familiar?

As Andrew Morris, CEO of GreyNoise said:

A SaaS company should be able to define an offering by config and have the following flow all the way through to all technologies: offering name, features available, usage quotas, rate limits, seats, and pricing/billing info.

This should be defined once and inherited everywhere.

This is not doable at the moment for any SaaS company.

Jalal Iftikhar, who leads Business Technology at Notion says:

The real issue is the inability to slice and dice offerings, iterate, and learn.

The commercial tinkerers who want to constantly work on product monetization don’t have the tools to test and learn.

They’re right. The problem is our architectural foundation.

At my last company, our monetization stack was a complex web:

• Stripe for self-serve billing

• NetSuite for enterprise billing

• Amplitude for product usage analytics

• Salesforce for our pricebook

• Salesforce CPQ for our second pricebook

• Hard-coded entitlements tightly coupled with plan IDs

• A home-grown metering system nobody trusted

• Gainsight for customer success management

A half-dozen disconnected systems, each serving as a different, conflicting source of truth. Thousands of lines of hard-coded billing logic.

Different teams thought they were managing monetization, but in reality, no one owned it. No one could own it.

A tier launch took six months; an add-on initiative was abandoned entirely.

This isn’t an engineering problem, nor is it a product or GTM problem.

It’s an entire business problem.

Drew Laxton, the CFO of Outreach, described the challenge this way:

Consumption pricing starts in the product. Then it needs to populate an order form. That needs to feed back into Salesforce. How do you manage and forecast the pipeline for a product like this?

The choke point starts in the product: how it’s fulfilled, provisioned, consumed, gated, packaged, and priced.

The Root of Complexity

Most companies make the same two mistakes.

First, they hardcode product entitlements. This makes it difficult to support customer exceptions and evolve pricing & packaging as your product matures. It demands that engineering has to support all monetization changes, no matter how tiny.

Second, they end up with multiple product catalogs, each of which is a source of truth for different teams. The CRO looks at the CRM. The CFO looks at the billing system. The CTO looks at the database. Changes to pricing & packaging require changes everywhere.

It’s understandable how this happens. Early on, it’s easy to hardcode and move on. People rarely make time to invest in architectural coherence.

That is, until they’re forced to.

At our Monetizing AI Summit, Shar Dara, who leads billing at Vercel, put it bluntly:

Traditional SaaS companies updated pricing once a year. In the AI era, that’s impossible.

At Vercel, we average five to six pricing changes per month: new SKUs, add-ons, packaging tweaks.

You need architecture that can keep up.

Ben Papillon, Schematic’s CTO, added:

If pricing agility is important, you need an architecture that lets you change how you price without rewriting your product.

The Four Pillars of Modern Pricing Architecture

The solution that Shar and Ben outlined is a decoupled, four-part architecture.

Pillar 1: Unified Product Catalog

A single schema describing every plan, feature, entitlement, and price, as the canonical source for billing, provisioning, analytics, and GTM.

This single catalog ensures that changes appear everywhere, instantly.

Pillar 2: Decoupled Entitlements

The heart of agility is to decouple pricing logic from code. Instead, feature access rules are stored centrally and queried at runtime.

Instead of this, repeated for every feature and plan:

if (planId == Plans.Enterprise) {
  enable_feature_x();
} else if (planId == Plans.Pro) {
  // ...
}

Your code becomes a single, clean call:

const { enabled } = useEntitlement(“reports”);
if (enabled) {
  showReportButton();
}

This simple pattern replaces hundreds of lines of brittle logic. It also allows business users to change packaging without ever touching code.

Pillar 3: Real-Time Metering

AI products have probabilistic consumption, introducing uncertainty. Real-time metering gives customers transparency and predictability.

This is critical because, as Marcos Rivera of Pricing I/O says, a core anxiety of consumption billing is the unknown:

The ‘oh shit’ moment everyone’s worried about is the huge, unpredictable bill. It’s nerve-wracking to go into an agreement and not know how much you’re going to use. Real CFOs don’t buy that way.

Your architecture must solve this anxiety. This is why you must provide tools to estimate and monitor usage. Visibility calms anxiety.

Pillar 4: A Control Plane for GTM Teams

Let non-engineers manage monetization. Your growth and product teams should be able to:

• Create new plans and SKUs

• Adjust limits or pricing tiers

• Trigger experiments or promotions

...all without filing an engineering ticket.

Before and After

These pillars dramatically increase agility. We built Schematic to make this possible.

Zep launched metered billing in ten minutes. Plotly shipped two AI products two quarters faster.

How fast could you run?

Treat Pricing Like Product

When most founders think about pricing, few of them picture architecture.

But your architecture determines how fast you can test and how confidently you can scale.

Kurt Smith, the CEO of Fexa, once said something that to this day I find both true and haunting:

Think about all the creative pricing & packaging conversations in SaaS & AI companies that simply stopped happening, because people accepted that even if they had good ideas, their systems would make testing those ideas impossible.

In the AI era, those who treat pricing like product will win.

Thank you for reading. As a reminder, to get your copy of You’re Leaving Money on the Table, click here (online) or click here (print).

This one phrase is so punchy, memorable, resonant. It promises explosive growth in 5 simple words. No wonder everyone chases it.

Teams obsess over defining their own “aha moment”.

They spend quarters trying to drive it.

In doing so, they fall into 4 traps.

Here they are and how to avoid them.

Brought to you by Adora.

See why product-obsessed teams like Notion, Canva, Replit, and Granola use Adora to map their entire product, document every screen in a live design library, and surface then fix user friction. Get started here.

It’s not an epiphany

The very phrase “aha moment” implies a single, dramatic epiphany.

It suggests if you improve it, everything downstream will improve.

But real experiences don’t work this way.

It’s impossible to improve the business by just focusing on the “aha moment”.

To consider how it actually works, I like to visualize the Panama Canal.

A waterway, with a series of locks.

Working on the “aha moment” is the equivalent of lifting a gate early in the canal.

Water sloshes forward.

Then it abruptly hits the next gate.

To keep it moving, you must release the next gate… and the next… and so on.

So, yes — you might help more users reach the “aha moment”.

But by definition, they’re all “marginal” users who flipped from “not reaching it” to “now reaching it”.

They are all lower-intent than users who already found value in your product.

As a result, they abruptly hit the next gate and drop off — unless you focus on what happens next.

Celebrate the journey

You have to consider what happens next… and after that… and so on.

Casey Winters and Shaun Clowes define these precisely. I consider these mandatory reading if you are driving activation:

The “aha moment” is when a user first experiences your product’s core value proposition.

To define this moment in your data, look for a signal that the user has taken the core action that delivers your product’s value prop for the first time within some initial time period.

The goal is to know what moment tells us if the user is likely to develop a habit.

— Define your aha moment

And:

The “habit moment” is where the user has established a habit around the core value proposition.

A habit implies that they’ve taken the action a repeated number of times within a certain time period.

The amount of time it takes to build a habit varies depending on the product, the complexity of the habit, and the individual person.

— Define your habit moment

Examples:

• Slack

  • Aha: 3+ messages exchanged in the first 7 days

  • Habit: 1+ message exchanged on 4+ of the first 7 days

• Airbnb

  • Aha: 4+ star review in the first 6 months

  • Habit: 2+ bookings in the first 1 year

Think about what the “habit moment” is before you work on the “aha moment”.

Think about what an “engaged” user is before you work on the “habit moment”.

Then consider how the full user journey must change for more users to succeed.

7 is a lie

“7 friends in 10 days” is spuriously specific.

Of course, “8” friends is better than “7”, “9” days is better than “10”, and so on.

A better mental model, per Adriel Frederick on Lenny’s podcast:

“7” and “10” turn an idea — accumulating friends over time — into a binary outcome.

This lets you construct a KPI: the % of each new cohort hitting the “aha moment”.

KPIs are useful to track progress, establish causality, confirm incrementality.

But to actually improve the customer experience?

Product managers, designers, and engineers are better off helping users accumulate as many friends, as fast as possible.

Lines in the sand

How should your startup think about this?

The above guides explain how to pick a great “aha moment” and “habit moment”.

But no matter what, you are ultimately picking lines in the sand.

At Superhuman we wanted a product action to define our “aha moment”.

From first principles and regression analysis, we chose “emails cleared”.

We then wanted a frequency and time window.

We looked at the percent likelihood that a user was retained for different numbers of emails cleared over different time windows.

But ultimately we just picked lines in the sand.

“50 emails cleared within 7 days”.

And the stronger mental model was: “as many emails cleared, as fast as possible”.

Pick numbers, accept their imperfections, and move on.

A Frankenstein experience

Teams that follow the above thinking rapidly hit a sharp edge.

Despite all the structure — or perhaps because of it — it’s too easy for teams to ship a disjointed, Frankenstein experience.

This is especially common when different people are asked to own different parts of the journey.

A growth marketer owns the website. A growth PM owns the FTUE. A PMM owns activation emails.

Together, they build something fragmented and half-baked.

One brain, one journey

A single brain must own the full new user journey.

This brain should think like a hotel manager responsible for the entire guest flow.

From booking, to confirmation, to arrival, to check-in, to finding the room, to finding the spa, and finally, at long last, hitting the jacuzzi.

Let’s say you assign a Growth PM to work on activation.

Give them the full new user journey. This might be the first 30 days. It might even be the first 3 months.

Whatever it is, it should all be theirs.

As a first priority, they should create and maintain:

• A simple paragraph describing the journey, including the “aha moment”, “habit moment”, and “engaged” state

• A document articulating each step of the journey clustered into areas like “first minute”, “first hour”, “first day”

Tools like Adora make this remarkably easy. Especially when your onboarding grows in complexity across use-cases, languages, devices, and more.

Now, with this one brain owning the whole experience, how should they focus?

How much effort should go into the first minute, versus the first hour, day, week, or month?

Use a sliding 80:20 rule.

80% of focus on the first minute, 20% on the first hour.

Ignore anything beyond the first hour.

When the first minute is extremely dialed in, shuffle the 80:20 along.

80% of focus on the first hour, 20% on the first day.

Keep doing this until you have built something remarkable.

One-trick pony

Nearly everyone misses the following:

Facebook’s “aha moment” quietly one-shots three of its growth levers:

1. Value: Each new user experiences core value

2. Virality: Each new user creates more new users

3. Network: Every connection strengthens the network, raising value for all

Of course it was good to drive: it was a three-in-one.

It’s so obvious when you see it.

You could spend months on the “aha” and “habit moment”.

But even if you move the needle, gains will be modest at best.

What’s better?

As many levers, as quickly as possible

Simply:

1. Identify all your growth levers

2. Figure out how they relate to new users

3. Tastefully add them to the new user journey

For example, Superhuman asks new users to invite their team before they land in the product.

I remember the pushback:

“Why are we prioritizing this? How will it improve activation?”

The team’s honest reply:

“It won’t.”

And it didn’t.

But that one step drives 40% of all team invites. It’s probably more valuable than any other activation work the team did.

Think about your growth levers, and how you might fold them into the new user journey:

• Revenue expansion → Get new users to buy the next tier

• Word of mouth referrals → Get new users to add friends

• Team expansion within a logo → Get new users to add their team

• User-generated content → Get users to create and share on social

• Collaboration network effects → Get users to collaborate

The key, of course, is to insert them tastefully.

If you keep it tasteful, you can ask for a lot.

Much more than most realize.

No footsteps, no sound.

You can barely see them. Just a shimmer in the air; a faint outline where a person should be.

Are they even there?

They’re holding something you sold them last week.

They seem… upset? Disappointed? It’s hard to tell. Their face is a blur.

They say nothing. But they want something.

A return.

Okay.

You hold your breath. Close your eyes.

You silently take the item. Hand them the cash.

Not a word exchanged.

They drift towards the door, fading away.

Ghostly quiet.

Gone.

Bizarre.

Except… this is exactly how most startups handle churn.

A customer clicks “cancel”.

“Are you sure?”. “Confirm”.

Not a word exchanged. Not even a glance.

Gone.

If this were in person, you’d talk to them. You’d try to understand.

“What went wrong?”, “Do you want something else?”, “How can we fix it?”.

Why don’t startups do this?

There’s so much you can do with that moment of churn.

It starts with a simple question:

What made you cancel?

Brought to you by Churnkey. High-volume subscription businesses like Superhuman, Gamma, and Veed use Churnkey to reduce voluntary and involuntary churn. Churnkey helps save 20%-40% of subscription revenue that would otherwise be lost.

Listen

About 80% of churn is voluntary.

Customers are making a choice. Choosing to cancel, versus a failed payment or expired card.

You have their attention, even if just for a moment.

In that moment, ask: “What made you cancel?”1.

What they tell you is critical. You’ll use that information over and over. So let’s make that information high quality.

Provide buckets:

• Too expensive2

• Don’t need it anymore

• Expectations not met

• Prefer an alternative

• Technical problems

• Missing functionality

Then go beyond the bucket: ask for an explanation. Even better, require an explanation3.

Make it mandatory. Impose a minimum character limit.

Sure, haters will hate, complainers will complain.

But you and your customers will be better off with this info.

Of course, you’ll receive some nonsense. Terse, one-word answers. Straight up gibberish.

So: why not have AI scan in realtime? Check if the explanation was specific enough to act on, and if not, prompt for more detail4

Save

Now that you have detailed reasons, save as much as you can.

Companies using Churnkey see a median save rate of 18%. The top 10% see save rates above 26%. Some see save rates as high as 50%.

This starts with simple logic. For example:

• The user says: Don’t need it anymore

• You suggest: Pause subscription for one month

But again, why stop at basic, linear rules?

You’re already capturing reasons and detailed explanations. You already know the customer’s plan history, usage, location, and more. The customer is churning — it’s now or never.

Imagine the customer typed: “I’m a student and I don’t need this over summer”.

We know they’re on the student plan. We know they were a happy and engaged user.

A good offer might be a free month, but the best offer might let them pause for three months.

Avoid the hassle of reinventing the wheel. Use off-the-shelf solutions like Churnkey’s Adaptive Offers.

Synthesize

With maximum churn saved, go back to those reasons.

Now batch it into discrete areas. For example:

Engineering

• Technical issues (bugs, failed flows, downtime)

• Usability issues (performance, UI/UX)

Product

• Use case not solved well enough

• Entirely missing use case

Growth

• Couldn’t reach value fast enough

• Price didn’t match value

Marketing

• Promising the wrong thing

• Bringing in the wrong customers

Then assemble the information to answer:

• What was the total amount of churn?

• How much was successfully saved?

• How did it break down by bucket? By ownership area?

• What are the big changes versus last week?

This packet should let you dig deep.

What were the specific features and issues? What were the exact customer quotes? Who were the exact customers?

Act

The hardest part.

Most companies fail to act on churn.

I think the failure is two-part: a lack of information, and a lack of prioritization.

If you’ve followed the above, you should have near-perfect information.

But prioritization is genuinely hard.

So let me suggest a framework.

First: do you have PMF?

There are many definitions for PMF.

From “40%+ of customers say they’d be ‘very disappointed’ if your product disappeared tomorrow”, to “customers are buying the product as fast as you can make it”, to “customers are spontaneously telling other people about it”.

In the context of churn, if your retention curve relentlessly slopes towards 0%, you don’t have PMF:

If you see this, your only job is to get to PMF.

To get there faster, pay close attention to what churning customers are saying.

If you do have PMF, your retention curves will flatten.

Then you should ask: how hard is it to raise the point where that flattening happens?

If it’s quick and easy, just do it.

Every churn improvement benefits all current and future cohorts.

Meanwhile, leaving issues unfixed means burning through customers you’ll never get a second chance with. This hurts revenue and brand.

So: fix bugs. Adjust pricing. Ship features. All of it, sitting right there in your churn data.

Now you’re left with the hard stuff.

You have PMF. You’ve picked the low-hanging fruit.

But your churn data is surfacing gaps. Most likely these are product swings that need real investment.

This is where you shift from reacting to churn to channeling what you’re learning into product strategy.

How to build that strategy is beyond this piece.

But here’s an obvious truth: most companies make these decisions with far worse information than you now have.

You have a stream of customers telling you exactly why they left.

That’s gold. Use it.

Why doesn’t every company do this?

There’s no technology barrier.

It starts with one question: what made you cancel?

Of course, I am a marketer right now.

I market myself. I market the companies I advise. I help marketers market.

But I specifically wouldn’t want to be one marketer, at one company.

AI is breaking the role.

Why? And what should founders and marketers do about it?

The case for humans

There’s an explosion in AI tooling for marketers.

Most seek to 10x, 100x, or 1000x existing output. Some want to automate the entire endeavor.

One can’t help but ask: is there still room in marketing for humans?

I believe the answer is and always will be yes.

Taste matters. Slop smells.

World-class marketing requires the unexpected. A flash of insight. A little je ne sais quoi.

Humans provide this raw creativity.

In-house marketers

The logical conclusion for most of the last century was to employ full time marketers.

Marketers take your message to your market. Your product reaching your customers. Beating your competitors in your channels.

This required an enormous amount of context.

The cost of learning all that context justified permanent roles.

Specialties, constantly covered. The social media marketer, the performance marketer, the content marketer1.

A permanent team of builders.

Always producing, always on.

Context shift

AI is increasingly capable of handling vast context.

With near-instant ramp-up time.

And they’re getting faster to adapt, more nuanced, and more self-directed on long-range tasks.

The context that necessitated in-house marketers is rapidly shifting from meatware to software.

This threatens the in-house marketer.

Not directly from software. Marketers won’t directly lose their jobs to AI agents.

The bigger threat is its humble namesake: the marketing agency.

AI-powered and fiercer than ever.

Cambrian-era explosion

Agencies aren’t new.

They’ve long bridged the gap between annoying work companies would rather skip, and business outcomes.

How are they changing?

A century ago, agencies were expensive to run, expensive to hire.

Affordable to only the biggest companies.

This made sense. Production required actual factories. Distribution required horse-drawn carriage. Media access was scarce and secretive.

The internet compressed this.

A few high-skilled professionals were all it took.

One Ad Managered, another Photoshopped, a third SQLed.

AI is compressing it yet again.

Now all it takes is an individual and an internet connection.

They don’t need to write any copy, push any pixels, crunch any numbers.

Their AI does it all.

Why is this better than an in-house marketer?

They get to try, fail, and learn across dozens of companies at the same time. The in-house marketer doesn’t stand a chance.

Tiny, AI-first agencies are flourishing.

We’re in a Cambrian-era explosion of agency diversity.

Every possible niche in the marketing ecosystem is being filled by Ai-powered agencies who do one thing, extremely well, over and over again.

There’s an agency for that

I’m seeing it first hand.

For every marketing need at startups I advise, a boutique agency meets the need 10x or 100x better than an in-house marketer.

Want to host B2B sales prospects at fancy dinners?

There’s an agency for that.

Flood social media with UGC?

There’s an agency for that.

Hack Reddit threads?

There’s an agency for that.

From organic and paid search, to LinkedIn ghostwriting, to memes on X, and beyond.

If you can imagine it, there’s an agency for it.

They exist because they supply human taste.

Intimate IRL dinners. Authentic Reddit posts. Funny X memes.

These aren’t automatable, but they sure are delegatable.

Strictly rational

This part is non-obvious to founders and marketers.

For startups, it is strictly rational to hire the agency.

On one hand, spend 30 hours over 3 months to hire one person.

On the other, spend 30 minutes over 3 days to find the world's best agency.

Your employee sees 1 company.

The agency sees 10 companies, sees 10x more, learns 10x faster.

The choice is clear.

By the time you’ve defined, opened, and filled the role, the channel’s changed. The W’s on X. The arb’s gone.

Hire a marketer?

You needed them last month.

Chief Marketing Orchestrator

As a founder or marketing leader, your role has shifted.

From:

• runner → coach

• executor → coordinator

• marketer → multithreader

You distribute context, pull strings, and uphold taste.

You are the Chief Marketing Orchestrator.

With a myriad of agencies at your disposal, your success depends on 3 things:

1. Defining a cohesive strategy

2. Finding the world’s best

3. Driving accountability

I cover defining a cohesive marketing strategy here.

So how do you find the world’s best and hold them accountable?

Finding the world’s best

Consider an analogy: painters, plumbers, electricians.

Knowledge-intense specialists who bring meticulous execution with a dollop of creativity.

How would you hire for excellence in those roles?

You’d ask your friends to get high-signal referrals.

Then you’d ask a few things.

How many times have you done this before? Answer: hundreds, if not thousands of times.

What’s your track record? Answer: endless 5* reviews.

How soon can you start? Answer: immediately.

How fast are you? Answer: unbelievably fast.

How expensive is this? Answer: surprisingly reasonable.

Litmus test: don’t trust a marketing agency that sucks at marketing itself.

Driving accountability

Underrated yet necessary skill: driving accountability.

On average, in-house marketers care more. Their upside and downside are higher.

This doesn’t mean agencies are worse, but they do need to be managed.

First, define excellence.

• Absorb industry benchmarks

• Observe the best in your industry

• Define your goal, in numbers, feel, or both

Write it down. Be precise.

Then, demand excellence.

For each effort, require written update weekly; accept nothing less.

Use this template:

Relative to our goal:

• What did we accomplish this week?

• What results did we drive?

• What did we learn?

• What will we do differently next week?

If progress is being made each week, up the stakes.

If stalling out, cut and replace.

It’s that simple.

Have a marketing problem?

There’s an agency for that.

Choose well. Delegate hard. Replace fast.

I navigated willingness to pay for Superhuman many years ago.

We did something remarkable. We established a premium price point for a product that had always been free. But getting there wasn’t easy.

I studied Van Westendorp. I read conjoint papers. I stumbled through conversations.

I wished for a simple guide.

This essay is that. What follows is the clearest explanation of willingness to pay I’ve seen. Every founder should internalize it.

This is a guest post from Fynn Glover, Founder & CEO of Schematic, from his pricing & packaging book: You’re Leaving Money on the Table.

To get the full book, click here (online) or click here (print).

After selling my first company — where, for full disclosure, I’d botched pricing and packaging — I was charged with leading pricing at a high-growth cyber company.

This gave me the opportunity to meet James Wood, who ran Insight’s Center of Excellence on pricing.

I learned a lot, but most importantly, I learned different approaches to conducting primary research on customer willingness to pay.

I believe that being able to talk to customers about willingness to pay is arguably the single most important tool in a founder’s toolbelt for finding PMF.

I also believe that conducting these conversations should never stop.

They should occur when you’re pre-product, after you’ve won initial customers, and then on an ongoing basis as you extend and deepen your PMF.

Before leaving the cyber company to start Schematic, I did dozens of WTP interviews to prove to myself the market would pay for the solution we envisioned.

When enough people confirmed WTP, I felt confident enough to go out and raise money and leave the good gig.

Over the years, I’ve done countless WTP interviews with customers.

But I’ve also trained hundreds of founders on WTP so that they are armed with the vocabulary of pricing and packaging and are therefore confident in their ability to position and sell their pricing.

This has evolved into daily Pricing Therapy: live, 30-minute conversations where I help founders learn the vocabulary of willingness to pay and, just as importantly, pressure-test their confidence in their own pricing narrative.

What I often find with early-stage founders is that a lack of confidence is the default state.

As Xano founder & CEO Prakash Chandran puts it,

“almost 100% of founders undervalue themselves... especially initially. You’re like, ‘please, will you pay me money for this? Like any money, like, can you give me $5 for it, please?’”

He’s right.

We so often set our first price from a place of desperation, and then, as he says, we’re “so scared” that “everyone is going to leave” that we never touch it again.

My “therapy” sessions with early-stage founders attempt to work through this fear.

It’s part therapy, part Socratic method, and part design exercise. What happens in those calls says more about a founder’s confidence in their value than any other exercise I’ve seen.

A Real Session

(The following story is drawn from an anonymized Pricing Therapy session with a company we’ll call “AdOptics” — a fast-growing ad-tech platform in the DTC e-commerce space.)

AdOptics had built something powerful: software that automatically stops brands from wasting ad dollars on non-incremental audiences.

It was delivering 30–40 percent improvements in customer-acquisition cost. On paper, an easy sell.

In practice, their founder — Alex — was frustrated. He’d gone through three pricing models and none felt right.

He told me:

“In ad tech, we look expensive to SaaS buyers and cheap to performance marketers.

We started at five to ten percent of saved ad spend — ROI-based and fair — but it was impossible to explain.

Then we went with one percent of ad spend — too unpredictable for CFOs.

Now we’ve landed on static tiers.

It works, but it’s finger-in-the-wind.”

He looked tired.

Pricing fatigue is real.

I asked him to role-play.

“Pretend I’m your dream customer. I ask, ‘Hey Alex, can you explain your pricing?’ Walk me through exactly what you’d say.”

He delivered his pitch — a clear explanation of tiered pricing tied to ad spend — then paused.

“Grade yourself,” I said. “How’d that feel?”

“B minus,” he answered.

“I know it should be shorter and more about value, but the math is hard to explain.”

“Okay,” I asked, “now put yourself in your customer’s seat. How do they feel hearing that?”

“Sticker shock. They’ve never seen a SaaS tool at six or seven grand a month. Even if it’s positive ROI, it feels foreign.”

We talked about what would make a customer excited to pay more next year.

Alex didn’t hesitate:

“If they cut acquisition cost forty percent and lift profit sixty, they’d happily pay ten grand.”

He knew the value; he just hadn’t built the story around it.

From there, the conversation shifted from price level to value.

I asked about value metrics:

“Is CAC your customers’ single north-star metric when they purchase your product, or are there others?”

“Mostly CAC,” he said. “Sometimes NCAC — new customer acquisition cost — but same idea.”

That’s when we got to the heart of it:

“You have a clear value metric — ad spend reduced, or CAC improved — but you’re forcing that value through a single lens: percent of ad spend.

What if you built a hybrid model? A base platform fee plus a performance-linked variable? Predictable for the CFO, scalable for you.”

His eyes lit up.

“That would actually calm the sticker shock. Two thousand base, plus usage or performance on top. Makes sense.”

Exactly. The session wasn’t about discovering the “perfect” price.

It was about aligning architecture, psychology, and confidence.

By the end, Alex said something every founder eventually says in these calls:

“I feel like I can finally explain it without tripping over myself.”

Why These Conversations Matter

One of the most important things to understand about pricing is that it is a communication discipline.

The way you talk about price — the language, metaphors, and structure — shapes how customers perceive value long before they ever see a proposal.

Try to surface three things:

1. Clarity: Do you understand what value you create and how to express it?

2. Confidence: Can you deliver your price narrative with confidence, poise, and persuasion?

3. Consistency: Does every team inside your company tell the same story?

If those aren’t aligned, the model doesn’t matter.

You’ll lose deals, underprice, or discount out of insecurity.

The Playbook: Eight Steps to Discover Willingness to Pay

Over time, I distilled these sessions into a repeatable flow.

It isn’t statistical; it’s conversational.

You can run it with customers, prospects, or design partners in under an hour.

1. Context & Framing: Explain your product’s promise. Get the buyer in the right mental model.

2. Perceived Value: Ask what problems they think it solves, and which features feel indispensable.

3. Relative Value: Compare your value to the tools they already pay for. If they spend $100K/yr on Salesforce and think your product is 50% as valuable, they’re indicating that they’d value your product at ~$50K/yr.

4. Price Metric: Explore which metric best maps to their perceived value — seats, usage, API calls, customers, etc.

5. Packaging Schema: Show draft tiers (Good/Better/Best). Ask what resonates and what feels off.

6. Maximum WTP: Either open-ended (“What’s the max you’d pay?”) or scenario-based (“At $X, what happens?”).

7. Influence Drivers: Uncover the real factors — budget, comparables, frequency, stakeholder buy-in.

8. Feedback & Emotion: End by asking how the conversation felt. You’re listening for discomfort, not just answers.

The power of this framework is its simplicity.

It requires curiosity and empathy.

You’re teaching customers and yourself to talk about value together without flinching.

Sidebar: When Van Westendorp or Conjoint Make Sense — And When They Don’t

I think Van Westendorp is a good tool when you’re selling a simple product with a single price point, and you want a quick read on perceived expensiveness.

It’s fast and directional.

But it doesn’t capture usage, value metrics, hybrid pricing, or real buyer psychology.

I think conjoint analysis is a good tool when you have a large statistical sample size, well-defined packaging options, and the time and budget to run a proper survey.

It can quantify tradeoffs across bundles — but only when your product and customer base are mature enough to support it.

For most founders and product leaders in high-growth companies, I’ve found this eight-question framework to be far more useful.

It helps you understand how customers think about the value you create relative to other tools, how they would intuitively think about & measure your value, how packaging would influence their decision, how sensitive they are to price, and what constraints — budget, internal politics, reputation — actually drive their willingness to pay.

This isn’t statistical.

It’s behavioral, practical, and grounded in real conversations with real buyers — which is what most startups need.

The Founder’s Psychology

What surprised me most after running hundreds of sessions is how much pricing anxiety mirrors founder psychology.

A founder who underprices often under-believes.

A founder who over-complexifies is usually compensating for insecurity.

As pricing expert Kris Szyszkiewicz puts it,

“I thought at the very beginning the pricing is 90% analytics... 10% confidence.

I wouldn’t stay by this statement anymore. It’s like more of a 50-50.”

And the best founders — the ones who can explain their price clearly, calmly, and consistently — project a kind of gravity that customers feel.

They aren’t guessing. They’re teaching.

In that sense, willingness to pay is reciprocal.

Customers mirror your confidence.

If you hesitate, they sense risk.

If you explain with conviction, they assume value.

Kent Keirsey, founder of Invoke.ai, framed this perfectly when he told me his philosophy:

“I want to earn every dollar that we make... as we earn the price increase over time, as we demonstrate the value... that is how we increase our prices.”

That’s what this chapter is about: building the evidence to earn your price, so you can broadcast that belief & confidence to the market.

Every founder eventually reaches the same moment Alex did.

They’ve proven value, won customers, but still feel awkward saying the number out loud.

That’s the sign you’re ready to do the real work: turning pricing from a source of anxiety into a source of conviction.

From Discovery to Design

Once you have this conviction, you’ve solved the first half of the problem.

You’ve used discovery to find the value.

The next step is to package that value.

This moves us from the “what” of WTP to the “how” of merchandising.

It’s time to take your newfound confidence in what your market will pay and build a GTM and packaging schema that captures it.

In the next chapter, we move from discovery to design. It’s time to talk about packaging.

Guest Perspective: Madhavan Ramanujam

“Price is more than just a dollar figure; it is an indication of what the customer wants — and how much they want it.

It is the single most critical factor in determining whether a product makes money.

Yet so often it is an afterthought, a last-minute consideration made after a product is developed.”

— Madhavan Ramanujam, Monetizing Innovation

Thank you for reading.

As a reminder, to get your copy of You’re Leaving Money on the Table, click here (online) or click here (print).

We were grinding hard.

But we barely had an MVP.

Our prototype was riddled with bugs.

We needed more engineers. We had 3 plus our CTO, but it was nowhere near enough.

We needed to triple to ship a basic MVP.

Being responsible for growth, I faced an impasse:

• To grow, we needed a product

• To have a product, we needed engineers

• To have engineers, we needed to speed up hiring

The bottleneck was clear. Hiring was our growth constraint.

So I did the responsible thing and jumped into hiring.

I spent nine months hiring founding engineers to reach our target team size.

In this write-up I share:

• What I wish I knew at the start

• Mental models that saved us hundreds of hours

• The spreadsheet that fully deconstructs the problem

This is for anyone scaling their team, particularly engineers1.

It works any time you’re urgently filling multiple roles, from your founding team to a late-stage pod.

Please share if you find it useful.

Share

Hiring is basically growth

They both involve:

1. Defining your positioning and message

2. Reaching a large top of funnel

3. Optimizing every conversion step

If you’re good at growth you should be good at hiring2.

You just need to understand the system and do the work.

Define your positioning and message

Your startup won’t grow if you can’t explain why customers should care.

Similarly, your startup won’t hire if you can’t explain why candidates should care.

Answer the following questions.

• What is your company? What is its mission and vision3? Why should someone care?

• What is your product? What problem does it solve? Why is it special?

• What’s exciting about your company? Why will it win? Who are notable teammates, investors, customers?

• What’s the cash and equity? Benefits? Culture?

These should collectively answer: Why should I join your company?

Particularly an engineer who can sleepwalk into a $600k+/year job at FAANGAMANGO4, or start their own company and own most of it.

Write down your answers, as you’ll use them throughout the hiring process:

• Messaging candidates before interviews

• Pitching candidates in early interviews

• Closing finalists

Next, refine your words and practice saying them out loud.

Why?

Impressions matter.

Candidates will forget 90% of what you tell them, but they’ll remember how they felt.

In early interviews, you only have a few minutes to pique or lose candidate interest.

In late interviews, you need to stand out amongst the crowd.

You want candidates to feel excited and impressed. They should think “I love that company!”

Knowing what to say makes every second count.

At Superhuman, everyone running interviews authored their own one-page pitch doc, and it worked5.

Reach a large top of funnel

You’ve defined your message. You now need to go to market.

The first step is understanding your channels.

Hires generally come from these four places:

• Inbound

• Hiring platforms

• Referral

• Cold outreach

Inbound

This channel is obvious: candidates who apply for the job.

These candidates are often prudishly dismissed as low quality.

People who say this are wrong. Possibly low quality themselves.

Inbound applicants have reliably been some of my best hires across every role, including founding engineers.

Get your job on as many boards as possible (LinkedIn, Glassdoor, Indeed, Teal, etc). Then market the hell out of it.

Hiring platforms

These are active candidates, findable on platforms, waiting to be messaged.

Tthey don’t know about your company so you must pitch them.

Like inbound, this channel is often dismissed as low quality.

But again, this is wrong. Again, many of my best founding engineers were sourced on hiring platforms.

Hiring platforms come and go, but some at the time of writing: Wellfound and Next Play. Campus recruiting is also a form of hiring platform.

Referral

Referral is what usually comes to mind when trying to scale a team.

Who’s in your network? Your network’s network?

Think broad. You, team, friends, family, investors, advisors, early customers. Run a memory palace with them all.

Use the latest tools like LiftOff and Happenstance.

Go beyond network: pay attention to industry events.

If a startup just got acquired or shut down, see who in your network will connect you to the talent list.

If an event is bringing candidates into one space, score an invite through your network and show up.

As we’ll see, this channel is usually the most important, so invest heavily in it.

Cold Outreach

These passive candidates are happily minding their own business. Few are job hunting.

You hope to pry them from their current occupation to focus on your startup.

This channel is the most scalable but the hardest to convert.

It can feel depressing. Hundreds of outreaches, only to get one or two “no thanks” replies.

Fortunately a litany of tools exist to help. LinkedIn Recruiter, Gem, Juicebox, Clay, Clado, Brix, and more.

You can either ramp up on these and DIY, or outsource to a recruiter.

I always recommend the latter.

Find a quality recruiter up to speed on the latest tools and techniques and judge on candidate volume and quality.

Note: While tooling isn’t the focus of this write-up, using the right tool can make or break your search.

But SaaS for recruiting is incredibly turbulent. Based on the last 10 years, half of these solutions probably won’t exist in another 10 years. YMMV.

Don’t see your favorite recruiting tech tool listed above? Message me to get it added. Did one above miss the mark? Let me know.

We’ve defined message and channels.

Finally, we must understand and optimize the funnel.

Optimize every conversion step

There are 5 stages to a typical hiring funnel6.

Outreach → Screen → Onsite → Offer → Accepted

Let’s now introduce the master spreadsheet.

This is your source of truth model; a reliable way to hit your hiring goals.

Click here to copy the spreadsheet.

Next, watch this Loom for an explanation.

Prefer to read? Keep going 👇

Hires needed

Everything starts with the number of hires needed.

In this example we’re going big: we need 10 hires.

Punch in your target at the top.

Time to hire

Assert your fastest possible time to hire.

I say “fastest” on the assumption that your startup will die without these hires, so you’re giving it everything.

If that’s not the case you should focus on something other than hiring.

For engineers my rule of thumb is 4 weeks for the first hire, and 2 weeks per incremental hire once the recruiting machine is running.

So:

• 1 hire should take 4 weeks

• 5 hires should take 4 + 2 + 2 + 2 + 2 = 12 weeks, or just under 3 months

• 10 hires should take 22 weeks or just over 5 months

Put your numbers in. Sense check the conclusion.

Channel Mix

We next must assert a channel mix.

For a batch of hires, what % should come from inbound vs. hiring platforms vs. referral vs. cold outreach?

For founding engineers I typically see:

• Inbound: 15%

• Hiring platforms: 10%

• Referral: 40%

• Cold outreach: 35%

Don’t like this mix? Change it. Just make sure it adds to 100%.

Conversion Rates

Finally we must assert conversion rates.

Per channel, what % will move from each interview stage to the next?

You might assume pushing to 100% is ideal.

But hiring obviously doesn’t work that way:

• Candidates won’t meet the bar

  • You want to minimize time spent on candidates you won’t hire

  • But hiring necessarily involves interviewing candidates who ultimately do not pass, even with strict top-of-funnel filtering

• Good candidates will drop out

  • You want to minimize regrettable drop-out

  • But you won’t convert everyone, even with perfect selling

These are the percentages I see in healthy founding engineer hiring funnels:

In the Loom I elaborate on why these numbers.

A key principle is that the percentages should increase as candidates get deeper in the funnel.

This means you are narrowing in on the best candidates in every step.

In reality, how good or bad you are at marketing, targeting, screening, and selling will dramatically affect these percentages.

The above are very good conversion rates. Consider them healthy targets to aim for, even if your starting point is completely different.

The model will now spit out detailed, concrete recruiting goals.

Total Outreaches

The top row tells you the numbers you must hit to close 10 hires:

In this example, you need to:

• Receive ~500 applications7

• Message ~100 candidates on hiring platforms

• Contact ~100 referrals in your network

• Cold outreach ~4000 profiles

If you do this and the conversion rates hold, you’ll hit your hiring goals.

Total Interviews

The rest of the first table tells you how much total activity is needed.

In this example you must run 192 screens and 34 onsites to reach 10 hires.

Remember these numbers when you design the interview plan.

If you make the onsite 4 hours, you’ll spend 136 hours or 17 days interviewing. If you make it 8 hours you’ll spend 34 days in onsites.

That’s a sickening 7 full weeks just doing onsites!

The takeaway is clear: design the most efficient interview imaginable.

Beyond onsites, you’ll need to deliver (at least) 13 offers to get to 10 accepts. Bitter pill to swallow, but best to recognize up front.

Weekly Interviews

The final section tells you how much weekly activity you need.

In this example, you must do (at least) 11 screens, 2 onsites, and give 1 offer each week to hit the goal.

Of course, it’ll be difficult to hit these numbers from a standing start. You’ll likely miss in the first few weeks.

The model takes this into account, creating enough buffer so that if you’re hitting these numbers by week 5, you’ll make the hires.

Where next?

The stage is set.

Admittedly we haven’t touched the messy part of recruiting: branding, making noise on social, rifling through your network, getting jobs live, outreach, hiring recruiters, running events, showing up for coffee chats, interviews, exercises, work trials, sell dinners, offer letters, negotiation, start dates, onboarding…

But at least you have a plan.

Back to 2016. Back to Superhuman.

This was the exact process we followed.

And it ran like a well-oiled machine.

Each week we tracked volume. We debugged conversions. We made the hires.

Now it’s your turn:

• Copy the spreadsheet

• Watch the Loom

Do the work and build your team.

We’re “people first”.

We think about people, not problems.

We reason through roles, not responsibilities.

We discuss what’s right in front of us, not what’s missing.

Our “people first” tendency shows up everywhere. Especially in hiring for your startup.

Slow eng culture? Intangible issue. Cracked engineer? It’s time to build™️

Mediocre brand? Meh, tomorrow’s problem. Designer from SF Tech Week? “Hired”.

Wildly inefficient

Hiring “people first” is wildly inefficient for your startup.

It works for cofounders. It works for founding team. And it works when the individual is top 10 in the world.

But 99.9% of the roles you’ll ever hire at your startup do not fall into these buckets.

If it becomes the norm, hiring “people first” can be company-destroying.

Here are three ways the mistake shows up. Are they happening at your startup?

Subscribe now

1. Role In Search of a Problem

Every hiring manager makes this mistake; most don’t realize:

Founder: We need a Growth Marketer

Recruiter: 🫡

Cofounder: Why do we need a Growth Marketer?

Founder: To grow

Cofounder: What are they going to do?

Founder: Not sure… run Facebook ads?

Cofounder: Why aren’t we growing?

Founder: Hmm not sure

Cofounder: I think it’s because-

Recruiter: Here are 20 Growth Marketer profiles

Founder: Great, let’s interview the top 10

Recruiter: 🫡

Cofounder: We know how to acquire, retain, and monetize customers — we’re just not shipping fast enough

Founder: So we don’t need a Growth Marketer?

Cofounder: Right, instead we need-

Recruiter: You now have 10 interviews on your calendar, the first starts in two minutes

See the issue?

Starting with a role skips the critical step of clearly articulating a business problem or opportunity.

Instead, it anchors discussion around pointless details of that role.

Job title, hiring manager, sample profiles, referrals, ideal companies, interview plans, hiring team, ATS updates, level, salary, equity, FTE vs. contractor, team structure, onboarding plan, and on and on.

Unless you are clear on business outcomes, these details don’t matter.

What if the outcomes were never a priority? What if a new role wasn’t ever the solution? What if someone already on the team could do it?

2. Made Up Role

Here’s how this plays out:

Cofounder: Who was in the office this morning?

Founder: Someone I’m interviewing

Cofounder: For what role?

Founder: I’m not sure yet … perhaps Head of Design

Cofounder: Huh? We’re a 5-person company and already have a Designer

Founder: But this person is incredible

Cofounder: Do we need a Head of Design?

Founder: They have over 100k followers on X

Cofounder: What are their salary expectations?

Founder: Probably ~$330k…

Cofounder: 😳

Founder: …plus equity

Cofounder: Since when were we hiring a Head of Design?

Founder: We aren’t

Cofounder: But we’d hire them as a Head of Design?

Founder: Yes

Cofounder: How did we get here?

Founder: Our VC made the intro

Familiar?

You or a teammate encounters someone strong.

Then create contrived ways for them to join your startup, even if you have no idea what they would do.

Small and scrappy startups (rightly) jump at promising opportunities.

But if this persists unchecked into hiring, your teams morph into Frankenstein beasts lacking in purpose and clarity.

3. Empire Builder

Perhaps most disastrously:

Founder: Please welcome our new exec!

All: 👋

Exec: I must now hire an army

Cofounder: Why?

Exec: For my empire, of course

Cofounder: How long will this take?

Exec: Only 2-3 quarters

Cofounder: “Only”?

Exec: Don’t worry, I know exactly who to hire

Cofounder: How?

Exec: I’ve worked with them all before

Cofounder: What roles?

Exec: I haven’t decided yet

Cofounder: Can we create interview plans?

Exec: We don’t need to

Cofounder: Why?

Exec: I’ve worked with them all before

What happens next?

The exec is a flop. So is the empire they built.

The entire thing needs ripping out a year later.

A better way

There is, of course, a better way.

From a16z’s The Hiring Process:

If you can’t define it, you shouldn’t hire for it.

Most searches fail before they even start because there isn’t a clear answer to the question, “Why are we hiring for this role?”

How?

Write a MOC — a document that outlines the mission, outcomes, and competencies of the job.

Articulate the business case for hiring, and what, specifically, the new hire should accomplish over the next 6/12/24 months.

MOCs are requirement documents for what success looks like.

They are for your team what PRDs are for your product.

a16z talks about MOCs as a tool for hiring execs, but I believe every hire should have a clear MOC.

The effort needed to create one is a small fraction of the effort going into a search.

JDs

Is a MOC just a JD?

No. A MOC should be written before the JD. The MOC is half of what goes into a JD.

The MOC’s mission, competencies, and title will show up in the JD.

But crucially, the MOC also defines clear business outcomes.

Examples:

• “Grow ARR from $3M to $15M+ within 12 months”

• “Accomplish and maintain <1 hour ticket resolution in first 100 days”

• “Build and launch ChatGPT for Doctors by June this year”

Meanwhile, JDs are full of marketing on why to join your company. Benefits, PR, investors, and so on.

Why aren’t MOCs identical to JDs?

Companies are understandably squeamish about revealing business outcomes in public JDs.

So most JDs are understandably vague on what the person actually needs to do1.

Flow

The correct sequence:

1. Identify important business gap or opportunity

2. Figure out if someone on the team can handle it

3. If no, write a MOC that scopes out a role’s title, mission, outcomes, competencies

4. Use the MOC to create a JD

5. Hire the person

6. Give them their MOC during the hiring process or on day one2

More here!

Don’t build your entire team “people first”.

Start with what you need to get done.

Schematic unlocks pricing agility. Trials, credits, usage data, custom deals, expansion triggers, & more. All in one place.

Easy setup: built on Stripe, no migration needed. Schematic gets it.

Schematic is one of the most interesting SaaS advancements in years. The industry will look back and wonder why the heck we asked internal teams to build these capabilities (often poorly, over and over).

— Michael Morrisey, CPO Sema4.ai & ex-CTO, Fullstory

Book a demo or create a free account here, or email Schematic CEO Fynn at fynn@schematichq.com.

Do you hear this at your startup?

• “Should we be freemium?”

• “How about a free trial?”

• “Why one, not the other?”

Weeks, months, years go by without answers.

Sometimes it goes beyond questions. Sometimes the startup builds for months.

Only to realize they had it all wrong from the start.

Existing literature underwhelms. Blog post fluff that haphazardly lists pros and cons without a clear decision framework.

Bad advice results in bad decisions.

This essay guides founders, product, and growth leaders on how to choose between freemium, free trial, and paid only.1

What is the difference between freemium and a free trial?

• Freemium is free product access that could last forever

• Free trial is free product access that ends after some time or usage limit

Here’s the decision tree 👇 and let’s dive in.

Freemium

First, decide whether to offer a free tier that can be used for free forever.

This decision is fundamental. It affects nearly every aspect of how your business operates.

On the other hand, free trial is a funnel optimization.

Therefore deciding freemium should happen before you decide on a free trial.

Many generational products are freemium. Consider ChatGPT, Spotify, Dropbox.

How should you decide?

Competitive Pressure

Do you face significant competitive pressure to offer the product for free?

Think about whether there are cheaper competitors, or competitors who might be cheaper.

Will they take the market before you?

If yes, consider freemium.

Astute readers might notice the logical leap…

Just because a competitor might undercut you doesn’t mean you should jump all the way to free.

Logically it might be better to price-match or marginally undercut.

But if you do, the competitor might undercut you again.

If you simulate forward enough times, you both reach the lowest price you can afford2.

This is typically free for software. (More on AI costs and costs in general later).

In competitive markets, freemium can cement users into your ecosystem. You can monetize later.

Consider that for every free product with massive mindshare, there are many competitors who might have taken that market.

Spotify had Pandora, Apple Music, and more. Dropbox had Box, Google Drive, and more. ChatGPT had Claude, Gemini.

What if you don’t face competitive pressure? Are there other reasons to be freemium?

Network Effects

Are you creating a network effect that needs many users?

Consider freemium if it creates a network effect that solidifies your market position.

Network effects are when product or business value increases with the number of users.

This mostly happens in products that connect people, like communication and collaboration tools.

Take Slack or Zoom. Mass adoption makes them more frictionless and valuable for everyone.

Or take Figma or Dropbox. The network effect starts with file sharing, pulling more users into the ecosystem. And increases with collaboration.

A more subtle network effect happens when a product generates massive word of mouth entirely off-platform.

Take ChatGPT’s launch in late 2022, or its 4o image launch in early 2025. Each launch sparked a word of mouth explosion. User-generated content fueled growth.

In all cases, freemium fueled the network effect.

But what if there is no network effect?

Data for Innovation or Monetization

Are you collecting a huge amount of data for innovation or monetization?

You might need a free tier to gather valuable data.

That data could improve the existing product or power brand new experiences.

Consider Grammarly or ChatGPT. Free user data both improves existing models and highlights brand new use cases.

Data can also open new and meaningful revenue streams.

The most common is advertising. In this case the “data” are the free users themselves. This is the primary business model for social media, search engines, casual online games, and more.

Some SaaS companies also monetize ads. Spotify’s free tier ads are ~12% of revenue; Duolingo’s are 5-10% of revenue.

Last but not least, you could directly sell the data. This is increasingly popular given AI companies are hungry for data to train models. Just be transparent with your users!

Free Trial

Having explored freemium, how should you decide whether to offer its distant cousin, a free trial?

Competitive Pressure

Do you face significant competitive pressure to offer a free trial?

We meet this concept again.

Only now we know the pressure isn’t so massive that you need a free tier.

Imagine two identical products.

Neither offers a free tier. One offers a free trial.

If the other doesn’t have a free trial, it would lose market share.

Consider Calm and Headspace as an example of this dynamic. Both have extremely similar time-based free trials.

Or consider each of Replit, Lovable, Bolt, and v0’s usage-based free trials.

When I worked with Replit earlier this year, the Agent pricing conversation was one of the quickest I’ve ever seen. Given the competition, there was no answer other than a usage-based free trial.

What if you have little to no competitive pressure?

Reaching New Customers

Do you need to reach lower intent customers to grow?

Imagine that you’re already reaching customers who are willing to pay up front.

You might feel you’d grow faster by also accepting lower intent customers on a free trial.

The hypothesis is that some of these customers will pay — but only if they experience it for free.

This is common for mature products that start to saturate their market. Salesforce, Adobe, and Tableau are examples.

What if you’re early stage and reaching lower intent customers isn’t a priority?

Value Perception

Does using the product for free result in higher perceived value than paying up front?

Users might feel your product is more valuable if they start for free than if they pay up front.

This might result in the same customer activating better.

This is a delicate dance between “the value you convey through marketing, pre-product” versus “the value you convey through product”.

If it’s extremely difficult to convey value through marketing, but easy through product, consider a free trial.

Take Readwise, which helps you remember what you read.

The value is hard to imagine just from marketing.

It is extremely obvious when you connect your Kindle highlights and receive a few emails.

This predisposes Readwise to a free trial instead of payment up front.

Note: This effect is extremely hard to untangle from “reaching new customers”.

Why? Because a free trial is guaranteed to introduce a lower intent persona that explodes funnel volume.

This makes it very difficult to observe how a single customer’s psychology is affected by paying up front versus free trial.

But you can get meaningful signal by manually onboarding customers.

Pay attention to their emotional state, behavioral cues, commitment signals.

Consider a free trial if it seems that your ideal customer only “gets it” after using the product.

Deeper Considerations

Global versus local optimization

You must be clear on whether something is a primary reason to be free, or just a local optimization.

Yes, perhaps you can gather data, expand audience, and so on.

But should you?

Or would you be better off focusing on the main product right now?

For example, Spotify’s ad model — at only 12% of revenue — is probably more of a local optimization, not the primary reason they have a free tier.

The main reason is likely more about competitor dynamics and reaching lower intent customers.

Most free debates I see at startups are due to differences in opinion on whether an idea is a global or local optimization.

Which neatly segues into evaluation…

Was it a good idea?

How do you evaluate if free was a good idea versus paid only?

Free trial versus pay up front is difficult to evaluate, but doable.

Good indicators are day 1 to 30 activated customers, and month 1 to 3 retained revenue.

Activated customers examines customer throughput and value experienced.

Revenue retained tells you if customers actually stick around.

You can run an A/B test where half of visitors get a free trial and half don’t. Then compare outcomes.

Freemium versus paid only is much harder to evaluate, since freemium affects every part of how your startup operates.

Thus there’s no easy A/B test to run.

This is why we need a clear decision framework, with clear hypotheses.

You can then evaluate outcomes against those hypotheses.

Did we beat competition? Did we create huge network effects? Did we gather massively valuable data?

And, ultimately, was the upside worth the cost?

Speaking of cost…

Free isn’t free

Both freemium and free trial assume the upside outweighs the cost of all those users.

Proceed with caution!

More users means more spend on infra, AI, handling support tickets, fixing bugs, handling discussion on social, and more.

AI costs are particularly important. Power users can run a small company out of business with aggressive use.

So you might need usage caps.

Initially this might be precautionary. Over time it becomes strategic and part of how you monetize.

Why didn’t Superhuman do freemium or free trial?

I get this question a lot — so let’s use the framework.

Freemium

Do you face significant competitive pressure to offer the product for free?

No.

There was (mostly) free Gmail and Outlook.

And there were lots of startups. Some had a free tier.

But the whole point of Superhuman is to provide significantly more value than alternatives.

The lack of competitive pressure for the value provided meant there was no need for a free tier to secure market share.

We did the opposite: deliver a far superior product as a wedge, and price high to reflect value.

Are you creating a network effect that needs many users?

No.

The core value of Superhuman does not need everyone to have product access.

In fact, quite the opposite!

Superhuman is backward compatible with existing email systems.

We flirted with the idea that “Sent via Superhuman” plus word of mouth might be a good marketing network effect that could justify a free tier.

We even experimented with this for nearly a year.

The specific hypothesis was that we’d spread faster inside teams if we offered free access.

So we launched a free seat type to existing paying teams, and we required payment past a certain amount of usage.

The virality uptick was huge: ~4x the amount of growth within teams.

Incredible!

But <25% of those seats monetized. <25% even activated.

On a revenue basis we were worse off.

Disaster.

So we pulled the plug … for now.

Are you collecting a huge amount of data for innovation or monetization?

No.

Free Trial

Do you face significant competitive pressure to offer a free trial?

No.

Do you need to reach lower intent customers to grow?

To an extent, but offering a universal free trial is a delicate choice, because…

Does using the product for free result in higher perceived value than paying up front?

Every time we experimented with a free trial, we saw customers take Superhuman much less seriously than when they’d paid up front.

As a result they activated, monetized, and retained less well.

This was not just because the free trial reached lower intent users that pulled average engagement down.

No. We saw the same customer type take Superhuman less seriously, due to lower value perception.

Customers need behavior change to succeed with Superhuman.

But the free trial seemed to cause less behavior change.

So we concluded that paying up front was a crucial part of the experience.

And if someone really wants a month free, they can get a referral.

Now it’s your turn.

Considering adding freemium or a free trial?

Or think you might have erred in your choice so far?

Use the framework.

Sponsored by Schematic: helping startups rapidly build & iterate their pricing models. Grow faster and free up engineering — what’s not to love?

Book a free pricing session with Schematic CEO Fynn here.

See why startups like Harvey.ai, Noom, and dozens more trust Top Funnel.

I’ve seen them fill growth, marketing, engineering, and ops roles at ridiculous speed.

My favorite part: they don’t just throw candidate profiles at you. They work all the way through the funnel, from source to close, to find you the best candidate.

Top Funnel can massively accelerate your hiring. Email CEO Tony (LinkedIn) at tony@topfunneltalent.com.

Mention this post and he’ll audit your ATS and roles, free.

Hell Yes

You should only hire someone if you are a “hell yes”.

That’s true in any role. And in growth, it’s non-negotiable.

Why?

Because growth roles are barrels, and an excellent barrel is many times more impactful than a merely good one.

But here’s the problem: growth roles T-shaped.

So your desired skills can quickly balloon into a laundry list of 20+ bullet points.

To get to “hell yes” we need signal amidst the noise.

What makes a great growth hire?

Over thousands of hours of interviews and hundreds of candidates, I’ve boiled it down to the following 2 skills:

• CPU 🧠 — raw intelligence and problem-solving firepower.

• Speed ⚡️ — fierce pace and extreme bias-to-action.

If an individual spikes on both, typically all else follows.

CPU 🧠

CPU — intelligence — is the most important factor.

The best growth hires I’ve ever made are problem-solving machines.

So, a vast proportion of the interview goes toward validating this requirement.

This skill can show up in many ways, from being structured and methodical, to being able to solve intricate problems.

The more this skill shows up in specific growth areas like data fluency and experiment design, the better.

But absent that, I’d still prefer a shaper yet less-experienced candidate over the inverse.

Speed ⚡️

A smart person should figure out the right direction.

The next unlock is how fast they get there.

I look for speed in every facet of how candidates show up.

Do they respond fast? How soon can they interview? How rapidly do they solve problems?

Call me silly for evaluating outside of interviews, but the trend is undeniable.

I recently helped a startup hire a Growth PM. The (excellent) candidate we hired was introduced over email in the late afternoon and offered to interview at any time the same day.

That’s speed.

Close relatives of speed are agency, drive, and work ethic. The best candidates know how much their input counts, so they push hard.

Look for this fire.

Interview Plan

This is my tried-and-tested growth interview plan.

This plan works. I’ve stress-tested it across dozens of startups, from seed to scale.

It’s specifically for growth product managers and growth marketers, but can also be adapted to adjacent roles like growth engineering, data science, and design.

1. First Round

   1. Recruiter screen [30 minutes]

   2. Optional hiring manager sell [30 minutes]

   3. Hiring manager screen [45 minutes]

2. Onsite 1

   1. Qualities [45 minutes]

   2. Execution [60-75 minutes]

   3. Capabilities [45 minutes]

3. Onsite 2

   1. Optional final skills [30-60 minutes]

   2. Optional founder, investor, or advisor AMA [30 minutes]

   3. Hiring manager AMA [30-45 minutes]

Ideal candidate volumes:

• 10±3 candidates at First Round

• 4±2 candidates at Onsite 1

• 2±1 candidates at Onsite 2

Total time spent with finalist: 5-7 hours.

Let’s unpack.

1. First Round

a. Recruiter screen [30 minutes]

This is the first step for nearly all candidates. The exceptions are referrals who might go straight to the next interview.

Recruiters should ideally exit this screen extremely excited for the hiring manager to meet a candidate.

I’ve seen recruiters reluctantly advance candidates. This is bad. Make sure you agree on requirements and where the bar is.

• 🎯 Goal: Gauge fit and sell

• 💡 Skills: Communication, experience, culture fit

• 👥 Interviewer: Recruiter

• 💻 Format: Phone or video call

b) Optional hiring manager sell [30 minutes]

This interview is only for candidates where more selling is required.

Usually, these are passive candidates whom the company targeted or who were referred.

Hiring managers and candidates should exit this conversation excited to move forward.

• 🎯 Goal: Build trust and validate mutual excitement

• 💡 Skills: Communication, passion

• 👥 Interviewer: Hiring manager

• 💻 Format: Video call or in person

c) Hiring manager screen [45 minutes]

Hiring managers should be extremely excited about candidates after this interview.

The Onsite is very time-intensive for all parties, so only advance Strong Yes candidates.

Some of my favorite tips on how to efficiently conduct this interview are in Andrew Chen’s write-up here.

• 🎯 Goal: Broad and shallow vet, and also confirm critical requirements like compensation and location

• 💡 Skills: CPU, speed, and a shallow check of every other desired skill

• 👥 Interviewer: Hiring manager or trusted delegate

• 💻 Format: Video call or in person

2. Onsite 1

Onsite 1 is where you confirm your “hell yes”.

It should be extremely clear if you want to present an offer after Onsite 1.

It’s hugely time-consuming for the hiring manager to run every interview.

So define the “most skilled person” (MSP) and give them specific interviews to lead.

The MSP for these interviews are usually ICs that the growth hire will work closely with, like engineers, data scientists, designers, and marketers.

a) Qualities [45 minutes]

• 🎯 Goal: Deeply validate innate qualities

• 💡 Skills: Speed, agency, initiative, drive, work ethic, organization, ideation

• 👥 Interviewer: Hiring manager or MSP

• 💻 Format: Video call or in person; Q&A with case studies

b) Execution [60-75 minutes]

• 🎯 Goal: Deeply confirm execution ability

• 💡 Skills: CPU, problem-solving, analysis, copywriting, engineering, experimentation, taste

• 👥 Interviewer: Hiring manager or MSP

• 💻 Format: Video call or in person; 5 minutes to prep, 40-60 minutes for candidate to execute solo, 5-10 minutes to jointly review

c) Capabilities [45 minutes]

• 🎯 Goal: Deeply confirm on-the-spot skills

• 💡 Skills: CPU, intelligence, problem-solving, rigor, numeracy, clarity

• 👥 Interviewer: Hiring manager or MSP

• 💻 Format: Video call or in person; Q&A with case studies

3. Onsite 2

Only “hell yes” candidates who you are trying to close should make Onsite 2.

Of course, you might have one or two final areas to vet. The first interview makes space for this.

Optional final skills [30-60 minutes]

• 🎯 Goal: Validate any remaining gaps

• 💡 Skills: Values, humility, receptiveness to feedback, self-awareness, or any other remaining question

• 👥 Interviewer: Hiring manager

• 💻 Format: Video call or in person; topgrading format

Optional founder, investor, or advisor AMA [30 minutes]

• 🎯 Goal: Build trust, sell, close

• 💻 Format: Video call or in person

Hiring manager AMA [30-45 minutes]

• 🎯 Goal: Build trust, sell, close

• 💻 Format: In person

Growth Hiring Traps

Growth PM vs. Growth Marketer

Deciding between a Growth PM and a Growth Marketer seems to flummox every startup.

And for good reason:

Guidelines:

• Both Growth PMs and Growth Marketers grow the business.

• Growth PMs (mostly) use the product, and Growth Marketers (mostly) work outside the product.

• Growth PMs touch many growth levers: activation, retention, churn prevention, monetization, virality, and more.

• Growth Marketers primarily touch acquisition through various channels: search, paid, social, affiliates, influencers, partnerships, virality, and more.

So, if you mostly want someone to reach more users through acquisition channels, hire a Growth Marketer.

If you mostly want someone to use the product to drive growth, hire a Growth PM.

This is crucial to define up front, as the skills are very different.

Sourcing

Where do awesome Growth people hang out?

1. Referrals: Comb your 1st and 2nd degree connections.

   1. Email all investors, advisors, and company friends asking for the best growth people they know.

   2. Use a tool like Teamable or Scoutr to easily mine your team’s contact lists.

   3. Sit down with each person in your team and use the mind palace technique to compile a list of individuals.

2. Inbound: Advertise the role in public.

   1. Demonstrate incredible PMF and that you take growth seriously. Strong growth hires will want to see these two things before joining.

   2. Want a template? Here, here, or here.

3. Outbound: Excellent candidates tend to be quite busy and quite happy.

   1. Work with an excellent recruiter like Top Funnel to source for you.

   2. Treat this like the funnel it is: track volume and conversion at every step.

Remember: Growth people tend to know each other since it’s part of the role to trade strategies, tactics, and benchmarks.

So, look for connectors in addition to directly identifying potential candidates.

Take-homes, presentations, and work trials

Please.

Avoid take-homes, presentations, and work trials like the plague.

Why?

• Doesn’t directly assess the top attributes of CPU or speed.

• Slows down the hiring funnel by up to a week per candidate. Disastrous for hiring speed.

• Biases towards candidates with more time. But free time is not a variable we care about.

You can test all important skills during Onsite 1. Use the interview plan above!

Next, you might wonder: “What exact questions should I ask?” and “What does good look like?”.

The trouble is, I actively interview at startups I work with, so I can’t give away all my secrets… yet 😉.

If you want to go deeper, just drop me a DM. Maybe one day I’ll stop interviewing and share the full playbook.

Until then, share this with people hiring for growth.

If you get a lot of DMs, you're likely losing opportunities in the mess. Kondo fixes that.

I use Kondo to find hidden opportunities lost in my DMs. Split inboxes, reminders, snippets, and keyboard shortcuts built in. Your first month free (worth $30) here.

Referrals are a growth dream:

• Low cost

• High conversion

• Low churn

If you have product-market fit, customers are probably already recommending your product.

Referrals let you double down on this love.

In its early days, customers were eager to recommend Superhuman, but they had no easy way to do it.

I did what any sensible growth person would do. I hacked together and shipped our MVP referral feature.

How it worked:

• Hit 🎁

• Enter email address

• Send

Before I knew it, referrals were an astounding 30% of weekly customer mix.

There were dozens of further optimizations. So we did those things.

Weekly referrals tripled.

The best part? The channel delivered year after year — and still does.

Startups always have and always will grow through referrals. It comes up constantly at startups I advise. And the playbook is long and intricate.

I got tired of repeating myself (!) so I wrote it all down.

To that end, I present: The Ultimate List of Referral Optimizations.

This is for founders, growth, and product people.

Preconditions for referrals

Get clear on definitions

If you aren’t clear on definitions, you and your entire team will go nuts.

Please let me save you the headache:

• Referrer: The person making a referral.

• Referee: The person being referred.

• Referral: What the referrer gives the referee.

Next, let’s clarify channels:

• Referral is when a customer recommends your product in a way that is measured and tracked by your product.

• Word of Mouth is when this happens in an untracked way.

• Do not conflate referrals with Add to Team, which is about growing within a logo. I’ll write a separate essay on Add to Team.

Ensure referrals make sense for your business

Referrals make sense:

• For B2C companies acquiring more customers.

• For B2B companies landing new logos.

Referrals are a good fit when:

• Your product is not so expensive that it would be weird to recommend to a friend.

  • Imagine the friction of referring a product that will cost the referee >$50k per year.

  • Rule of thumb: under $500 a year is the sweet spot for B2C, and under $5k for B2B.

• It is easy for users to identify who specifically would be a good person to refer.

  • In B2C, this is mostly a question of customers identifying “who do I know who is like me?”

  • In B2B, this involves your customer identifying both which company needs your solution and which individual to contact.

    • Easier in product-led B2B, where your users can just think about other users like themselves.

    • Harder in sales-led B2B, where your users might not know the right person to contact.

Ensure you are ready to invest in referrals

Typical conditions:

• Product love: high product-market fit, high retention, high net promoter score.

• You already see word of mouth: customers share your product organically.

It might be too early to invest in referrals if you don’t see these signs.

Work on product instead. Boost customer love.

Buy vs. build.

By the time you read the full playbook, you might be nervous about the amount of infrastructure necessary.

Consider integrating a referral service from this list to save effort.

Now, the playbook.

The Ultimate Referral Playbook

Step 1: Get referral basics in place

The first step is to create low-friction ways for customers to refer.

Customers already want to recommend your product.

So create pathways to create a behavior loop.

1. Give users a referral link that they can share anywhere they want.

   1. Copy link to clipboard in one click or tap.

   2. Optionally include some persuasive text when the link is copied (“I think you'll like Kondo. Here's a link for 1 month free https://app.trykondo.com/referral/2FB3V1”).

2. Create clear referral entry-points throughout the app.

   1. Make entry-points extremely visible from the app home screen. Making it visually indistinguishable drops engagement 3x. Burying it beneath a click drops engagement 10x.

   2. Use words (“Refer”) to make the action crystal clear.

   3. Use iconography (“🎁”) to create visual identity.

   4. If appropriate, pick a color for further visual identity.

   5. Add a keyboard-driven method accessible anywhere in the app. For example, if your app has a Command Palette, implement Cmd+K → Refer.

   6. Support all platforms: Desktop, mobile, web admin, and any others.

3. Instrument and optimize everything:

   1. When existing customers engage with the new referral entry points.

   2. When the referral link is copied to clipboard.

   3. When the referral link is clicked by referees.

   4. When a signup converts via the referral link.

   5. How well those customers retain.

For now, we are not remotely concerned with adding incentives, financial or otherwise. We are simply tapping into existing motivation.

But wait. Isn’t it obnoxious for customers to refer if there’s no incentive?

Not if they love your product so much that they want to share it with friends. If you really have found product-market fit, your customers will already organically refer other customers.

You might well have some storytelling around why customers might want to refer in the absence of an incentive:

• Perhaps you’re running a waitlist in closed beta, and you’re making it so that referrers can help referees jump the queue.

• Perhaps your product’s value increases when there are more connections between users, like in consumer social. In this case, giving customers the ability to refer directly benefits them.

Step 2: Layer referral incentives on top

With the basics in place, we turn to incentives.

The critical thing to remember is that incentives are primarily an excuse to communicate with your customers about the ability to refer.

The details of the incentive also matter. But not as much as the excuse to message them about it.

1. Create a two-sided incentive.

   1. Both the referrer and the referee receive something in return for a successful referral.

   2. Unless you have an extremely strong reason otherwise, make the incentive symmetric. If nothing else, it’s easier to express in copy: “You and your friend both get XYZ”.

2. Make the incentive conditional on a successful referral, not just making the referral.

   1. The referee must get deep enough into the user journey to be considered high-quality growth.

   2. Usually, this is the point where the referee would pay, or when they have reached sufficient product usage depth.

   3. For example, Wispr Flow unlocks the symmetric referral bonus when the referee either pays or hits 2,000 words (which is the primary threshold to shift from free → paid).

3. Express the incentive as cash.

   1. In SaaS, you might be used to seeing: “You and your friend both get a free month”. But humans typically respond more favorably to cash: “You and your friends both get $20”. Cash feels instant and concrete. A free month requires context.

4. Don’t just express a free month incentive as cash; actually make it cash.

   When it comes to incentivizing customer referral programs, there are two options: offer discounts on your products or offer cash (or an equivalent). In most cases, cash is king.
   
   — Cash is king for customer referrals, Tremendous

   1. Use a vendor like Tremendous to abstract away logistics.

5. Update the journey to remind referrers and referees of the incentive.

   1. Referrers should see mention of the incentive at all points in the journey of referring. Every bit of explanatory microcopy should remind them.

   2. Referees should be notified of the incentive in all messages, landing pages, and sign-up screens.

6. Notify referrers and referees when their incentives kick in.

   1. Thank referrers when they successfully refer and when the referee successfully joins. In these messages, remind them about the incentive.

   2. Notify referees when their reward is unlocked. In this message, prompt them to start referring.

7. Create incentive depth. Remember, incentives are primarily about reasons to message, so it’s a good thing to keep changing them.

   1. Add more incentive as the referee gets deeper into the product.

      1. You might give both sides $10 when the referee starts a subscription, and another $10 when they pass a usage threshold.

      2. This is especially powerful if the referrer has some ability to influence the referee’s usage.

   2. Increase the incentive as referrers refer more to drive continued activity.

   3. Layer non-monetary incentives like swag giveaways, invites to exclusive events, and public recognition.

An incentive will inevitably add volume to the top of your funnel.

However, it will also cause average referral quality to decline. This will show up in conversion and retention.

Unless you’ve really messed something up, the former should outweigh the latter. So don’t be alarmed when you see referral to customer conversion tank, and month 1 customer churn spike.

Keep calm and measure incremental impact on the number of new customers retained past month 2.

Step 3: Market the heck out of your referral program

Your referral program lives and dies by how many of your customers know about it.

So make sure your entire customer base is made aware.

1. Launch the referral program to the entire customer base.

   1. Notify customers in-app, via notifications, over email, and any other communication channel available.

   2. Allow them to opt out of this type of communication without harming their ability to receive other messages, of course.

2. Routinely run referral campaigns with your entire customer base.

   1. I find the sweet spot is 2-4 times a year. Much more and customers might experience message fatigue and run out of people to refer.

   2. Connect promotions with meaningful customer moments. For example, at Superhuman, we ran referral promotions during Inbox Zero Week and in the first week of the year to tap into the “new year, new me” feeling.

   3. Use a deadline to create urgency.

3. Front-load the ask at the beginning of the new user journey.

   1. Create an interruptive full-screen ask that tells all new customers about the ability to refer, and the incentive.

   2. Founders often balk at asking new customers to refer so early in their journey. But when an incentive is at stake, many customers are very happy to hear about the offer — especially in B2C. Everyone else will ignore it without further thought.

4. Prompt users to refer at moments of peak happiness.

   1. This makes the most sense when they’ve completed some action that demonstrates product love. Think reaching a streak milestone in Duolingo, or giving a 5* Airbnb rating.

5. Mention the ability to refer throughout lifecycle messaging.

   1. Advertise the ability to refer as a “ps” or “by the way” in your activation emails.

   2. Include a dedicated email in your activation sequence about the ability to refer so that all new customers are made aware.

   3. Make mentioning referrals a standard part of your new product announcement messages.

   4. Create an ‘insiders’ group of your top referrers with ongoing fun and community-oriented messaging.

Step 4: Make your app a referral machine

You now have a well-structured and well-advertised referral mechanism.

Next, turn your app into a referral machine.

1. Don’t just rely on users sharing a link. Create the ability for your app to directly message referees on behalf of referrers.

   1. Pre-write the message to make it easier for users to hit “send”. Let referrers tweak it if they desire.

   2. Make this message look like a real message from the referrer, not a generic-looking message from a company.

   3. Include information that goes beyond “I like this tool”. For example, include usage statistics that show the referrer is getting value.

2. If possible, suggest specific referees. Make this as simple as tapping a few names and faces. Of course, this requires that your app has access to a user’s contacts.

   1. Add a way for users to import contacts and provide a strong reason to do so. Maybe the product functions better with it, or maybe the referral incentive is reason enough!

   2. Sort the list to put the ‘best’ recommendations at the top. Use whatever signal you might have. At Superhuman, this is easy: we look at who the customer emails a lot. Other techniques include cross-referencing the contact list with your “signed up, but hasn’t paid” list, or prioritizing a pre-defined list of logos.

   3. If you’re extremely confident in the recommendations, add a “select all” checkbox, and pre-select it, removing all friction!

3. Suggest referees in targeted and specific contexts.

   1. For example, Kondo has a “refer” button in the sidebar when you look at a particular user’s profile, and in the conversation with that user.

      

4. Create a referral dashboard for users.

   1. A basic requirement is showing the list of referrals made, in progress, and completed. See early Dropbox for inspiration.

      

   2. The dashboard should make users feel good by showing total credit earned. It should make them take action by showing total credit on the table.

   3. Finally, it should have a “remind” button to nudge referees.

Step 5: Combine referrals with other channels: casual contact, social, community, influencer, affiliate, and add to team

At the heart of every great growth strategy is a collection of mutually reinforcing channels.

Referrals are often one of them.

1. Combine referrals with casual contact loops. A casual contact loop is any time a user using your product results in someone new seeing it. Think sharing a Loom, Figma, or DocuSign.

   1. Embed your user’s referral link into any content they might share using your app.

      1. For example, “Sent via Superhuman” email signature is each user’s referral link, letting them emit referral links and earn incentives just by using the app.

      2. It’s common to provide the ability to turn this off, but only on higher-tier plans.

2. Combine referrals with social.

   1. Make the referral link extremely easy for customers to post to social. Two clicks max: One to express intent, one to post.

   2. If you have user-generated content or brag-worthy usage statistics, have a low-friction way for users to share on social, and include the referral link.

3. Combine referrals with community.

   1. Create a public micro-site that identifies and thanks your top referrers. Reset this list every month to keep it fresh.

      1. Notify them and let them do their magic in bragging about it.

      2. Notify others to get them to compete for primacy.

   2. Create a private community for your top referrers.

4. Combine referrals with affiliate and influencer.

   1. Use the same referral program and infrastructure with affiliates and influencers. Build once, use thrice!

   2. Advertise these programs in your app to continuously recruit customers with large audiences. The ‘refer’ screen is a perfect place for this.

5. If you also have an ‘add to team’ flow to drive team expansion, make it work alongside your ‘refer’ flow.

   1. Start by deciding which to prioritize. Why? There will be situations where you can only pick one. Examples include limited pixels on screen or a finite ability to blast your customer base.

   2. In my experience, a dollar of team expansion is worth more than a dollar of new logo. In this case:

      1. Re-capture referrals as team adds where you think the user might want to grow their team, for example, if a user makes a same-domain referral like adam@acme.com referring beth@acme.com.

      2. Prompt referees to join existing teams, where relevant. Yes, this voids any promised incentive, but the growing team probably benefits from the new member joining, and the referee avoids payment forever by joining that team.

   3. Revisit any previous assumptions made about your referral program. For example:

      1. Deeply consider the copy. You might want to explicitly label them “Refer a friend” and “Invite a teammate” to clarify who each is for.

      2. Update suggestion lists. You might want to remove any contacts from the referral suggestion list that look like teammates.

Weird Things 🥴

• Don’t cap referrals to create a sense of artificial scarcity.

  • Referral curves are exponential. Your top referrer will do 100x the volume of your median referrer. Let whales be whales.

• If you’re running a waitlist, you might consider asking signups to refer to skip the wait.

  • This only works if you have extreme marketing aura and an extremely delightful product to justify the hack.

  • Not impossible, but very delicate.

• You might worry about incentive abuse and exploits.

  • Put sensible safeguards in place, but don’t spend too much time on this. It’s easy to get bogged down in imaginary issues.

What tactics did I miss? Comment or reply ✍️

Further Reading

• Tremendous’s blog is full of good resources, such as What is referral marketing and Getting started with referral marketing.

• Viral Loops has several great referral program teardowns, including Dropbox, Airbnb, and Uber.

Track AI visibility across ChatGPT, Perplexity, and other LLMs, monitor how often AI bots are crawling your site, and create content that LLMs love.

Get 20% off Profound Lite with code GAURAV20 here.

The ground has shifted

ChatGPT hit 500 million monthly active users in May 2025.

This was the same month that Google announced AI Mode, their version of ChatGPT, directly within Google search:

This is the defining moment. Google is admitting that LLMs are the search interface of the future.

Every brand is noticing the open-mouth “alligator effect” in their SEO metrics:

Impressions are up, and clicks are down. And AI is to thank — or blame, depending on how salty it makes you feel.

I see these patterns at startups I advise: ChatGPT and other LLMs are rapidly rising in “How Did You Hear About Us?”. Some of my clients have an astounding 30% of their weekly traffic from ChatGPT.

Just as important is the speed of change. For example, Vercel saw 10% of their signups come from ChatGPT in April 2025, up from 4.8% in March 2025, and 1% in October 2024.

ChatGPT and LLMs aside, most startups I work with see traditional search (i.e., Google) account for 30-50% of their signups. And as above, Google is shifting towards AI Overview and AI Mode.

Search has decisively shifted towards LLMs.

Traditional SEO is dying.

GEO (generative engine optimization — a mouthful, I know — I’m sorry, I didn’t name it) is here to stay.

What should you do about it?

First, we’ll debunk the top two misconceptions surrounding GEO. Then we’ll dive into what to do about GEO.

This post is for anyone responsible for acquisition: founders, growth, and marketers.

Subscribe now

Debunking common GEO misconceptions

Misconception 1: GEO is still small

Here are the facts:

1. Google’s AI Overviews appear at the top of over 50% of searches. This is up from 25% just 10 months ago (source).

2. ChatGPT is already 3% of Google’s traffic. This might surpass 10% by the end of this year, and could match Google within the next 5 years (source).

3. Google’s AI Mode is currently in beta, but if AI Overviews are any proxy, it will see rapid rollout and adoption.

GEO is already extremely important for any company whose customers find them online.

Misconception 2: GEO is the same as SEO

This is directionally true in that you need content with authority to feature in both traditional and LLM search.

But execution tactics differ.

To understand why, let’s compare traditional search and LLM search.

Difference 1: Words per query

The average traditional search is 3-5 words.

The average LLM search is extremely detailed and nuanced at 20+ words, not counting follow-ups. In traditional SEO, they would be “long tail searches”.

A traditional search strategy might manufacture content aimed at winning clicks for high-volume but low-signal traffic, like “AI notetaker”.

An LLM search strategy needs to maximize chances of being cited in a long tail search like “List the top 10 AI notetakers and give me their cost, pros, and cons”.

It must do this with rich and authoritative content with facts, benefits, advantages versus competitors, and specific use cases.

Difference 2: User results

Traditional search gives you a pile of blue links to explore.

LLMs provide one answer composed of many different sources, aiming to fully answer your question. You likely will not click through to multiple sites.

LLM traffic that does click through tends to convert extremely well. For example, Ahrefs sees LLMs represent only 0.5% of visitors but over 12% of signups. LLM search converts 12x better than traditional search.

Brands first need to win citations, then optimize websites for high-intent traffic.

Difference 3: Ranking factors

Kevin Indig, SEO expert, studied the types of content that rank well in LLMs.

He found that traditional SEO factors like backlinks, keywords, and total traffic were not significantly correlated with LLM citations.

Instead, LLMs pay attention to factors like domain authority, content comprehensiveness, and readability.

While GEO has directional overlap with SEO, the finer nuances of how to win GEO differ:

How to determine GEO priority

Hopefully you’re convinced by now that GEO is a priority.

If you’re still unsure, ask:

1. How fast are your customers adopting AI to search?

   1. Size traditional search. Use Google Keyword Planner, or third-party tools like Ahrefs or SEMRush. Assume half of those searches are influenced by AI Overviews.

   2. Size LLM search. Use Profound’s Conversation Explorer to quantify how often your problem space appears in LLMs. Currently available to Enterprise customers.

   3. Directly survey your customers and ask how frequently they search using LLMs like ChatGPT, Google AI Mode, and Perplexity.

2. How do customers currently find you?

   1. Ask “How did you hear about us?” (HDYHAU) in your new user journey. This reveals when users might have heard about you in one place, for example ChatGPT, even if your attribution doesn’t track that.

   2. Read more in this Growth by Gaurav essay.

Either of these should be a trigger for you to prioritize GEO:

1. If >10% of your customers search with AI.

2. If >5% of your customers already find you via AI.

Action plan for GEO

If you’ve decided GEO is a priority, here is a step-by-step playbook a single person on your team can follow.

1. Audit your foundations

Run a full technical GEO and SEO audit.

• Learn the rules. For example, see Step 8 in Aleyda Solis’ guide here.

• Speed up your site, nail meta tags, and page titles.

• Add structured data, kill broken links, and fix duplicate content.

2. Nail your brand positioning

Mixed messaging confuses AI and users.

• Clarify what you stand for. Define your core message and tone.

• Make every channel reflect it: Website, social, blog, and beyond.

• Tell your story the same way everywhere.

3. Identify and answer top customer questions

Answer questions early to win AI citations.

• Learn from sales, support, and search forums what questions customers ask before buying.

• Answer them better than anyone else.

• Predict emerging questions and answer before others do.

4. Refresh existing content

Turn old pages into new growth drivers.

• Audit old posts: Update statistics, fix links, and add depth.

• Update language to be more precise and descriptive to help AI.

• Improve structure, clarity, SEO tags, cross-links.

For an incredibly detailed guide, again refer to Aleyda Solis’ checklist here.

5. Create new authoritative content

Be the best answer on the internet.

• Publish detailed landing pages, competitor comparisons, how-to guides, and answers to top questions.

• Include real statistics, unique insights from subject-matter experts, and compelling case studies.

• To move fast, repurpose existing media. For example, cut and transcribe existing podcasts and video clips.

6. Go beyond your website

Build an AI-friendly footprint across the internet.

• Study what websites LLMs cite most often for prompts related to your business.

• Profound studied the top domains per LLM from 30M citations. The top 5 sources for ChatGPT and Google AI Overviews were Wikipedia, YouTube, Reddit, Quora, and LinkedIn. Beyond this, high-trust sites like Forbes, G2, and Gartner feature heavily.

• Build a web of content that features across these sites. Participate in forums, commission influencer content, publish op-eds, and more.

7. Monitor, measure, and optimize

Search is a compounding game.

• Track LLM visibility, share of voice, sentiment, clicks, and HDYHAU.

• Fill gaps with frequent new content and technical tune-ups.

• Revisit conversion optimization and retargeting strategies to make the most of your MOFU.

Timeline to get started on GEO

Last but not least, want a concrete timeline for your GEO efforts this quarter?

Remember, it’s not whether LLMs will affect your funnel. It’s when.

Forward this to a founder, growth person, or marketer.

Until next time.

For more:

• The Great Decoupling (or Why Your Clicks Are Down and Impressions Up) — Ryan Law, Ahrefs

• What content works well in LLMs? — Kevin Indig, Growth Memo

• AI Search Content Optimization Checklist — Aleyda Solis, SEOFOMO

• AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information — Nick Lafferty, Profound

• Homepage Traffic is Up 10.7% from AI Overviews and LLMs [50-Site Study] — Ross Hudgens, Siege Media

• Are AI Chatbots Replacing Search Engines? A 2-Year Data Study on Web Traffic — Sujan Sarkar, One Little Web

• 7 Steps to Reclaim Your Traffic with an Informed AI SEO Strategy — Will Melton, Xponent21

• Google’s AI Overviews Surpass 50% of Queries, Doubling Since August 2024 — Will Melton, Xponent21

Superpower is now out of beta. Join here and enter GAURAV at checkout for $50 off.

Founders, marketers, growth people1, we need to have an honest conversation about waitlists.

Early-stage startups frequently hit these questions:

• Should we do a waitlist?

• Will a waitlist boost growth?

• Will a waitlist make us cool, edgy; hype, vibey?

Reality: A waitlist is rarely the right thing for your startup.

Waitlists are permissible over short timeframes. They should exist for a few months at most. Anything longer and waitlists hurt conversion by >3x. The longer you keep it around, the worse these problems will become.

Ultimately, if you invoke a waitlist, you create a ticking time bomb that you must diffuse before it explodes.

Supply Constraints

When does a waitlist make sense?

Invoke a waitlist when you have a genuine supply constraint that prevents you from delivering your ideal customer experience.

That is it.

Nothing else about optics, brand, or hype should be your primary rationale.

What are the three forms of supply constraint?

1. Hard product constraint

There are constraints, often physical, that stop you from serving more customers.

Examples:

• Hardware businesses have a fixed number of units.

• Capacity businesses like airlines or concert venues have a limited number of seats.

• Marketplace businesses like Uber might not have enough drivers.

• Software businesses might not support a customer’s device or platform.

It is perfectly fine to waitlist customers that you’re just not able to serve. Practically speaking, you have no other choice.

2. Soft product constraint

Parts of the product experience aren’t where you want them to be.

Examples:

• Software products may not cover all desired use cases yet.

• Security companies might want to fix a backlog of known bugs.

• AI labs may want to ensure model safety and alignment before scaling.

It’s reasonable to have a waitlist that limits how many new customers you onboard until you meet the quality bar.

3. Service constraint

You don’t have enough teammates to deliver some service component of the customer experience.

Examples:

• Enterprise SaaS companies might not have enough implementation specialists to set up new customers.

• DTC companies might not have enough support operations to guarantee smooth customer experiences.

• Healthcare companies might not have enough clinicians to meet demand.

A waitlist that buys time to scale your team is very reasonable.

In all three cases, alleviating the supply constraint must be a top priority to drive sustained growth.

Waitlist Costs

Having explored legitimate reasons to invoke a waitlist, what are the downsides to running a waitlist?

1. Lost customer interest

Waitlists cause signups to lose interest.

Signups have the best likelihood of converting to customer the moment that they sign up.

A waitlist halts that momentum. A meaningful percentage of those signups will stop caring, pick a competitor, or become unreachable over time.

When I was in charge of Superhuman’s waitlist at the same time as its live funnel, I was able to compare conversion rates:

• Waitlist signup → paying customer (within 7 days): ~3%

• Live signup → paying customer (within 7 days): ~10%

The live funnel was >3x as performant as the waitlist funnel.

There were no other differences in the populations in each group. The only difference was the waitlist had been made to wait.

Furthermore, performance sharply decayed the older the waitlist was.

3% conversion to paying customer was for signups who had been on the waitlist for approximately 6 months.

Conversion halved for each year further back the waitlist ran, from 3% → 1.5% → 0.8%, and so on.

2. Foregone startup momentum

Momentum is oxygen for your startup. Momentum is everything.

A waitlist by its very definition slows momentum.

Any momentum loss comes at great cost — often difficult to quantify.

Who knows which use cases you could have discovered, what sales you could have landed, or hires you could have closed, if it weren’t for the waitlist?

These costs should be weighed heavily against any potential upside.

3. Distraction maintaining the apparatus

Maintaining a waitlist is work.

You spend precious brain cycles managing a queue when you could have been building and shipping.

The waitlist becomes an entity unto itself; a beast that needs feeding; a tail wagging the dog.

I regularly see this in my advising. When a startup begins to optimize the waitlist it is clear they are falling into this trap.

They might ask questions like: “Should we send nurture emails, run webinars, and host contests to sustain waitlist interest?”.

These questions come from a good place: A sense of duty to potential future customers. If we don’t keep the waitlist engaged, won’t they get upset? And if they are upset, won’t they think badly of us, harming future growth?

But the reality is that customers signing up for a service don’t want to be kept entertained.

They just want the product.

There is rarely a message worth sending a waitlisted signup other than an invite to the product.

The other common question indicating the startup is getting distracted is: “Should we hire somebody to keep growing the waitlist?”.

This comes from a desire to hit growth milestones at the same time as building the product.

But “size of waitlist” is little more than a vanity metric; a weak proxy for true demand, retention, and revenue.

Local Optima

Here are the most common (bad) reasons to run a waitlist.

Be careful any time you hear these arguments.

They are all legitimate reasons. But I consider them local optimizations when faced with a supply constraint.

They are not ways that a startup should globally optimize growth.

1. A waitlist enhances user experience

Humans love anticipation.

We talk excitedly about things we are waiting for. We are taught that “good things take time.”

This effect is twofold.

1. Those on the inside feel special for making it past the waitlist

2. Those on the outside feel anticipation for eventually getting in (the velvet rope effect)

If you’re running a waitlist, it’s smart to play into these feelings. There are all sorts of fun and games to be had:

• Brag about the size of your waitlist

• Let outsiders see their waitlist position

• Give insiders a way to prioritize outsiders

• Limit the number of outsiders that insiders can bring in

• Allow outsiders to skip the waitlist in exchange for acts of virality

But these games don’t last forever. They do eventually get stale. You should kill off your waitlist well ahead of this fatigue setting in.

2. A waitlist enhances our brand

We all know products where limited supply elevates brand perception. Consider Supreme, Air Jordans, early Tesla.

In some cases, ‘limited edition’ is a long-term growth strategy, especially in fashion or luxury goods.

However, this is not the case for most scaling businesses.

Eventually you need to ramp up production, or customers will question your competence — eroding the brand you worked so hard to build.

3. A waitlist lets us nurture prospects, increasing conversion

The hypothesis is that a waitlist lets companies educate prospects.

As a result, more may convert to paying customer than if they had been able to purchase right away.

The fallacy in this thinking is forcing everyone to wait in order to educate some prospects.

A better approach is to give high-intent prospects an immediate pathway to customer while capturing contact details for low-intent prospects.

Then, low-intent prospects can be intensely nurtured toward conversion.

4. A waitlist creates hype, increasing reach

The hypothesis is that the notoriety of a waitlist might generate more signups than any other marketing effort.

The short-term hype of a waitlist might reach more customers, but this effect lasts for at most a few months to a year — and sometimes much less.

Ultimately, if the product is genuinely good, then authentic word of mouth is a far more potent growth engine than any waitlist tactic.

5. A waitlist can manufacture a specific growth curve

A waitlist might give a startup the ability to create an impressive looking growth curve.

Specifically, the startup can allow 5 customers off the waitlist this week, 10 customers next week, 20 the week after, and so on.

The usual reason founders might want to do this is for an upcoming fundraise. This is fine as a short-term hack, but it must be time-bound.

Bonus: All of the above

What if the combined effect of all the above reasons yields a globally better outcome than giving customers immediate access?

This might be true for a brief amount of time. But after that window, we stray into the realm of fuzzy thinking.

A collection of weak reasons for a waitlist can easily obfuscate the core truth that it’s slowing down growth.

Case studies

Superhuman

I ran Superhuman’s waitlist from start to finish.

We had hard product constraints, soft product constraints, and service constraints.

The hard product constraints were not supporting all mail providers and mobile devices.

The soft product constraints were not having all the features customers expected. We also had a strong commitment to deliver incredible speed and reliability to every customer.

The service constraint was our desire to manage all customer interactions exceptionally well as a core differentiator versus incumbents.

Highlights

We pulled every lever to make our waitlist fun!

We told existing customers they could refer to help prospects skip the waitlist. We told prospects they just needed a referral from existing customers.

As a result, a constant stream of referral match-making unfolded over email, on Twitter, and in real life.

The waitlist ultimately accrued well over 500,000 signups and drove significant brand momentum.

Lowlights

Our waitlist’s success resulted in it sticking around far longer than the supply constraints.

At first, this was fine. It wasn’t broken, so we didn’t fix it.

But the shine wore off over time. Little by little, the waitlist created more fatigue than excitement with prospective customers.

The shift was subtle. Like a frog in boiling water, it was difficult to notice the exact moment it became a burden. Some even saw it as a key part of our brand.

Removing it was technically trivial — but, to my surprise, culturally momentous.

OpenAI DALL·E

OpenAI opened a waitlist for DALL·E 2 in April 2022.

The underlying reasons were a blend of hard and soft product constraints: GPUs, cost, and a desire to responsibly deploy new AI technology.

Highlights

Website, email, and Twitter/X messaging around the waitlist’s purpose, timing, and mechanics was extremely clear. The official OpenAI account and key individuals like Greg Brockman led from the front with clarity on invite status.

Most importantly, OpenAI sunset the waitlist just 5 months after invoking it.

Lowlights

Competing tools like MidJourney and Stable Diffusion gained huge traction during this period: X was awash with compelling output from DALL·E 2 alternatives.

One can’t help but wonder whether a direct release might have performed better than a waitlist? See also: Images in ChatGPT 4o.

Superpower

Superpower launched a waitlist in 2024 for their all-in-one health super app.

Superpower’s reasons for a waitlist were hard product constraints and service constraints. The service includes a blood draw which involves human and shipping logistics, as well as clinician support to interpret results and take action.

Highlights

The waitlist quickly amassed over 150,000 signups. This gave early market validation without the company having to fully solve the product and service supply chain.

The team artfully positioned this signal. This drove further brand awareness, helped close crucial hires, and allowed the team to raise capital.

And best of all, the waitlist is no more! Sign up here, use GAURAV at checkout for $50 off.

Lowlights

Like DALL·E 2, Superpower could have provided more clarity on why the waitlist was in place, invite mechanics, and timeframe. More transparency would likely have banked even more trust with future customers.

No startup’s product is ready on day 1.

Waitlists help capture demand while ramping supply. And like any growth tactic, waitlists can be optimized.

However, startups with waitlists should intensely focus on fixing their supply constraints so that they can kill their waitlist.

Know a founder contemplating a waitlist or running one too long? Forward this and help them out.

What did I miss? What was your favorite waitlist experience? Hit reply or add a comment.

If you enjoy this essay, you’ll enjoy Wispr Flow. Get a free month of Pro here.

Product thinkers go deep within their space.

Deeper than their product shows. Deeper than 99.9% of others.

I did this at Superhuman: I obsessed over team communication.

Every startup faces these questions:

• How should we communicate? Which tools should we use? When should we Slack versus email versus talk?

• How responsive should we be? What response time is too slow or too fast? Is there even such a thing as too fast?

• How should communication work for remote teams? Teammates constantly in meetings? OOO teammates?

These all resolve to one core question: What helps the team move the fastest?

You’d think these are solved problems. After all, every startup faces them.

Amazingly, no.

Reality: Startups stumble onward. Local optima emerge. Tension brews across teams.

Some startups claim intentionality. But they, too, land on suboptimal equilibria.

A classic example is startups mandating that 100% of communication goes through Slack. Sure, sounds good on paper.

But speak to any individual to see that they’re miserably drowning in an endless onslaught of comms.

A minority of startups think deeply about the problem. Shining examples include Stripe’s email transparency, and scaling email transparency, and Amazon’s Write Like An Amazonian.

But these cases are vanishingly rare.

Why is startup communication so hard?

Part of the problem is human inhibition. Fear of embarrassment, mostly.

Sometimes it’s a skill issue. A junior hire who hasn’t yet learned how to write succinctly. An executive who is chronically slow to respond.

And part of it is systematic: A simple absence of guidelines.

Subscribe now

This essay documents foundational principles of good startup communication.

These principles should be of practical use to you.

You could personally start using them today. Or you could institute them for your entire team.

Principles of Good Communication

1. Effort Conservation Principle

Good communication conserves effort as much as possible.

Teammate attention is a finite resource. It is perhaps the team’s most precious resource.

Senders should be conscious of their own effort spent communicating, as well as the effort they demand of others who receive their communication.

Senders should seek to minimize both.

2. Efficient Expression Principle

Good communication is concise, but not at the expense of clarity.

The Efficient Expression Principle flows directly from the Effort Conservation Principle.

Verbose or unclear communication wastes teammate attention that could have been better spent elsewhere.

Concise and clear communication minimizes effort.

3. Appropriate Artifact Principle

Good communication uses appropriate artifacts to boost message salience.

Example artifacts might include:

• A Notion page

• A Loom voiceover

• A spreadsheet

• A quick diagram

The Appropriate Artifact Principle builds on the Efficient Expression Principle: Let the artifact do the talking so that you don’t have to.

4. Recipient Specificity Principle

Good communication explicitly addresses a set of recipients. Good communication is never vague on who is being addressed.

Recipient specificity ideally uses in-built tag functionality: “@” in most tools.

This principle requires that teammates understand each other’s roles and responsibilities well enough to know who to address.

This principle should be upheld for all communication types: Sharing information, posing a question, requesting work, and beyond.

And it should be upheld in all forums: Email, Slack, Notion, Drive, Figma, GitHub, Linear, and beyond.

5. Recipient Minimization Principle

Good communication does not just specify recipients, but actively minimizes the number of recipients required to consume information.

Good communication never requires an individual to consume information that they do not need.

The Recipient Minimization Principle flows directly from the Effort Conservation Principle combined with the Recipient Specificity Principle.

6. Information Access Principle

Good communication defaults to public channels for all valuable and non-sensitive information. It is bad communication to sequester such information in private silos.

Crucially, per the Recipient Specificity Principle and Recipient Minimization Principle, the Information Access Principle does not suggest that every teammate is forced to read every communication.

It merely requires that individuals are able to access information if they want.

A practical example of the Information Access Principle would be starting a growth-related conversation in the public #growth channel, rather than DMing the growth PM.

Of course, only the relevant individual(s) should be tagged in the channel.

There is a critical consequence of the Information Access Principle:

For most, this comes as a huge relief: Freedom from drowning in an endless sea of comms.

The longer-term benefit of the Information Access Principle is that information can be easily discovered by others.

This safeguards against the reality that roles and responsibilities are ever-changing, such that information may be relevant for a different set of individuals tomorrow than today.

The Information Access Principle requires that individuals proactively shift conversations into public forums, rather than letting them proliferate in private.

What defines valuable and non-sensitive communication?

Valuable is anything that might further the company.

Sensitive requires private discussion because public discussion would distract the team or create undue risk. This might include personnel matters like feedback, hiring, or compensation. It might involve private information like API keys or PII. Or it might relate to delicate projects such as a fundraise or M&A.

Sensitive topics should be the minority of a typical teammate’s total communication. However, some roles like Recruiting or Corp Dev might have a higher share relative to other roles.

7. Independent Discovery Principle

Good communication recommends that individuals should not request information that can be easily looked up.

The Independent Discovery Principle is unlocked by the Information Access Principle.

Consider a fast-growing organization.

There will be an ever-growing body of knowledge. There will also be an ever-growing number of individuals who need said knowledge.

The team suffers if teammates disrupt each other every time information is needed that could have been independently discovered.

Therefore, individuals should seek to independently discover information they want, and only ask if it is not found after a quick look.

Finally, after receiving the answer, individuals should make high-salience information easier to find in the future. This might look like updating keywords on existing information to make it appear faster when searching.

8. Thorough Response Principle

Good communication responds to every point made in a prior communication.

Furthermore, good communication makes clear precisely which part of a message is being responded to, for maximum clarity.

This ideally uses in-built quote functionality, for example typing “>” in most rich-text editors, or a feature like Quick Quote in Superhuman:

The Thorough Response Principle upholds the Effort Conservation Principle: It is extremely burdensome for a sender to double-check that all of their points were handled. This principle relieves senders of that burden.

9. Closed Loop Principle

Good communication closes the loop on all open topics of discussion. Good communication never leaves threads hanging.

The Closed Loop Principle is a special case of the Thorough Response Principle, for tasks that naturally involve a delay.

Of course, it is not practical to chase down every idea to completion in the sense of actually completing the work.

Therefore it is reasonable and even encouraged to close loops with the simple message: “I’m deprioritizing this topic.”

10. Channel Norms Principle

Good communication depends on teams defining shared channel norms. Each teammate should understand those norms so they can communicate with intentionality.

Teammates should never feel like they are picking randomly when deciding which channel to use.

What is a channel? The concept can exist at several levels of abstraction.

For example, teams can communicate in different platforms like email or Slack.

Teams can also communicate in different spaces within platforms, such as different Slack Channels.

Channel norms should exist for all relevant levels of abstraction.

Channel norms should be clearly written down, for example in a company Wiki or handbook.

They should be considered required reading for all new teammates, and any meaningful updates should be pushed to the entire team.

The most important norm to agree on is a responsiveness level per channel. Responsiveness levels should be set to the fastest possible turnaround that is not disruptive to meaningful work.

It is better to respond faster than slower. If information traveling through an organization represents small increments of progress, then slow responses hold back that progress.

But at the same time, communication can slow down other forms of progress, notably individuals making strides forward in their own work.

Responding to every Slack within 1 second would be disastrous for meaningful individual work, despite being excellent at ensuring that collaboration never stalls.

Different channels also naturally tilt towards different responsiveness expectations. The form factor of Slack lends itself to semi-synchronous discussion, whereas the form factor of email promotes more reflective asynchronous exchange of ideas.

So what should responsiveness norms be?

A good starting point is that everybody responds to Slack within 3 working hours, and to email within 1 working day.

To uphold these, equip your team with appropriate tools like Wispr Flow and Superhuman.

11. Channel Nuance Principle

Good communication permits individuals and sub-teams to adopt more nuanced norms per channel, so long as they do not subvert the team’s wider channel norms.

Let’s say the team agrees that emails should be read and responded to within 1 working day. It is obviously good for the organization if some individuals and sub-teams choose to respond faster.

On the other hand: It would not be good if an individual or sub-team consistently responded slower.

In line with the Information Access Principle, individuals and sub-teams should transparently share any channel nuances they have adopted, to make it easier for others to communicate with them.

However, because nuances should never subvert the wider company’s norms, these should not be considered required reading for others.

Individuals should be able to freely communicate without needing to learn everyone’s peculiarities and preferences.

12. Acknowledgment Principle

Good communication has recipients send acknowledgment of receipt as fast as reasonably possible. Acknowledgments inform senders that communications have been seen.

Valid forms of an acknowledgment include a Slack emoji reaction, an iMessage Tapback, a blue double-checkmark in WhatsApp, or simply responding “ack”.

To minimize disruption, acknowledgment must take the recipient as little time as possible.

Acknowledgment combats the inherent uncertainty in asynchronous communication that arises from a lack of non-verbal cues.

When senders cannot physically see a recipient’s subtle head nod or thumbs up, efficiently communicating “I saw this and it’s on my radar” goes a long way.

13. Notification Sovereignty Principle

Good communication requires each individual to take full responsibility for their notifications. Notifications include alerts, banners, badges, counts, sounds, email updates, and more.

The Notification Sovereignty Principle relieves senders of the need to consider the notification settings of their recipient. After all, that is the responsibility of the recipient.

In turn, senders do not need to feel responsible for the consequences of those notifications. This frees senders from the mental overhead of constantly having to consider the timing of their communication.

Senders can communicate whenever they want, which is usually as soon as they can. This prevents communication from being arbitrarily delayed for fear of disturbing the recipient.

In the extreme, the Notification Sovereignty Principle implies it is perfectly reasonable to send communications at 3 AM or the in middle of the weekend if desired. It should also be perfectly fine to tag a teammate who is Out Of Office.

Sender mental overhead and arbitrary message delay are significant taxes in organizations of all sizes.

The Notification Sovereignty Principle nearly fully eliminates these taxes.

14. Notification Suppression Principle

As a crucial counterpart to the Notification Sovereignty Principle, good communication encourages teammates to suppress notifications so that they are minimally disruptive.

Unfortunately, most communication tools have the exact opposite as their default: Red badges, stressful numbers, moving banners, noisy sounds, lock-screen widgets, and more.

Most teams, particularly those whose work depends on a maker schedule, unlock improved productivity when individuals minimize notification disruption.

How can a team uphold the Notification Suppression Principle yet also be responsive?

Individuals should proactively check communication channels. This should not happen in response to notifications. It should instead align with their team’s Channel Norms Principle.

For example, if the team has agreed that emails will be responded to within 1 working day, then each individual should proactively check their email at least once per working day.

All individuals should carve out at least one “in case of emergency” channel where notifications will reliably disrupt them. Usually, this is “walk over to my desk” or “call me”. Senders should use these channels sparingly.

Similar to the Channel Nuance Principle, some roles might elect into a higher degree of notification disruption. For example, an engineer would reasonably allow PagerDuty to always make a noise. A salesperson might reasonably want email banner notifications.

Individuals may communicate their specific notification settings to others. However, thanks to the Notification Sovereignty Principle, senders should not be required to understand this information before communicating.

15. Responsible Recipient Principle

Good communication requires that all recipients responsibly handle their inbound communications.

Specifically: They should catch up on all messages since the last time they were online. This should be upheld regardless of how long they were offline.

The Responsible Recipient Principle relieves senders of the need to chase down replies.

It further relieves senders from having to create systems to facilitate said chasing, such as setting reminders on outbound communications.

If upheld, this principle releases senders from a significant amount of mental overhead.

Adoption

Some of these principles are uncontroversial, like the Effort Conservation Principle or Efficient Expression Principle. Any reasonable person thinking about them for a moment would buy in and adapt behavior.

Many are a question of skill, like the Appropriate Artifact Principle or Responsible Recipient Principle. Luckily, teams have self-reinforcing mechanisms that strengthen the adoption of these principles. The more individuals uphold them, the more effective they become in their role, deepening their adoption while influencing those around them to do the same.

But some resemble a Prisoner’s Dilemma, like the Information Access Principle or Channel Norms Principle.

When everyone participates, everyone benefits. But as organizations scale, individuals might subvert these principles for personal gain, even if it hurts the team.

Consider the executive who hoards information to build political influence, subverting the Information Access Principle. Or the salesperson who constantly swings by desks to move their deal forward even if majorly disruptive to others, subverting the Channel Norms Principle.

Of course, if everyone adopts these habits, everyone suffers.

While there is no silver bullet for adoption, several factors go a long way:

1. Hire intentional individuals. Seek characteristics like “looks before leaps”, “capable of deeply engaging with ideas”, and “thinks from first principles”.

2. Train the team. Write down norms and expectations, and make them required reading for all existing and future teammates.

3. Foster a culture where everyone has permission, and indeed, feels an obligation to remind others of the principles. Individuals should focus on the mutual benefit of the principles versus dogmatic rule adoption.

4. Encourage reflection. Is the system breaking? Revisit and iterate.

5. Pick good tools and sensible defaults. For example, at Superhuman, we discourage using Google Docs due to its inherent low discoverability, and instead recommend Notion.
   
   We furthermore encourage new teammates to carve out space for non-sensitive thoughts. This default helps teammates rarely make the mistake of failing to invite collaborators to their pages — upholding the Effort Conservation Principle and the Information Access Principle.

AI

It is clear that AI has the potential to help vast numbers of people uphold good communication principles.

Tools already help: Glean accelerates discovery; Superhuman boosts responsiveness; Wispr Flow unlocks speed; Grammarly sharpens writing.

But can tools go further?

What if AI noticed when you forgot to reply to every point, violating the Thorough Response Principle?

Or prompted you to exit Slack DMs and move to a channel, to uphold the Information Access Principle?

And why stop there? These principles don’t just apply to human-to-human communication. They are just as relevant for AI-to-human communication.

AI agents — especially those showing up in our workspaces like Devin, Lindy, and Risotto — should be programmed with good communication principles in mind.

They should instantly acknowledge messages, respond thoroughly to all points, be specific in who they are addressing, and more.

If anything, well-designed AI agents may end up teaching us: Exhibiting communication so principled that we improve through pattern recognition and imitation.

Thank you Matt Waters for draft input, and for your intellectual fortitude in helping me interrogate the most arcane minutiae of good communication principles.

If you find this write-up useful, use Default to set up “How did you hear about us?” today.

There’s one question every startup should ask new customers.

But it’s astounding that so few do.

That question: “How did you hear about us?”

There are all sorts of ways to determine which channels are working: Referrer attributes, URL params, affiliate codes, Stripe coupons, Google Analytics, Google Search Console, CDN logs, Fivetran imports, DBT ELTs, ad platform dashboards, and much more.

That data can be assembled into an attribution model as gloriously complex as you can imagine.

But “How did you hear about us?”, or HDYHAU1, is a machete through the jungle of trying to figure out what's driving your growth.

Instead of spending weeks assembling disparate data sources, you can stand up HDYHAU in less than a day, and start using same week.

Every startup serious about growth should ask this simple question during signup.

Read this if you’re a founder, marketer, or in growth.

Subscribe now

Requirements

Pre-defined options or a free text box?

TLDR: Pre-defined options. Include an optional free text box for added color.

Let’s explore this more comprehensively.

We should work back from the decisions that will be made. HDYHAU unlocks two grains of decision:

1. Strategic: Entire channels to further invest in or pull back on.

2. Tactical: Niche opportunities to further explore, such as a podcast or viral tweet that mentioned you.

Most of the value is in strategic decisions: For example, deciding to double down on paid and organic search, because a surprisingly high percentage of customers find you by searching.

Remaining value is in tactical decisions: For example, noting that there's a growing amount of “ChatGPT” in your "Search" bucket, so you might want to start investing in AEO (AI Engine Optimization), beyond SEO and SEM.

Given the strategic decisions you need to make, you'll want to periodically review a stacked bar chart2 that will start to look like this:

To make tactical decisions you will occasionally want to investigate more deeply to get a feel for underlying details, nuances, and trends.

What is the best way to get this data?

• Pre-defined options most rapidly get you to the visual. The data collected is the data charted.

• An optional free text box adds further signal and nuance.

Purists might argue that pre-defined options are a cludge and that all grains of signal can be extracted from a free text box.

While it is true that LLMs can mostly aggregate free text entries into broad buckets, this is a level of ongoing data manipulation beyond what most startups should spend time on when they're very early.

That said, an off-the-shelf LLM solution that is 95%+ reliable will exist within the next year. At that time, this recommendation will change.

How many options? What should they be?

TLDR: Anywhere from 5-10 works great.

They should more or less be:

• Search

• Worth of mouth

• Social Media

• Content

• Ads

• Other

For each area: Break out sub-channels if they are individually important to track:

• Search might usefully be broken down into Google, Bing, ChatGPT, etc.

• Word of mouth might split into Friend, Family, Colleague, etc.

• Social Media might live across Instagram, X, TikTok, etc.

• Content might be YouTube, newsletter, podcast, blog, etc.

• Ads might be digital, broadcast, out-of-home, etc.

Of course, include any further but more niche channels that are relevant to your startup, such as influencers, customer review sites, events, etc.

Last, make sure you randomize the order to avoid the top option being over-represented.

Mandatory or optional?

TLDR: Mandatory.

While optional is fine, it will result in low coverage and data skew.

Lower intent channels like paid social will end up under-represented if the question is optional. And you won't know by how much.

Why flirt with ambiguity? Just make it mandatory: It takes customers five seconds.

Perhaps most important to dispel is the myth that asking people where they heard about you is going to destroy your signup conversion rate.

Any customer so low intent that they drop out at this step was never going to be a good customer in the first place.

I'm gathering data. Now what?

Visualize

Create a stacked bar over time.

You should look at two views: Absolute and Relative stacked to 100%.

Together they will look like this:

Make edits within the first few days of gathering data:

1. Get rid of options that are <5% by combining them (unless, of course, monitoring them is key to your strategy.)

2. Split apart options that are >50%.

3. Create options to keep “Other” to <5%.

The goal is to get a view that is detailed enough that you can make decisions, but coarse enough that it doesn't take forever to decipher.

Understand

First: Is the data what you expected? Where are numbers bigger or smaller than what you expected?

Most founders I advise are surprised at how “quiet” channels like Search are so big, and how “noisy” channels like Social Media are so small. Reflect on what you’ve learned from the data.

Next: If you’re gathering free text data, spend time more deeply exploring the data. Use an LLM or tool like Default to aggregate the data and double check your buckets.

Are there any notable sub-channels? Do any specific names get mentioned a lot?

Build a deep mental map of what is working for your company.

Quick Wins

If this is your first time looking at this data, chances are you are early in the journey of creating your marketing strategy. If so, there are probably some immediate quick wins:

In the context of HDYHAU, high volume is any channel above 20%.

• Low cost, high volume: Double down
  Deepen your investment in these channels and see how much volume you can drive at acceptable costs.

• Low cost, low volume: Autopilot
  Let these channels hum along until you have enough capacity to drive them more.

• High cost, high volume: Optimize
  These channels are working, but you can probably trim cost as you continue to push volume.

• High cost, low volume: Kill off
  These channels are not working: Get rid of them.

Revisit

There are two lenses through which you’ll want to revisit HDYHAU:

1. New cohorts over time
   Set up a cadence of glancing at HDYHAU every week or month (depending on how fast-moving your funnel is). Pay attention to any major channel shifts — expected or unexpected.

2. Older cohorts aging
   Over time you’ll build a robust picture of which channels lead to customers with the best and worst activation, monetization, retention, expansion, and more.

The latter is particularly illuminating. Some channels might be in the sweet spot of low cost, scalable, and result in the best downstream customer metrics.

Of course, it is extremely worthwhile to double down on these channels!

(Hint: it will be Word of Mouth.)

Limitations

Supplementation

HDYHAU is extremely useful, but it is far from a panacea.

Rand Fishkin outlines 2 criticisms here.

1. Human memory is fallible
   Results might not accurately represent reality.

2. Answers won't identify specific parts of the marketing journey
   HDYHAU will not distinguish between when a new customer first heard about you (creating initial awareness) vs. when they most recently heard about you (resulting in them taking action), for example.

These are valid concerns to keep in mind.

That said, I vouch that for 90%+ of startups, the errors introduced from these inaccuracies are significantly less than the errors introduced by not having any visibility whatsoever — which is where most startups start.

Rand rightly argues for two further components to your attribution: Customer interviews, and passively collected data at scale.

By the time you are investing serious money in marketing — roughly speaking, over $50,000 per month — these should be “next up” on your marketing ops to-do list.

With a broader data pipeline in place, HDYHAU can be usefully merged into your attribution algorithm.

For example, between passive data collection and HDYHAU, you might learn the following for a single customer:

• First touch: Referral from existing customer.

• Middle touch: Clicked retargeting ad.

• Last touch: Clicked organic search result.

• HDYHAU: Social media.

Each startup might view this data in different ways, but a reasonable approach could be to allocate weight to the three channels of “word of mouth”, “paid social”, and “SEO” — with extra weight placed on “paid social”, thanks to HDYHAU.

Bigger picture

Rand also pushes on a higher altitude strategic consideration:

The people who didn’t reach you, the people who never heard about your company, can’t possibly tell you what they pay attention to that you’re missing out on.

So if you use “how did you hear about us?” to inform your marketing efforts, what will happen?

You will only ever invest in the kinds of marketing that have already been successful for you in the past. And you won’t find new channels and sources.

This is correct. So, once you’ve done the quick wins, take a step back and look at HDYHAU as one of several inputs to your marketing strategy. Consider running an exercise like The Unfair Advantages Framework with HDYHAU in mind.

Growth by Gaurav

The Unfair Advantages framework

Every startup must figure out: “how do we grow…

Read more

a year ago · 21 likes · Gaurav Vohra

And remember, the beauty of HDYHAU is it is:

• Something you can stand up in less than a day.

• Something you can start using same week.

Further reading:

• Don’t reinvent user profiling during onboarding. This is what it should look like — Elena Verna

• The 3 Big Problems with Asking “How Did You Hear About Us?” — Rand Fishkin

• The Unfair Advantages Framework — Gaurav Vohra

• The State of Booking Inbound Demos: Insights from 51,219 Meetings — Default.com

Thanks to Preet Anand for draft input.

The most important question for every startup: How long will customers stay?

If your hard-won customers slip away you have a leaky bucket, stalling ARR. But keep them, even grow them, and ARR accelerates.

Retention has always been important, but it's taking on new urgency for founders. AI hype cycles are getting larger, resulting in "vibe revenue" and "annual curiosity revenue."

Startups need to be dialed in on retention now more than ever. And of all the levers a founder can pull to improve retention, onboarding may be the most effective.

The moment a new customer touches your product, a clock starts ticking. For subscription products, the majority of usage drop-off happens within the first few days. It's most disastrous on mobile: The average app loses 77% of its daily active users within the first 3 days after installation.

Customers make up their minds on whether to return within minutes of encountering a new product. So founders, product, and growth teams rightly obsess over their onboarding.

As they do, they run into the same questions:

• What is the best approach to onboarding?

• Should I onboard customers myself? Is that scalable?

• When should I invest in self-serve onboarding? How do we make it special for customers?

I've spent the last decade obsessing over these questions — both at Superhuman and in my role advising 30+ startups. These experiences have shaped a specific point of view on how startups should approach onboarding. These lessons will help today's founders who are troubleshooting retention.

Many have written about Superhuman's unique approaches to product and growth. This viral essay on product-market fit from Rahul Vohra, Superhuman's CEO, published on The Review, is a good primer, as is this write-up by Kevin Kwok. Meanwhile, this teardown from growth.design is a great deep-dive into Superhuman's human-led onboarding.

But the more detailed story of how we scaled white-glove, human-led onboarding — then transitioned to product-led — has never fully been told.

I spent five years on Superhuman's first phase, starting, and then scaling our human-led onboarding. In this phase, we might have spent more time onboarding customers 1:1 than any other startup in history. I personally onboarded hundreds of customers — from the cofounder of Instagram, to Justin Bieber’s manager — before hiring our first Onboarding Specialist. At peak, we had dozens of Onboarding Specialists, who onboarded thousands of paying customers per week, and tens of thousands per year.

Then, to achieve the next level of scale, I spent another three years building the in-product onboarding, transitioning to fully self-serve. We migrated all of the hard-earned lessons into scalable product experiences. And we re-focused white-glove time to our highest-value prospects and B2B customers.

Our focus on human-led onboarding was unconventional, but we weren't alone. For example, Clay's Varun Anand recently recounted a similarly opinionated human-led strategy with their early "reverse demo."

In the age of AI, a human-led approach might seem more counterintuitive than ever. Yes, tools can help you build a self-serve onboarding in a day. But startups that invest in human-led onboardings build a sharp edge.

Each conversation is an opportunity to develop a deeper, more intimate understanding of your customers. And companies with deep understanding — from real conversations — have an advantage that no LLM can replicate.

When you begin scaling, you'll understand your customer's first mile deeper than anybody else in the world.

This essay is for you if you're an early-stage founder, product, or growth hire, working on product onboarding.

If you're in product or growth at a later-stage company, this essay likely has nuggets for you. And many of the lessons also apply to high-velocity inbound sales and customer success.

I'll cover the biggest lessons from Superhuman's journey, including:

• Part 1: Do things that don't scale. Why every startup should practice human-led onboarding, a detailed cost-benefit analysis with actual figures from Superhuman, and a framework to figure out when to transition to self-serve.

• Part 2: Unconventional wisdom for productizing onboarding. Unconventional wisdom on building self-serve onboarding, with inspiration from real life and video games to create memorable experiences that are opinionated, interruptive, and interactive — and overcoming roadblocks that might stand in your way.

Part 1: Do things that don't scale

I believe every founder should spend time manually onboarding their earliest customers. This advice is also true for product owners at later-stage companies launching new products.

Even if you couldn't possibly scale human-led onboarding, you should do it at the beginning.

The specific questions become: What should you focus on? And how long should you do it for?

Let's unpack the first question.

Five reasons to practice human-led onboarding

1) Customer success: world-class customer experiences

This is the most powerful reason of all.

Startups live and die by customer retention, and retention starts with activation.

Nothing activates a customer better than manually onboarding them to your product.

At Superhuman, we took inspiration from Apple's Genius Bar. You might walk into an Apple store expecting to buy a new MacBook Pro to watch Disney+ with your kids. But through a guided conversation, you'd walk out with a much better solution: a fully set up iPad, a kid-friendly case, and the knowledge of how to use it.

We also deeply considered premium experiences such as 5-star hotels. You might check in, exhausted, not knowing where to go next. But the concierge can pleasantly guide you to your room, or the spa, bar, or beyond.

As the founder or product leader, just like the Genius Bar staff or hotel concierge, you know the product best. You know every feature and every bug. You can anticipate every moment of delight and every sharp edge.

In other words, you can craft the perfect experience for each customer.

Let's examine a classical framework for activation, from Reforge's Retention Series:

A human-led onboarding can push every new customer to make it through 'Signed up,' 'Setup Moment,' and 'Aha Moment.' If done well, it can set them up with 'Habit Moment,' and put them well on their way toward 'Engaged.' And it likely can be done in half an hour or less.

Over 65% of new customers fully transitioned their email to Superhuman after their human-led onboarding.

This was more than double what we observed by simply dropping users into the product, absent any in-app onboarding, and letting them figure it out themselves.

What indicators suggest customers may especially benefit from human-led onboarding?

Look for the following four attributes.

A) Changing a pre-existing workflow

Maybe the customer gets the job done today with different tools. Maybe they have existing data that needs migrating. Maybe they just have strong opinions.

If this is the case, they need to change their pre-existing workflow to adopt your product.

90%+ of early Superhuman customers had been using Gmail for over 10 years. They came with a tremendous amount of email baggage. It was common to see 100,000+ emails in the inbox, 10,000+ unread, and no good workflow for handling email.

Humans are exceptional at pattern-matching to swiftly identify issues and propose solutions. We could rapidly diagnose each customer's current setup in our human-led onboarding. Then, we could provide exact migration steps to persuade them to adopt a new and better workflow.

B) Learning a new physical paradigm

Perhaps your product demands that customers build new muscle memory. Or maybe it introduces new surface areas, never before seen.

Products with this property might have new interaction modes. Consider apps taking user input through novel swipe gestures, voice, movement, or image recognition. AI companies focusing on multimodality or robotics would fall into this bucket.

As an example, Superhuman was famous for popularizing keyboard shortcuts for everything.

Learning keyboard shortcuts is a very unusual experience. Like a piano lesson, our customers benefitted from a real human encouraging them to try new interactions, helping them turn awkward physical movement into fluid muscle memory.

C) A desire to see specific functionality

Your product might exist in a space with a wide variety of jobs to be done. This might be because the incumbents support a huge number of use cases, such as Superhuman tackling email or VimCal tackling calendar. Or maybe your product flexibly supports hundreds of potential workflows. Consider, for example, go-to-market tools like Clay or Iterable that have the potential to become company infrastructure.

If this is the case, customers likely have different features they want to see. And humans can jump straight to those features.

At Superhuman, an onboarding for a busy founder with an overflowing inbox might prioritize Auto Labels, Split Inbox, and Remind Me. Meanwhile, an onboarding for an outbound salesperson might deep dive into Write With AI, Snippets, and Send Later.

Personalization shortens time-to-value from days to minutes. Customers are delighted at how quickly they discover their new favorite feature and are more likely to become engaged.

D) A need to build repeated habits

Finally, your product might require that customers commit to recurring usage. Many products require repeated use to build growing value.

Examples of products that require customers to build a habit to unlock compounding value include gym memberships, e-learning, and productivity apps.

A live conversation is the most effective at persuading customers to make that commitment.

To bring this to life, the internal Superhuman onboarding playbook included the following list of tactics that Onboarding Specialists could use:

• Have customers verbally commit to the Onboarding Specialist that they will use Superhuman every day for the next 30 days

• Log into all email addresses in Superhuman

• Create a browser bookmark for Superhuman

• Pin the Superhuman Chrome extension

• Set Superhuman as the default desktop mail app

• Add Superhuman to their iPhone "home row"

• Set Superhuman as the default phone mail app

• Delete Gmail and Outlook bookmarks

• Delete Gmail and Outlook from the browser URL autocomplete

• Delete Gmail, Mail, and Outlook phone apps

With the most customer-centric reason for human-led onboarding now articulated, let's explore the company-centric reasons.

2) Product feedback: workflows, features, issues

The first company-centric reason is that each onboarding lets you rapidly and deeply learn about your customers, and how they interact with your product.

In Superhuman's early days, onboardings ran up to 90 minutes:

• The first 30 minutes were devoted to discovery: Learning about customers, their pain points, and their email and broader productivity workflows.

• In the subsequent 60 minutes, we would do the onboarding, during which time we would obsessively capture every feature request, bug, and point of friction, including ones the customer hadn't even noticed.

It was common to come out of each conversation with 10 pages of notes, 5-10 feature requests, and 5-10 bugs to fix before the next onboarding.

This depth of insight is impossible to gain from analytics alone, even with a session recording tool like Amplitude or FullStory.

Of course, as we scaled, we tightened all aspects of the operation. We pared onboardings down from 90 minutes to 30 minutes. Learning about customers shifted to a pre-onboarding Typeform survey, plus two minutes at the start of each call. And the number of feature requests and bugs encountered drastically went down as the product matured.

But the value remained: human conversation yields a continuous lens into who your customers are and what they care about.

3) Drive loyalty: methodically manufacturing evangelists

Spending live time with customers generates a tremendous amount of goodwill.

Customers see that you are real people putting in real work to make their lives better.

You can build personal rapport by going above and beyond to help them. During Superhuman onboardings, we would give productivity advice whether or not it had to do with email.

And you can establish connections to follow up on later.

We'd listen to what was going on in their work and lives, take notes, and make a point to ask about it later. For example, one customer had recently become a parent, and we'd follow up a few weeks later to see how they were doing. Another said they would be running a marathon, and we set a reminder to ask them their time. We'd connect with our customers on social media, bringing them closer to Superhuman, and invite them to events at our office.

This has the effect of making each customer, on average, far more loyal than if they spoke to no one at all.

Spending time with customers methodically manufactures evangelists.

This, in turn, results in the following positive outcomes over the customer's lifetime:

• Customers are more likely to stay engaged and retain.

• Customers are more likely to recommend you to friends and family.

• Customers are more forthcoming about their needs, feeding your product development cycle.

• Customers are more likely to report bugs they encounter, increasing your product quality.

4) Sidestep issues: from moments of disappointment to moments of delight

One of the best-kept secrets of the early Superhuman onboarding was how little the product initially supported. The earliest version was:

• Gmail only (no Outlook)

• macOS only (no Windows)

• Chrome only (no native app)

• Desktop only (no mobile app)

In addition, there were missing features that many customers simply expected — unified inbox, CRM integrations, and Grammarly support — to name just a few.

And yet, we were still able to delight a majority of customers.

How?

A major benefit of speaking to customers in person is being able to directly share why functionality is missing, while also noting their request so you can delightfully inform them later on.

Instead of the disappointing experience of discovering gaps on their own, customers are usually happy to hear that the startup knows a feature isn't ready yet but has it on its roadmap.

Then, they're intrigued when they see you write down their request. And they're blown away when you follow up to let them know their feature is available.

A similar flow can be used for any bug encountered: Observe it, troubleshoot in the moment, take note, and close the loop later on.

An in-person conversation can turn moments of disappointment into moments of delight.

5) Resource arbitrage: solving the hardest problems first

Last but not least, if you are resource-constrained, it's possible that you are better off dedicating build cycles toward core functionality.

Consider that an engineer could spend a week building a critical missing feature, or, they could spend that same week building the self-serve onboarding.

Humans can gracefully fill in for the lack of self-serve onboarding. But humans cannot fill in for critical missing features.

If push comes to shove, it's better to onboard the customer manually and build core functionality, than to automate an onboarding into an incomplete and potentially disappointing product.

Mandatory vs. optional

There is a critical choice to make between making human-led onboarding mandatory or optional for customers.

Companies with a sales-led go-to-market already face this dilemma: Should all new customers be required to 'talk to sales', even if the product could be self-served?

At Superhuman, we initially mandated that all new customers attend a 30-minute human onboarding to access the product.

Our unique combination of ideal customer profile (time-sensitive founders, who would later become sales decision-makers in a land-and-expand motion), competitive landscape (a fallback in Gmail), and price point ($30 per month), pointed to a "no customer left behind" mentality.

We would not leave it to chance that customers might hit critical issues in their first few moments. We wanted to provide everything possible to succeed with Superhuman.

We experimented with making the onboarding an optional step after accessing the product. Even with our most pushy lifecycle marketing, the typical attendance rate was 15% — down from 100%.

Cost-benefit analysis

We have explored reasons to conduct human-led onboarding. But in many founders' minds, human-led onboarding sounds so expensive that they might never consider it.

A practical approach is to fully reason through a quantitative cost-benefit analysis. Let's unpack the trade-offs using Superhuman's numbers as a starting point. Take this structure and apply it to your own product.

Benefits

1) Revenue

Each fully ramped, full-time Onboarding Specialist was able to add up to $650k of topline ARR per year.

Assuming enough customer demand, each full-time Onboarding Specialist could lead 8 to 12 half-hour calls per day, or 4 to 6 hours per day. This provides room for asynchronous activity such as supporting new customers over email, as well as team rituals, and breaks.

The biggest inefficiency was unexpected customer no-shows. At Superhuman, this averaged 15% and spiked as high as 30% during public holiday weeks such as Thanksgiving.

Net of no-shows, each teammate aimed for 40 completed calls per week, over 45 weeks per year. The remaining weeks were expected to be vacation, public holidays, company offsites, and other out-of-office.

At Superhuman's price point of $30/month, this set of assumptions amounted to 40 customers * 45 weeks * $30/mo * 12 months = approximately $650k per Onboarding Specialist per year.

2) Improved activation and retention

As mentioned above, human-led onboarding saw nearly 2x the activation rates versus self-serve.

It is worth noting that this was hard to perfectly test. Due to selection bias, the segment of customers willing to show up to a 30-minute call was always a subset of the segment willing to self-serve into a product. So we paid extremely close attention to qualitative signals: Customer feedback both during and after the call.

Even if the actual uplift was half of these numbers, a 1.5x lift would still blow any normal self-serve onboarding experiment out of the water.

3) Virality

We saw the same 2x uplift in customer virality: Double the number of referrals per customer who was manually onboarded.

Again, to be cognizant of selection bias, we paid special attention to social signals.

Early "whale" referrers publicly shared their referral ability on social media, sometimes garnering over 300 referrals per customer.

And we continuously see founders bring Superhuman to their teams many years into the future.

4) Accelerated learning

We were able to learn about customer needs several orders of magnitude faster. Feedback loops that might have taken weeks could be observed within minutes and closed the same day.

5) Brand boost

Superhuman undoubtedly benefited from the brand benefits of manufacturing legions of evangelical early adopters. The downstream benefits are far-reaching and many.

Costs

1) Operating expense

A full-time Onboarding Specialist based in the Americas might cost anywhere from $60k to $130k fully loaded.

2) Manager expense

At scale, each squad of 10 Onboarding Specialists required one full-time manager, and approximately 10% of a full-time executive (at the time, me).

A sensible concept for evaluating these costs is sales efficiency: A team generating at least 3x the revenue as what it costs is generally considered good.

Of course, manager time carries a further cost that must be considered: The opportunity cost of that time not spent elsewhere.

3) Supply and demand mismatch

Human-led onboardings introduce a challenge that is very typical for fixed asset businesses, such as airlines or doctor's offices, but somewhat alien to most startups: A supply and demand mismatch.

The problem is that at any point in time, customer demand and onboarding supply never perfectly equilibrate.

Most founders rightly worry about customer demand exceeding onboarding supply, ultimately throttling growth.

But the concern also runs the other way: Too much supply creates operational inefficiency.

These challenges were most apparent for customers in different timezones and over weekends. They were also apparent for large demand spikes such as right after a launch, and large demand troughs such as over Thanksgiving and Christmas.

4) Customer experience

Some customers experienced frustration waiting for and attending their onboarding.

For a suitably short wait and suitably incredible onboarding, this was a non-issue. But, there will always be customers who are better off figuring out how to use the product themselves, especially in certain industries. For example, it's commonly the case that engineers prefer to try products themselves, which is why most developer tools avoid 'talk to sales', and have a self-serve option.

Of course, each startup will vary drastically in these benefits and costs. At any moment in time, it's important to understand yours, and identify your single decisive reason for picking one path over another.

With costs and benefits fully explored, let's zoom out to a broader framework of how long to practice human-led onboarding.

How long to practice human-led onboarding

As above, I believe every startup should start out doing human-led onboarding.

In the beginning, the benefits of generating revenue, delighting customers, accelerating learning, and building a brand most likely outweigh the costs. But there is a question of when, if ever, to transition to self-serve.

This can be boiled down to a 2x2. On one axis is the price point, which can be thought of as Average Revenue Per User (ARPU) or Average Revenue Per Account (ARPA). On the other axis is product complexity, which is a proxy for the benefits previously explored.

High price point, high complexity: potentially forever

Imagine complex B2B SaaS at a high price point. Examples might include a CRM or marketing automation suite. Or imagine a high-end direct-to-consumer service, like personalized healthcare.

These products might benefit from human-led onboarding for a very long time — potentially forever, via an Implementation or Solution Engineering team.

Implementation can be baked into your expected cost to acquire a customer (CAC), in turn supporting a high expected lifetime value (LTV).

High price point, low complexity: as much as desired for ongoing learning

Imagine an expensive product that is also extremely easy to understand. Examples might include car insurance or a talent marketplace.

These products generally don't need human-led onboarding after the startup has reached product-market fit.

These startups can decide how much they want to continue onboarding manually. The most durable reason is typically to support ongoing learning.

Low price point, high complexity: as long as economically viable

Consider a startup intentionally keeping prices low to win market share, even if the product is complex. Imagine a new CRM aggressively discounting, for example.

These startups should manually onboard customers for as long as economically viable. Their objective is to make customers successful while winning the market.

Over time, they can increase prices to justify continued human-led onboarding. Or they might do the challenging work of building a self-serve onboarding.

Low price point, low complexity: as long as needed to reach PMF

Finally, imagine a tool with an extremely simple value proposition, that is either very cheap or free to get started with.

Most consumer software falls into this quadrant. Consider gaming, social media, or utilities like photo editing.

These startups should manually onboard early customers only as long as necessary to reach PMF. Beyond that, the costs of onboarding likely will not be further justified beyond ad hoc product learning.

Superhuman: existing in superposition

In general, startups exist in a superposition across this 2x2. Different tiers or customer types might reasonably require different onboarding motions. The free tier might be entirely self-serve, while the enterprise tier might be entirely human-led.

To design your onboarding flow, consider each SKU and persona as though they were their own distinct product.

Superhuman began in the top-right quadrant: High price point, high complexity. For example, a CEO with fifteen years of email baggage to migrate, whose thousand-person company represented a future six-figure revenue opportunity. These were customers we wanted to spend one-on-one time with: The beginning of a sales motion. And we still seek to manually onboard these customers.

But at other times, we existed in the lower-left quadrant: Low price point, low complexity. For example, students on a discount tier, who are already hyper-efficient and able to hit Inbox Zero. We implemented a carve-out to let all students self-serve as soon as we found product-market fit.

As we scaled beyond the highest-expectation customers of Founders and CEOs towards go-to-market professionals and wider teams, we increasingly found ourselves in the lower-right quadrant: Smaller long-term revenue opportunities, but sufficient complexity to require human-led onboarding.

We scaled human-led onboarding until the costs, particularly supply and demand mismatch, overtook the benefits. This precipitated a steady transition towards a productized onboarding.

Part 2: Unconventional wisdom for productizing onboarding

Nearly all products — except for the most expensive and complex — eventually transition to some kind of self-serve onboarding.

But the transition can be tricky to get right.

Plenty has been written on the topic of what a good product onboarding entails. Some of my favorite materials include The Elements Of User Onboarding, What is a Good Activation Rate, and the Activation sections of Reforge's Retention Series.

I recommend readers absorb these guides to build a good mental model for how to productize onboarding.

But despite these materials and more, I noticed an under-served area in onboarding literature: What specific qualities should product onboardings contain? How should they feel?

Product and growth teams — particularly those who have spent limited time 1-1 with customers — often struggle with these questions. This results in watered-down, unmemorable experiences.

But coming from human-led onboardings, the answers felt obvious.

What you'd tell customers in person

When building your product onboarding, imagine you were right there with customers, just like a human-led onboarding.

• You would comprehensively guide customers to accomplish their setup moment.

• You might have a predefined workflow or reverse demo to get them to their aha moment.

• You would carefully watch for issues, and jump in if you notice any hiccups.

In other words, you'd be extremely attentive to the customer's every need.

Reforge likens this to removing one's marketing hat and thinking like a personal assistant.

This helps with function, but it doesn't fully explain form. How might software further replicate what a human might do?

We needed a second source of inspiration. So we asked: Where else are there countless examples of software guiding users through a brand new world?

We turned to a long-standing source of inspiration at Superhuman: video games.

For decades, video game creators have been perfecting the art of dropping players into new and complex worlds and then setting them up for success. They teach players to learn the controls, take action, and embark on their adventure.

I discovered that the most effective product onboardings embody three attributes that product and growth teams typically shy away from: opinionated, interruptive, and interactive.

1) Opinionated

What it means

There are many ways to use your product, but there is likely a best way. And you owe it to your customers to help them down that path.

This is the path that most rapidly accomplishes the setup moment, reveals the aha moment, and primes the habit loop.

Nan Yu, Head of Product at Linear, describes the following: When you adopt business software you're often adopting a practice along with the tool. You're adopting packaged wisdom and putting it into practice.

This perspective applies just as readily to consumer software as it does to business software. For example, Duolingo has a point of view on how you should learn a new language. Airbnb has deeply held beliefs about what it means to travel and host.

At the core of every great product is a problem that exists in the world, and its creator's opinionated solutions to that problem.

New users come to your product looking for those solutions.

Inspiration

We drew inspiration from one of the most iconic video game experiences in history: 1985's Super Mario Bros 1-1.

This level is a masterpiece in opinionated design. Without instruction, the player is encouraged to move right. Then, as the Goomba trundles towards you, you must either jump or perish and restart the level.

It opinionatedly tells the player: "This is a game about jumping. You will learn to jump perfectly."

As you proceed, you learn everything needed to succeed — movement, enemies, power-ups — all in pursuit of your goal.

How to do it

Consider the following simple exercise: What would you tell the customer if you only had 10 seconds? What about 1 minute? How about 10 minutes?

Progressively follow this thinking and capture the most salient messages.

Some of my go-to opinionated onboarding elements include:

• A mandatory introduction video that gives the product elevator pitch and tees up the user's next steps.

• An on-rails experience where users must take certain actions to build familiarity.

• Dropping users into a stripped-down version of the product containing only the flows relevant to brand-new users.

Case Study

Superhuman's first self-serve onboarding was hyper-opinionated. Similar to Super Mario Bros 1-1, we opted for an experience that taught users to move in a controlled environment.

The first step drew inspiration from requiring Mario to move right: We required users to hit "enter" to get started. Clicking buttons made them jiggle helplessly, encouraging users to abandon the mouse.

The demo then stepped through basic navigation, such as "j" and "k" to scroll through emails.

This first version worked. But we knew it could be much better.

Samuel Hulick of useronboard.com even told us at one point: Teaching navigation this way felt like we were forcing users to "eat their vegetables."

In our next iteration, we strongly prioritized our most opinionated objective: Inbox Zero. We removed navigation like "j" and "k", and prioritized the controls necessary to hit Inbox Zero: Marking done with "e" and setting reminders with "h."

This shift had astounding results: "e" and "h" usage increased by 50%, and self-serve activation rates improved by 25%, from 40% to 50%.

2) Interruptive

What it means

Great onboardings dare to be interruptive.

In the last 15 years, unobtrusive elements like checklists, tooltips, and in-line nudges have somehow become an industry standard.

Particularly on desktop, it is typical to click through a series of tiny tooltips when onboarding to a new product.

Over time these elements become clutter, piling up in the corners of your app. Cruft that slowly degrades the new user experience, dividing user attention between guidance, and the product itself.

If an experience is tucked away, it will be ignored. And if it is ignored, it may as well not exist.

Great product onboardings arrest a user's attention with something important to say.

Inspiration

Our source of inspiration was the unforgettable "Hey, listen!" call from Navi, Link's fairy helper in 1998's The Legend of Zelda: Ocarina of Time. Navi shouts when there is salient information necessary to make progress.

It might be a meme, but this guidance goes beyond simple help, forming an emotional bond that many players remember decades later.

How to do it

Again, consider what you might say in person. If your customer had missed a crucial concept, you'd earnestly jump in. The product should be no different.

At the outset, it's important to establish a golden rule: If you are going to disturb the user, it needs to be for something extremely valuable.

This filters out 90% of what you might interrupt the user with. It leaves you with only the ideas that you absolutely must convey.

These interruptions might be at the very beginning. Think about, for example, when you set up a new iPhone. The entire screen is dedicated to core setup: Face ID, Location Services, iCloud login, and more.

Or they might be deeper into the journey. Consider Headspace greeting you with an explanatory video from the founder after your third meditation, or Simply Piano enticing you after your first lesson with a preview of becoming a proficient musician.

It is critical to identify concepts that are useful to a high enough percentage of users. And these moments must be timed extremely thoughtfully — optimizing for moments of high user receptiveness, and avoiding moments when they might lose context or progress.

Case study

At Superhuman we experimented with checklists, tooltips, and timed nudges. These yielded zero impact.

The checklist was the most surprising. We hypothesized that users would be completionists, seeking to accomplish every step in a checklist.

So we gave them "side quests" to complete: Clean up old email, activate Superhuman AI, and more.

Only 50% of users completed each task, and only 30% of users completed all tasks. And we saw no change in activation rates. Worse, we now had an experience obscuring a core piece of functionality.

Reflecting on our sources of inspiration — Navi's interruptive call, the iPhone onboarding mentioned above, and more — we rebuilt the checklist as a full-screen experience.

We shifted from steps in a tucked-away checklist to full-screen panels before hitting the main product. And we made them impossible to skip, instead providing smart defaults so users could quickly zip through.

This drastically elevated total completion rates from 30% to over 98% — because users were required to complete the steps or bounce entirely. And opt-in rates for key features rocketed from 45% to nearly 80%.

We then took this insight to several other areas of the user experience. Perhaps the most extreme is "Get Me To Zero." This step interrupts customers right before they hit their "real" inbox, persuading them to pick a date, such as three days ago, and Mark Done everything before then.

For anybody who takes their email seriously, this might feel incredibly scary.

Despite this, an astonishing 57% of new users opt in to "Get Me To Zero," and it is responsible for archiving close to a billion emails and counting.

3) Interactive

What it is

A common objection to being opinionated and interruptive is that it removes agency from users.

The antidote is to make these experiences interactive.

The pedagogy mantra of "do > show > tell" can be applied: Customers retain far more when they have directly taken action.

Moments of interaction have the added benefit of building up user muscle memory.

And last but not least, they are fun!

Inspiration

Video games, by definition, put players in control. Our specific source of inspiration was an experience from Super Smash Bros: A Training Mode that lets new players try every button in a safe environment.

The magic of this experience is giving players 100% agency at the same time as 100% safety.

How to do it

When pushing a message during the onboarding, ask yourself: What action could a user take to internalize this message?

This might be a setting to toggle, CTA to click, or feature to play with. These actions help customers internalize and remember your message.

Then ask: How can you make it as safe as possible to take this action? One of my favorite mental models for new user onboarding is "the user is drunk." So it's important to prioritize simplicity and safety. You must make clear what the action is, when it is needed, and how it can be done. And you must make sure the user is protected against adverse outcomes, like losing work.

Finally, it's best practice to have customers use your product to learn the product. A brilliant example of this is Readwise's read-it-later app, Reader. Reader pre-loads an article from the founder explaining how to use the product. Throughout, it self-referentially explains how the user should interact with that article, including highlighting, notetaking, and ultimately archiving.

Case study

The centerpiece of Superhuman's onboarding is a synthetic inbox within which users can clear emails and hit Inbox Zero. It is fully interactive and entirely safe.

This was such a success that we wanted to replicate interactivity elsewhere in the onboarding.

In Superhuman we teach new users how to "Undo Send." The best time to learn this isn't during setup, or even just before sending. It is in the split second after sending, when context around sending is at an all-time high, and when users may safely try the feature.

When we notify users about "Undo Send," we don't just explain the feature. We make it so that tapping "Z" actually undoes the send. Meanwhile clicking "Got it" safely dismisses the alert, while allowing the user to "Undo" for another 10 seconds.

This arrests attention with the initial interruption, and then safely allows the user to play with the feature, unlocking a sense of satisfaction.

Why is this hard?

One of the most fascinating aspects of product onboarding is unpacking why teams shy away from building opinionated, interruptive, and interactive experiences.

Founders, product, and growth leaders should be vigilant against these traits, and work to erase them from company culture.

Lack of conviction

A common challenge is the startup lacking conviction in its ICP, and their jobs to be done (JTBD).

It is understandably difficult for the onboarding to be opinionated if the product itself is not.

This is especially common when a startup is bringing a technology or capability to market, prior to being specific about who, exactly, they are building for.

It is also common when a startup has multiple customer profiles with a wide variety of jobs to be done. This creates decision paralysis where the startup does not know how to balance the various personas.

In both cases, it is critical to pick the most important ICP and their most important JTBD, and then build the best possible flow for that. These decisions are rarely one-way doors.

Better to pick a path with conviction and pivot if required, than to never pick one at all.

In the case of products serving multiple personas, once a primary ICP has been chosen, it is much less work to branch the experience for different personas. This is typically done by customizing suggested content and workflow for new users.

For example, social apps like Pinterest and X and media apps like Netflix and Spotify commonly ask new users to pick their interests to seed their recommendations. Meanwhile, SaaS apps like Miro, Airtable, and Notion all have some version of asking users for their use-case such as education, work, or personal, then suggesting templates matching those details.

Fear of backlash

It is extremely common for teams working on onboarding to fear backlash, either from customers or from within. Shaping the onboarding for 100% of future customers is already a big deal. Being opinionated, interruptive, and interactive only heightens that tension.

Consider controversial moves such as making an explainer video mandatory, changing a setting to be on-by-default, or removing 'skip' options from the onboarding.

The antidote is rapid testing to reduce risk. Tests, by definition, are reversible, with a clear framework in place of when to proceed versus when to backtrack. It is natural for stakeholders to balk at a controversial onboarding step — until it yields a 10% improvement in activation.

Faulty ecosystem

Typical plug-and-play elements for onboarding such as checklists and tooltips approach onboarding the wrong way. They are simply widgets that are the easiest to drop in.

They do not represent what customers need most, and they certainly do not match the nuances of your product.

Instead of being swayed by the typical menu of options, think deeply about what customers need to understand your product. You might be able to create these solutions through low-code solutions, but consider that your solution might require a more bespoke approach.

Create delight

At the root of everything I've described is an obsession with the customer.

Customer obsession is one of those squishy values: Oft-touted, rarely operationalized. We built Superhuman around the concept so deeply that our core value, Create Delight, permeates every part of the company.

There is no better way to build the best experience in the world for your customers than speaking to hundreds, if not thousands of them. Our Onboarding Specialists did just that, onboarding thousands of customers. Special appreciation must go to Kyle-Prescott Ogunbase with the most individual customers to his name at over 3,900. Others including Laurel Johnescu and Matt Waters, conducted thousands of calls, then became managers, leading the growing team that delivered countless more delightful experiences. And our Growth Product Manager Ben Ophoven-Baldwin and Pod obsessively drove product improvements from the insights they gleaned. Their customer dedication astounds me to this day.

We made a simple but important decision to place the customer at the center of everything we did at Superhuman — and from there, it shaped everything we do.

“I did that.”

But what does this look like in practice?

There are two very different kinds of growth you can drive.

This write-up is for founders, growth, and marketing professionals driving results, and finance leaders forecasting and tracking outcomes.

We’ll typify the two modes of growth, and clarify misconceptions about each.

Subscribe now

1. The Burj Khalifa

These are the astronomical spikes that blow everything else out of the water.

Then dry up in a few weeks.

Think:

• PR / going viral

• Product launches

• Raising prices

Myth → Busted

These gains last → They generally don’t

It’s easy to be tricked into believing one-off bursts will yield a long-term trajectory change.

They rarely do.

Spikes usually linger, then simmer down.

If the underlying driver was promotional, these spikes might even pull forward growth, creating a post-burst lull. This is visible in the graph above.

There are two scenarios where one-off bursts do last:

• When your growth engine is built on network effects. Then, short-term bursts do contribute to sustained growth.

• When you distribute and capture extraordinary new value all at once. An example is launching a higher price tier, packed with new features. This lets you migrate existing customers (the Burj Khalifa) and see ongoing upgrades from future customers (the Rising Tide).

Of course, bursts that last should be top of your growth priority list.

Not repeatable → Entirely repeatable

If you’re seeing Burj Khalifas in your historic data and questioning whether you can recreate them: you can.

The height of these spikes are driven by reach and resonance. Both are tractable.

Resonance is art of understanding human emotion. What feels like magic? What stops someone in their tracks? What compels them to share? The ability to deliver resonance is a skill that, with sustained effort over time, can be developed.

Reach is the science of understanding spread. Resonance certainly aids reach in the form of word of mouth. But there’s more. Understanding social media algorithms. Optimizing share paths. Playing with incentives.

Develop resonance and reach and these spikes will become more and more frequent.

Growth drives → Marketing drives

Assuming the company is big enough to have both teams, these one-off bursts are typically driven by Marketing.

To execute spikes requires “campaign” mentality.

This is something Marketers tend to be excellent at. Meanwhile, most Growth people are miserable in this execution mode, as they prefer to ship changes that drive impact forever.

When planning, get the right people on the right things.

One team can own it → Takes several teams

On the flipside, I’ve seen the mistake of believing that one team, or worse, one individual exclusively can pull off huge spikes.

Marketing drives the effort, but meaningful spikes usually need a blend of Product, Growth, and Marketing, to build, optimize, and distribute value. Marketing distributes but, usually, isn’t building.

The best bursts are built on real value, which usually takes several teams. So, plan accordingly.

2. The Rising Tide

These are gains built slowly and surely over time.

Driven by high-velocity experimentation.

Think:

• Optimizing referrals

• Improving activation

• Driving network density

Myth → Busted

Takes weeks → Takes quarters

As a chronically impatient Growth person, I’ll be the first to admit I hate when anybody tells me something will take multiple quarters.

But this is the reality of creating a Rising Tide.

The fundamental bottleneck lies not in your ability to conceive of, design, or build experiments. Nor in your ability to analyze data. (Though these certainly can slow things down.)

The fundamental bottleneck is in a high enough volume of customers experiencing the change. Whether you’re observing changes pre vs. post, A/B testing, or testing qualitatively, waiting for users to flush through is just slow.

Even if you have a huge amount of traffic, there’s good reason to let tests run for 1-2 weeks. Amongst other things, this flushes repeat visitors and captures weekends.

And if you have low traffic, you might need to pipeline tests to actually learn anything.

It’s reasonable to expect to work on a metric for 2 quarters, perhaps longer.

If that sounds like it’ll leave you twiddling your thumbs — add other areas to work on in parallel.

High specificity → High stochasticity

You will try 30 things, and only 10 of them will work. But at the outset, you have no idea what those 10 will be.

Furthermore, you have no idea what new ideas you will prioritize, based on those 10.

Where this myth causes the most friction is in creating plans — and evaluating performance versus plan.

Take quarterly planning. It’s common for teams to list quarterly initiatives. And as companies mature, it’s common to put impact estimates against each effort. These estimates collectively should surpass an overall goal.

This planning method more or less works if you deliver Burj Khalifas.

It’s mostly a shitshow if you deliver a Rising Tide.

You might estimate +$1M of incremental ARR each year from an idea. But there’s a 50%+ chance it doesn’t happen. (More precisely, there’s a range of potential dollar outcomes which, 50%+ of which won’t be exciting enough to keep the change around.)

What then, of initiative-level plans?

This isn’t to say you shouldn’t plan efforts or anticipate impact. Rather, the roadmap and its impact estimates should reflect the work’s stochasticity. They should be appropriately coarse vs. spuriously precise.

Meanwhile the best planning outcome is a team hellbent on executing and learning as fast as possible.

Just optimizing → Stacking gains

It’s common to dismiss this type of growth work as low-value on the basis of it simply optimizing the thing you have today.

“We don’t want 10% improvements, we want 10x improvements!”

Compound interest is hard to reason with, so a little illustrative math can ground us.

Many are familiar with “1% better every day”:

If you get one percent better each day for one year, you’ll end up thirty-seven times better by the time you’re done.

The business equivalent: a 10% improvement every week for 2 quarters does in fact deliver a 10x improvement (1.1x^26=11.9x).

So the question becomes, how do you deliver a 10% improvement every week?

Simple rule of thumb: if roughly 1 in 3 initiatives will succeed, launch 3 experiments each week.

Again, if you don’t have enough customer flow to do this for one metric, tackle multiple areas in parallel.

Takes a village → Takes one Growth pod

The beauty of this category of work is that it can be executed by a very small group of individuals.

The skills needed are relatively diverse: product, design, engineering, data. But find yourself a PM Who Designs, and an Engineer Who Analyzes, and you’ll be surprised what a tiny team of 2 can accomplish.

Of course, a pod of 2 is overly idealistic. A more typical growth pod structure is a Product Manager, Designer, Data Scientist, and 3-7 Engineers. But still, we’re talking a single Two Pizza Team that, with sufficient agency, can change business trajectory forever.

Putting it together

The best plans combine frequent Burj Khalifas and rapid Rising Tides.

Both simply planning to make them happen at the same time. And better, prioritizing ideas that drive both growth modes. Examples:

• PR hype built on content that yields evergreen traffic

• Price changes that create ongoing user upsell

• Features that attract new audiences and lower existing churn

We’re entering a new era where starting a company is easy, but succeeding is exponentially harder.

The winners won’t be the ones who can build the best product — that’s a commodity now.

The winners will be the founders who deeply understand customer psychology, growth engine optimization, and scalable distribution strategies.

— Sean Ellis, How AI-Generated Code Will Reshape Everything

Whether new to PMF or storming revenue, startups want growth.

And they want it reliably, predictably, sustainably.

Nearly every single startup I talk to is actively or soon hiring for growth.

Last week Elena Verna published a timely writeup on good and bad growth job descriptions.

But what about the next step?

What should you watch for when interviewing?

Here’s my list of interview red flags. And, my playbook for the perfect growth interview plan.

Read this if you’re taking a growth role or hiring for growth.

Subscribe now

1. Moves Slow 🚶‍➡️

2. Resists Ideas 🙅

3. Pointless Take-Home 😒

4. Idle Chit-Chat 🥱

5. Sleight of Hand 🫳

In detail…

1. Moves Slow 🚶‍➡️

The most visible red flag is if the hiring process moves slow.

Success in growth demands velocity.

And while moving slow is concerning for any startup role, it’s a red flag for growth.

What does moving slow look like?

• >1 business day to respond to comms

• >3 business days to hear back from an interview round

• >10 business days between interview rounds

Here’s what those numbers should look like:

• <3 business hours to respond to comms

• <1 business day to hear back from an interview round

• <4 business days between interview rounds

(Obviously, use your judgment to make exceptions for one-offs, like a key interviewer is unexpectedly out of office.)

But what if the startup that moves slow wants to bring you in to intentionally boost velocity?

Your mileage may vary.

A fast-moving growth hire might shift a small startup’s culture.

More likely, there’s a lack of urgency baked into the team.

This is important to tease out, which brings us to our next category.

2. Resists Ideas 🙅

As you proceed through conversation, how open to new ideas are the team?

Pay particular attention to the founders, as at most companies, the buck stops with them, especially on high-risk high-reward efforts.

First, assess the job description (JD).

The thesis of Elena’s write-up is that too many growth JDs are kitchen sinks of requirements. And it’s common to see growth JDs rammed with everything from PMM to Sales with everything in between.

If you find this, test if the team is resistant to new ideas. Ideally, they welcome input, including the possibility that the role’s expectations need updating.

The second place to gauge this is during interviews.

Top growth candidates can confidently generate hundreds of ideas until they find the one channel (or product) that makes all others irrelevant.

To nail landing a role, I recommend presenting those ideas — whether asked for or not. Maybe it’s explicitly part of an interview, but if not, I recommend writing up your thoughts and sharing a 10-20 minute Loom voiceover.

Then evaluate: how does the founder respond?

Do their eyes light up? Do they lean in? Do they ask more questions, begin shaping ideas with you, figuring out next steps?

Or do they seem uncomfortable? Do you hear why ideas won’t work? Do you discover a part of the company you’re not allowed to touch? (Hint: it’s usually product).

In growth, the smaller the company, the more critical it is for everyone to exhibit high affinity for new ideas. And crucially, growth roles must be able to weigh in on any part of the company.

3. Pointless Take-Home 😒

Take-homes signal a company that doesn’t get growth.

Why?

First, take-homes introduce massive latency to the hiring process. Even a 24 hour turnaround can add a few days to the hiring process when you add in time to review, weekends, vacations. And we covered above why Moves Slow is a red flag.

If a startup hasn’t figured this out, they likely don’t understand that moving fast is crucial to growth, which is concerning.

Second, take-homes test depth of thinking — deep, reflective, systems 2 consideration. And they validate synthesis of those ideas.

While these skills matter in growth, they’re far less important than:

• Quick instinct and velocity

• On-the-spot problem solving

• Identifying tests you could launch today, not crafting beautiful roadmaps

Is this to say that the entire interview process should be live? Not at all.

One of my favorite interviews is to sit a candidate down with a prompt, give them 40-60 minutes to work, and then discuss their output for 15 minutes at the end.

Killer candidates who move fast will demolish these exercises, and will be ready to talk in as much depth as necessary about their findings.

My go-to picks for this semi-async interview:

• Test data-literacy and analysis: Present growth data in a spreadsheet. The data contains some jank. The candidate is prompted to calculate a set of KPIs, such as conversion rates. They then figure out strategic next steps that they’re ready to talk through.

• Tests writing and taste: Present customer-facing copy, such as activation emails, in a document. The emails have some issues. The candidate is prompted to re-write them according to a brand guideline to optimize for an outcome such as activation. They then figure out what else they’d do to drive that KPI.

• Test hypothesis formation and experiment design: Present some part of the company’s funnel or product, and a few internal hypotheses. The candidate must create a viable test that confidently prosecutes the hypothesis. They then explain what they’d do after the test, given different outcomes.

These are what Andrew Chen calls “infinite depth questions”, where the candidate could elaborate forever, yet you get signal on outlier candidates in <30 minutes.

4. Idle Chit-Chat 🥱

Unstructured, vague, “tell me about yourself”.

Maybe a long walk-and-talk. Maybe a carousel of unprepared teammates.

These sessions are tolerable in moderation, either early or late in a process. They help with rapport, trust, selling.

But it’s concerning if this is the entire interview process.

Reason: idle chit-chat indicates a lack of hypothesis clarity, or, lack of ability to test your skills.

A company unclear on hypotheses or unable to test will be much further behind on growth.

If you’re the company: I recommend at most 3 “chit-chat” sessions:

1. Early-stage "get to know you” with the hiring manager

2. Mid-stage “collaboration and communication” interview with a peer

3. Late-stage “sell” conversation with the hiring manager or leader

5. Sleight of Hand 🫳

This last red flag might be hard to spot, but is critical.

Does leadership talk openly and honestly about their product, growth, and customer feedback? Do they highlight opportunities and challenges?

If you detect this, it indicates there’s willing or unwitting obfuscation of truth at the company.

Objectivity in the form of hard data, and clarity on customer sentiment — is the lifeblood of succeeding in growth roles.

How do you gauge this?

First, validate how well leadership understands their business, and how honest they are with that information.

Ask direct questions and expect clear answers. Questions might include, but aren’t limited to:

• Financing: What runway does the company have? What was the last round of funding? How much, from whom?

• Growth: What is the company’s ARR growth rate? What is the current goal, if any?

• Retention: What is our NRR (B2B)? What is our 1mo and 12mo usage and dollar retention (B2C)?

• Economics: What is our CAC? How is that calculated?

• Sentiment: What is our NPS or “Very Disappointed” score? What’s the best and worst customer feedback we’ve recently received?

They might not have all the answers — and that might be why they’re hiring you. Or they might want you to sign an NDA.

But watch for cageyness and inconsistency.

Second, validate answers across teammates.

Inconsistency across the team is more of a yellow than red flag.

It’s a red flag if leadership has shaped information to their team in a way that makes things sound much better than they actually are.

It’s more of a yellow flag if teammates simply don’t know the answers. While the optimal environment for growth is an extreme degree of information symmetry, this is a more tractable problem than deliberate obfuscation.

If you’re a hiring manager, be ready to answer, and prepare your team.

And in the meantime, if you know someone interviewing for growth, send this their way.

So what constitutes a good growth interview plan? Guide coming soon.

What area of hiring should I unpack next?

Reply or comment ✍️

Taste.

So hot right now.

Though, hardly new.

Taste has been dissected through the ages, and certainly as long as Silicon Valley has existed.

Steve Jobs identified taste as heading in the right direction — and what Microsoft lacks — in 1995. Paul Graham articulated taste as principles of good design in 2002. Julie Zhuo detailed concrete techniques for developing taste in 2013, as did Emil Kowalski in this elegant memo in 2025.

The topic is ever-present in technology’s collective consciousness. But it’s experiencing a cultural renaissance.

Why?

Spurred by the tidal wave of AI content flooding our everyday lives.

Anu Atluru astutely observed in September 2024:

In a world of scarcity, we treasure tools.

In a world of abundance, we treasure taste.

The barriers to entry are low, competition is fierce, and so much of the focus has shifted — from tech to distribution, and now, to something else too: taste.

— Anu Atluru, Taste Is Eating Silicon Valley

Experience the AI tidal wave and you will quickly recognize the tasteless hallmarks of AI slop.

In writing: meandering, delving. In images: high-gloss yet lackluster. In audio: metallic and interruptive, barely bridging uncanny valley despite claims.

Quadrillions of calculations hurtle towards AGI’s final frontier: Will Smith devouring spaghetti. But for now, in our million-year-old brains, a different quadrillion calculations sense deception.

Synthetic Smith slurps; we squirm.

Resonance

The pendulum is swinging back towards craft, quality, authenticity — both culturally and economically.

Code is cheap.

Money now chases utility wrapped in taste, function sculpted with beautiful form, and technology framed in artistry.

— Anu Atluru, Taste Is Eating Silicon Valley

Taste has always been a primary concern for designers. Accordingly, most discussion of taste has focused on capital-D Design.

But taste is now increasingly prized across the board. This includes Growth, where not long ago, mere functional competence sufficed.

Growth grew up in a world where it was fine, even encouraged, to disregard everything except for the number.

Remember “throw spaghetti at the wall and see what sticks”?

I assure you that spaghetti did not taste good. Synthetic Will Smith would agree.

Production is cheaper than ever; frameworks are commoditized. You and you competitor with comparable product use the same gen-AI adtech, studied the same onboarding playbook, attended the same virality course.

In this world, standing out demands a specific manifestation of taste: resonance.

Reach is how many people see you. Resonance is how much they care.

Awareness is a proxy for what we really want. We want affinity.

We don’t just want others to be aware of us. We want them to love us.

— Jay Acunzo, Two Revealing Definitions to Inform Better Work

So much of Growth focuses purely on reach.

Maximize impressions. Grow followers. Rank in search results.

But resonance is how we form lasting impressions.

Resonance is the message that sits with you. The experience you must share. The inspiration to take action.

Resonance is Duo’s death then resurrection. It’s Linear feeling more like a social movement than B2B SaaS. The Superhuman onboarding you recall 8 years later, the psychedelic journey through Superpower’s homepage, the awe of Stripe’s Black Friday cash register.

It is possible to deliver reach devoid of resonance. Distribution channels open up to deep enough pockets. A sufficiently well-resourced party can push any message to the world, no matter how insipid. They merely need to imagine it, genmoji it.

A spectacle is also prone to go viral, but ultimately lack emotional connection. Cheap thrills flood our news feeds.

Conversely, it is possible to deliver resonance that lacks reach: a hidden gem. Certainly, resonance can amplify reach through word of mouth. But in today’s algorithmically policed world, this is far from guaranteed.

Both resonance and reach is, of course, the ultimate prize.

And a sense of taste is the foundation for delivering resonance.

Taste is how we anticipate, ahead of time, that we are producing experiences beyond utility — striking ads, viral stunts, sites that sing, opinionated onboardings, copy with character.

Practical Foundations

So how might one develop taste in pursuit of resonance?

Let’s first set realistic expectations. Jae Lubberink aptly notes:

The mimetic urge to simply ‘acquire Taste’ reveals why it does not come naturally to many. It’s oxymoronic.

Taste, like wine, demands terroir: the grit, the grind, and the unglamorous labour of digging deeper

At its core, taste is a love letter to effort.

— Jae Lubberink, Capital-T Taste doesn't exist

Nonetheless, it’s useful to identify which effort.

I hope this provides a practical foundation for founders, Growth professionals, and beyond.

1. Learn The Rules

Learn the rules like a pro, so you can break them like an artist.

— anon, commonly attributed to Pablo Picasso

If you’re already in Growth, this should feel natural.

The entire field builds on the idea that companies can be grown by applying well-defined frameworks. These frameworks are composed of rules across mathematics, technology, and psychology.

So in pursuit of taste, learn the rules.

Spend time to fundamentally understand how things work.

Start within Growth. Subscribe to Substacks, read Reforge, wander Wikipedia. Don’t just A/B test: understand the underlying statistics. Don’t just send emails: dig into deliverability, domain reputation, responsive design.

Next, pursue interdisciplinary influence. Growth lives between Product and Marketing, so study these disciplines. What questions should your homepage answer? What frameworks create deeply cherished products?

Go further still. At Superhuman we encourage depth and mastery. It is common to recommend niche textbooks from The Art Of Game Design to The Visual Display Of Quantitative Information.

2. Gather Data

Design-focused advice recommends following the greats, and surrounding yourself with their work.

The same advice holds in Growth.

Voraciously explore products.

Seek growth flows. Explore branching pathways. Observe macro decisions and minuscule details.

Let’s acknowledge that growth flows can be hard to conjure at will. Unlike hammering a feature, it is laborious to redo an onboarding just to see if it’s changed. Other flows are fleeting and cannot be re-invoked; a transient notification, for example.

For precisely this reason, resources exist to curate growth artifacts. Subscribe to these to efficiently gather data.

User Onboard details product onboardings. Good Better Best snapshots pricing pages. Growth.Design chronicles end-to-end growth flows. Siteinspire captures homepages. Really Good Emails has … really good emails.

If you’re starting out, gather data as an undirected search. This is acquisition for acquisition’s sake, as you don’t know what you don’t know. You will inevitably focus over time.

3. Practice Discernment

While gathering data, evaluate the goodness of each experience. Use reach and resonance to do this.

Reach is easy to assess.

How many views, likes, and comments does a piece of content have? How many shares? Spot and absorb numerical proof points.

Resonance is less obvious, but still tractable.

How did that campaign make you feel? Why? Could you remember the message weeks later? How about months? What do others say, particularly high-taste testers?

I recommend active evaluation.

Next time a product asks you how likely you are to recommend it to a friend, write down why your answer is what it is.

When you next onboard to a new product, record in Loom and verbalize your thoughts.

4. Effortful Creation

I hear and I forget. I see and I remember. I do and I understand.

— anon, commonly attributed to Confucius

You must create as part of developing taste. An hour of practice is worth a day of study.

I put emphasis on effortful creation. Muscle is built through exerted effort.

You might use AI to speed up your process, but it cannot tackle the core of creation. You must.

In Growth, creation is commonly parceled out to other individuals. Engineers engineer, designers design. And rightly so, for efficiency.

But to build taste you must be part of production. You must know when to take the pen and let the ink flow.

In Growth Product, scaffold the experience and write the copy. In Growth Marketing, articulate the emotions and draft the message.

5. Seek Feedback

Feedback refines taste.

Your creations are already measured on reach. You pursue goals and track metrics. You know when you hit the number.

So how do you get feedback on resonance?

Proactively ask.

Ask hard questions, particularly of those you admire, or seek to reach. How did it make them feel? What worked, what didn’t?

What we make people feel is just as important as what we make.

— BMW, commonly attributed to Rahul Vohra

Don’t only listen to what people say to your face. Hear what strangers tell each other.

Some of my biggest realizations of when we’d missed the mark at Superhuman were from passive but public critique.

You’ll hear things that don’t feel good; things you outright disagree with.

Maybe you deliberately decided to not polish before shipping. Or maybe you’re criticized on an effort that is a numerical success.

Amidst the noise there will be salience, and that salience will refine your taste.

6. Repeat

Iterative repetition unlocks increasingly sophisticated production.

By the time you have executed your 100th viral campaign, landing page, or product launch, you’ll know a thing or two about doing them very well.

Little by little, you might shift from taster to taste maker. You might receive requests to review. You might notice your work curated, referenced, or outright copied by others.

This is good!

Whether or not this happens, remember that developing taste and pursuing resonance is lifelong. “A love letter to effort”.

There is no ceiling: keep learning; keep creating.