Short answer: the first ten customers of a B2B AI startup aren't bought with marketing — they're recruited with a single founder doing a hand-built integration per customer, in exchange for design partnership rights and a quarterly check-in for the next 18 months. Distribution is the product for the first year. Build for that.
Most B2B GTM playbooks were written for SaaS where the product was deterministic and the buyer knew what they wanted. AI startups have neither. The product hallucinates this week. The buyer can't tell you what a "good" output looks like. Your competitor is the customer's existing engineering team trying the same idea on a Friday.
So the playbook is different. Here's what's worked.
Move 1: Pick a workflow you can run in a meeting
In your first ten sales calls, you should be able to run the product live on the customer's actual data, in the meeting. If you can't, the workflow you've picked is too abstract.
This filters which problem you go after. "We help analysts" — too vague. "We turn 40-page legal briefs into a one-page summary your associate would sign their name on" — runnable, demonstrable, decisive.
The customers who say yes after a live demo close in days. The customers who say "send a deck" close in months and often not at all.
Move 2: Trade integration work for design-partner rights
The unfair offer for early customers:
"We'll build the integration into your stack, free, in two weeks. In exchange: you give us weekly feedback for 90 days, the right to feature you in case studies, and a quarterly check-in for 18 months."
This is not a sales motion. It's a research deal dressed as a contract. Both sides know it.
Customers say yes because the risk is zero (you build it free, you maintain it) and the upside is a real solution. You say yes because you're getting more than money — you're getting the user behavior that lets you build a generalizable product.
After ten of these, you have evidence. Not "we have ten logos." Evidence — workflows you've watched, edge cases you've fixed, language you've stolen for your marketing.
Move 3: Charge real money on day one, but make it psychologically cheap
The instinct is to give it away. Wrong. Free customers don't behave like real customers — they don't push back, don't churn, don't reveal what they actually value.
Price small enough that the buyer can authorize it without procurement. $500 / month is a great early price. That's a corporate credit-card-and-forget number. The amount almost doesn't matter — what matters is the customer made a real decision.
Then, separately, give them the design-partner deal. That's not the same conversation as the price.
Move 4: One founder, ten customers, no salespeople yet
For the first ten, the founder runs everything. No SDRs. No AE. No customer success. One person walking the entire arc — first cold note → demo → integration → onboarding → 30-day check-in.
This is unsustainable on purpose. It's the period where you're learning the shape of the sale. What objections come up. What the actual close trigger is. What the customer says when they tell their CFO why they bought.
You can't outsource that learning. The moment you hire a sales team is the moment those learnings stop accumulating in the founder's head and start getting filtered through someone else's instincts. Wait until you have a playbook.
What signals "ready for #11+"
You're ready to scale past hand-built when:
- You can predict, before the demo, whether the customer will sign — within 70% accuracy.
- The integration takes one day instead of two weeks.
- The customer's success metric becomes legible — you can name what they'll see in 30 days.
- You can name three companies you definitely shouldn't sell to. (If you can't, you haven't found the wedge.)
Until all four are true, hold the line. Hand-build the eleventh customer. The hand-built ones compound. The pipeline ones don't.
The thing nobody says out loud
The first ten customers of a B2B AI startup don't pay for the product. They pay for the founder's attention. They get a private engineer/PM/designer for six months for $500/month. That's the actual deal. Both sides know.
The trick is to extract enough generalized learning from those ten that you can build the product the other 990 customers actually buy. If you don't — if customer #1's hand-built integration just becomes a custom branch you maintain forever — you didn't build a startup. You built consulting with a logo.
The line between "design partner" and "consulting contract" is whether you're allowed to build the same thing for someone else. Negotiate hard on that one. Everything else is decoration.