Finding pmf for an AI startup – a playbook without doing YC

We’re running 2-week validation sprints to find product-market fit. this is what YC without YC actually looks like. This is week 3.

Everyone says “you’ll just know when you have PMF” which is the least helpful advice in startups. First Round Capital has this thing called Levels of PMF that breaks it down into actual stages. We’re somewhere around level one.

Here’s the framework we’re testing:

You pick your exact target. not “logistics companies” but “VP of Sales at 1k+ employee logistics companies using Microsoft Suite for RFPs.” we try to add 3+ specific firmographics. the more narrow, the better apparently. The theory is if you’re specific enough, someone will eventually care.

We use Claude Projects to research the industry and build a brief. it’s about 60% accurate which turns out is enough to not sound completely clueless on calls. condense it down to a 2-3 sentences pitch for your target.

We connect with 25 people daily via HeyReach. Reach out 1 seniority above and 1 below to maximize luck (or desperation, hard to tell), no connection message. When they accept we send something short and direct. Follow up every 2 days until they block us or respond.

We log everything in Linear using MCP because Hubspot UI is too slow and makes us want to throw laptops.

When someone actually agrees to a call we panic-build a prototype in Cursor. Doesn’t need to work perfectly but it must show how it solves the 2-3 sentence pitch. Most of our prototypes are held together with a mix of duct tape and prayer.

What we’re looking for is customer pull, in order: will they pay? will they loop their boss in? will they take another call without us begging? So far: no, no, and rarely.

We added cold calling this week. EU block in the morning, US in the afternoon (Hubspot for calling, FullEnrich for data). Also brought our product manager in to help with this instead of just founders doing everything. Turns out most people don’t enjoy being cold called. Shocking.

we ignore anything that isn’t customer calls or building prototypes. I bucket tasks (in my brain) into Leverage (5-10x outcome), Neutral (1-2x), or Overhead (0-1x). most things are overhead pretending to be leverage. Allocate only 20% od time to Neutral and Overhead.

Things we’re still bad at: email outreach, consistent follow-ups, building prototypes early enough to send demo videos, getting engineers involved.

Our kryptonite is overintellectualising. Thinking that thinking about the problem will somehow reveal the answer, it won’t. nothing beats 100 customer development calls. We’re learning this the hard way.

GOAL: 30 customer development calls in 2 weeks.

This has been YC without YC Week 3.