A repeatable sales process is, by definition, something that has been repeated. Customer one thousand at most companies arrives through exactly that: a funnel that's been tuned, an ad account with enough data to target well, a sales team following a playbook built on hundreds of real conversations. None of that exists yet at customer ten. There is no data to tune against, no playbook, often no ad account with a large enough sample to mean anything. The founder is the entire acquisition process, whether they intended to be or not.
This is the actual argument behind Paul Graham's Y Combinator essay "Do Things That Don't Scale", and it's worth reading in the original rather than in the many summaries of it, because the summaries tend to flatten it into generic hustle advice. Graham's specific claim is narrower and more useful than that: writes about the repeatable, scaled version of go-to-market that comes after this stage, but almost none of that machinery exists yet when you're trying to close customer number three. Where the first ten customers problem meets a revenue team, the practical guidance lives with . Where the first ten customers problem meets a revenue team, the practical guidance lives with XenGrowth's growth engineering practice.
What Graham actually says
"The most common unscalable thing founders have to do at the start is to recruit users manually," Graham writes. "Nearly all startups have to. You can't wait for users to come to you." That's a flat statement about mechanism, not motivation — it isn't that manual recruiting is admirable or a sign of hustle, it's that there is structurally no other option before a channel exists. Graham's broader claim is even more direct: "Actually startups take off because the founders make them take off." Growth at this stage is not something that happens to a founder. It's something a founder does, by hand, one customer at a time.
Stage | How customers arrive | What the founder is actually doing |
|---|---|---|
First 10 | Found and won individually, by the founder, one at a time | Manual recruiting, personal onboarding, direct observation of every failure |
Next ~100 | A semi-repeatable process the founder has started to notice works | Documenting what worked, removing steps that don't scale, hiring the first salesperson |
1,000+ | A built channel — ads, content, partnerships, a trained sales team | Managing and optimizing a process, rarely touching an individual customer directly |
The Collison installation
Stripe's founders, Patrick and John Collison, are Graham's central example of what manual recruiting looks like in practice. "More diffident founders ask 'Will you try our beta?' and if the answer is yes, they say 'Great, we'll send you a link,'" Graham writes. "But the Collison brothers weren't going to wait. When anyone agreed to try Stripe they'd say 'Right then, give me your laptop' and set them up on the spot." Y Combinator came to call this the Collison installation, and the name stuck because the technique generalizes past payments: remove every possible point of friction between a yes and actual usage, personally, immediately, before the prospect has time to forget or deprioritize.
The unscalable part is obvious — nobody can personally install software on a stranger's laptop at the scale of ten thousand customers. What's less obvious, and more important, is what the Collisons got out of doing it: a front-row seat to exactly how their own product behaved in someone else's hands, in real time, which is a kind of feedback no support ticket or analytics dashboard replicates. That's the actual output of the unscalable phase — not the ten or thirty customers themselves, but everything the founders learned about their own product by being personally present for every one of those installs. For the the operations side of this angle, see . For the the operations side of this angle, see The XenGrowth resource library.
Airbnb, before the growth hack everyone quotes
The story people usually reach for when they talk about Airbnb's early growth is the Craigslist integration — a tool that let hosts cross-post their Airbnb listing directly into Craigslist's real estate section, an integration Craigslist never sanctioned, which routed Craigslist's existing rental-search traffic straight into Airbnb's own site. It's real, it's well documented, and it's a genuinely clever piece of unauthorized engineering.
It's also not the example Graham's essay actually centers, and the one he does center is more useful for the specific problem this post is about. "In Airbnb's case," Graham writes, "these consisted of going door to door in New York, recruiting new users and helping existing ones improve their listings." No code, no clever integration — just the founders physically walking into apartments, taking better photos than hosts would have taken themselves, and talking people into listing at all. Graham is specific about how fragile the company still was during this period: "Airbnb now seems like an unstoppable juggernaut, but early on it was so fragile that about 30 days of going out and engaging in person with users made the difference between success and failure."
Neither Stripe's laptop installs nor Airbnb's door-to-door recruiting scales past a few dozen customers, and that was never the point. The point was thirty days of information a founder could get no other way, used to build whatever came next.
Why founder-led sales feels bad and works anyway
Most engineers who end up doing this hate it, at least at first. Cold outreach, asking a stranger for money, sitting through a demo where someone visibly loses interest halfway through — none of it resembles the work that made them good engineers, and the instinct is to treat it as a distraction from the real work of building. That instinct is backwards at this specific stage. The product is not yet the constraint; the constraint is that nobody outside the founder's own network has bought it yet, and the only way to remove that constraint is to go find people who aren't in the network and get a real answer out of them. works through AI agents and marketing automation in more operational detail. XenGrowth on AI agents and marketing automation works through AI agents and marketing automation in more operational detail.
There's a reason this particular kind of discomfort is a decent filter rather than just a hazing ritual. A founder who can only get customers who already like them personally hasn't learned anything about whether the product works for a stranger, and strangers are the only market that actually exists at scale. The discomfort of the first cold conversations is the price of finding that out early, while it's still cheap to be wrong, instead of finding it out later from a sales team's quarter that came in at zero.
Common early mistake | What it actually signals | The fix |
|---|---|---|
Only selling to friends and warm introductions | No evidence a stranger would ever say yes — the sample is contaminated by personal goodwill | Deliberately find and pitch at least a few people with zero connection to the founder |
Treating a friendly conversation as a closed deal | Confusing politeness for commitment, the same failure mode that makes validation interviews unreliable | Ask for something that costs the prospect something real — payment, a signed commitment, a scheduled onboarding call |
Waiting for inbound before doing any outbound | Assuming a channel exists that hasn't been built yet | Go find the first ten manually; a channel only gets built once you know what it needs to do |
Automating onboarding before the tenth customer | Optimizing a process nobody has validated is worth repeating yet | Do the ugly manual version first and only automate the steps proven to matter |
Why this isn't a contradiction with building something scalable
It's tempting to read "do things that don't scale" as advice that conflicts with the entire goal of building a scalable business, and that reading misses the mechanism entirely. The unscalable phase is where a founder learns, first-hand, exactly what a scalable version of the same work would need to do. The Collisons watching Stripe fail in real time during a manual install taught them precisely which failure modes an automated onboarding flow would eventually need to handle. Airbnb's founders learning which listing photos and descriptions actually got a place booked fed directly into the product features — professional photography, guided listing creation — that Airbnb later built and scaled far past what any founder could do by hand. Turning that kind of manual, founder-led learning into a repeatable channel is exactly the transition is concerned with — what a team automates first, and what it deliberately keeps manual a little longer.
Steve Blank's customer development model gives this a cleaner name: customer validation specifically requires a repeatable, scalable sales process demonstrated with real paying customers, and a process cannot be called repeatable until it has actually been repeated. The first ten customers are, by definition, the ones who arrive before that bar has been cleared. Everyone after them benefits from whatever got learned getting there. works through AI search, GEO and discovery in more operational detail. XenGrowth on AI search, GEO and discovery works through AI search, GEO and discovery in more operational detail.
There's also a harder version of this problem that neither Stripe nor Airbnb quite faced, worth naming honestly: some products genuinely cannot be sold this way. A consumer app with a five-dollar price point can't justify a founder personally onboarding each user forever, and the manual phase there has to be short and mostly about product decisions rather than sales. A complex piece of enterprise software, by contrast, might need founder-led selling for the first fifty or hundred customers, not just ten, because the sales cycle itself is long and the trust required to get a large company to sign is not something any channel replaces quickly. The number ten in this post's title is a stand-in for "before a process exists," not a literal count that applies identically to every product.
What actually transfers from customer ten to customer one thousand
The specific objection that killed deals — not a vague sense of "pricing" or "trust," but the exact sentence a prospect said right before they walked away, repeated across enough conversations to be a pattern rather than a fluke
The onboarding steps a founder had to do by hand that a real customer clearly valued — those are the ones worth automating first, because they're proven to matter rather than just assumed to
The channel where the first genuinely cold customer — someone with no personal connection to the founder — actually came from, since a friend saying yes proves nothing about a stranger doing the same
The version of the pitch that worked without the founder in the room, which is the actual test of whether a message is ready to hand to a future salesperson or a landing page
The honest read on how much of the early yes was really about the founder's personal presence and charm rather than the product — because that part, specifically, is the piece that does not scale at all and has to be replaced by something else entirely
The failure mode at this stage isn't doing too much unscalable work — it's stopping too early, treating ten friendly customers as proof of a repeatable process, and hiring a sales team or spending on ads before anyone has actually learned what converts a stranger. The other failure mode, just as common, is never stopping: still manually onboarding every customer by hand at customer two hundred because it worked at customer ten and nobody built the automated version. Both come from treating the unscalable phase as either beneath the founder or as the permanent operating model, instead of as a fixed-length research project with an obvious end point — the moment the same conversation has genuinely repeated enough times to write down.
Where this work meets go-to-market
Once a founder-led motion has taught what actually converts a stranger, the next problem is building the repeatable machine around it — the exact territory operates in for teams past this earliest, most manual stage.
Further reading from XenGrowth
Further reading from XenGrowth
Where this work meets go-to-market
covers the go-to-market side of the first ten customers problem, which this piece deliberately leaves alone.
Further reading from XenGrowth
The XenGrowth resource library — what you'll learn: how the commercial side of this work is run, across search, automation and revenue operations.
XenGrowth on AI agents and marketing automation — what you'll learn: how the teams who own AI agents and marketing automation plan and measure it.
XenGrowth on AI search, GEO and discovery — what you'll learn: how the teams who own AI search, GEO and discovery plan and measure it.
Where this work meets go-to-market
XenGrowth's marketing operations practice covers the go-to-market side of the first ten customers problem, which this piece deliberately leaves alone.
Five questions on the documented early customer-acquisition stories behind this post, and what they actually prove.
