Instead of cold emails, we measured 200 restaurants. The study is now our entire GTM

My studio builds websites and a booking system for local businesses. Classic bootstrapped services plus SaaS mix, with the classic problem: cold outreach from an unknown studio converts terribly, and every competitor's email looks exactly like ours.
So last month we stopped pitching and measured.
We took 200 Amsterdam restaurants, drawn at random from 1,631, and tested one question with a browser agent: can an AI assistant actually complete a reservation? Find the place, understand it, check availability, reach the final booking step. We never submitted a real booking.
The results turned out to be better marketing than anything we could have written:
• 163 of 200 have a working website
• 41% show a booking path a machine can see
• 16 of 200 could be taken to the final booking step by an agent
• 0 of 163 sites publish a booking interface an assistant can call
That last zero became the whole strategy.
Why this beats cold email for us:
1 We now have a real reason to contact every business in the sample. "You were in our study, here is your row, tell us if it is wrong." A correction request converts better than any pitch, and every reply improves the dataset.
2 Journalists take a measured number seriously. The same letter that reads as spam from a vendor reads as a tip from a research team.
3 AI assistants cite studies. Our thesis is literally that machine-readable content wins, so the study ships with schema, open aggregate data and no email wall. It has to pass its own test.
4 The product is the conclusion, not the pitch. We build a 49 euro a month booking layer that assistants can actually call. The study never sells it. It just proves the gap exists, and whoever asks "so what do I do about this" finds the answer one link away.
We launched this week: essay on Medium, the study on our site, letters to the Dutch trade press go out next. The next dataset (Rotterdam, Utrecht, The Hague) is already collecting, so the loop repeats city by city.
The study, if you want to steal the format: https://g-lab.studio/research/amsterdam-restaurants-2026
I will report back with reply rates and what the press does with it. Happy to answer anything about the agent harness or the sampling.