519 people showed up. One got a message. My own numbers.

FirstCustomer's whole bet is that the person who needs your product is already writing about their problem somewhere, and that if you can find that post, you can find the customer. So I pointed it at my own product. It surfaced 519 people who matched. I contacted one.

That is not a rounding error and it is not a bad week. It is the real number, from the real database, on the day I pulled it: 2,541 signals collected across 25 real users who had actually run the pipeline, snapshot taken August 23. Out of those signals, 519 reached the stage where the product considers someone a customer, a person worth reaching out to. One of those 519 was moved to contacted. That is a 0.2 percent rate, and it is not FirstCustomer's rate on behalf of its users. It is mine, on behalf of myself, using the product I built to find my own first customers.

For context on how fast this happened: the idea came out of a hackathon on August 10, and the product was live in production, with payments and enforcement already wired in, four days later. By the time this audit was taken, the codebase had 674 commits behind it in 24 days, built by one person. None of that speed is the point of this post. I am only naming it so the next number lands correctly: this was not a slow, careful rollout where outreach had months to catch up. The finding tool was fast. The human step next to it was not, and speed was never the variable that was missing.

The part I expected to be hard

Before I shipped anything, the assumption I built the company on was that finding people would be the hard part. Founders do not have a channel to their first ten users, and going looking for one, one post at a time, is slow, manual work that most people give up on before it produces anything. That assumption held up. The pipeline did what it was supposed to do. It went out, read real posts, scored them, and handed back a list of people who looked like they had the exact problem the product solves. It did this 519 times over, across 25 different founders using it for 25 different ideas, and it did it without me touching a keyboard for any single one of those 519.

So the part I expected to be hard was not hard. Finding people is not the bottleneck. The audit that produced these numbers says it plainly: 22 of those 2,541 signals were ever opened by the person the pipeline surfaced them for, which is 0.9 percent. Zero were marked as a favorite. Zero were marked with a thumbs down, which means nobody was even bothering to tell the product it was wrong. Not liking a result and not looking at a result are different failures, and this one is the second kind. The list was sitting there. Nobody was reading it, including me.

The part I did not expect

Here is the uncomfortable part, and it is not a product bug. Having the list did not make me act on it. I had built a machine specifically to remove the excuse of "I do not know who to talk to," and once it removed that excuse, I found a new one. I did not open the list. When I did open it, I did not write the message. When I wrote one, in the entire audited period, I sent it once.

I want to be precise about what this is not. It is not that the 519 were bad matches and I correctly ignored them, because I never looked closely enough to make that judgment. It is not that I was too busy building, because building has a closed loop: you write code, it runs, it either compiles or it does not, and you get an answer in seconds. Reaching out to a stranger and asking about their problem has no such loop. Nobody tells you within the hour whether the message worked. There is a wait, and there is a chance of silence, and there is a chance of an answer you did not want. I built a tool that automates the part with a closed loop and left the part with no closed loop exactly where it was: on me, unautomated, and easy to defer for one more day.

That is the actual finding. The bottleneck did not stay where the product's design assumed it would. It moved from "can you find the person" to "will you send the message," and the second question is not one a pipeline can answer for you, because the pipeline is not the one who has to sit with the possibility of being ignored, or worse, of being told no by someone whose problem you thought you understood.

Why I am not calling this solved

I already wrote about the founders whose posts I read to design this product, and that corpus told a version of the same story from the other side: people running actual outreach campaigns, hundreds of cold messages, most getting single-digit reply counts. What I have here is a smaller, meaner version of it. Those founders at least sent the messages. I had a list ready-made for me, curated, scored, delivered without effort, and I still only acted on one row of it.

I do not have a clean explanation for why, and I am not going to invent one to make this post feel resolved. I could say I was heads-down on shipping and tracking got deprioritized, and that would be true, but it would also be true every week since, so it explains a pattern, not an excuse for one week. I could say the 519 did not look promising enough on a skim, but I did not skim closely enough, at 0.9 percent opened, to have formed that opinion honestly. The honest answer is that reaching out asks something building does not: it asks you to be the one who might get a bad answer, in person, from someone real, instead of getting a bad answer from a compiler where nobody is watching.

What this changes about how I read the rest of my own numbers

I have been treating the funnel from signup to activation as the number that mattered, and by that measure the early weeks looked fine, roughly a third of new signups ran the pipeline at least once. But activation, in this product, means someone pressed a button. It says nothing about whether the button's output ever turned into a conversation with a customer. On my own account, the honest funnel is not signups, or activation, or even the 519 that got surfaced. It is 519 to 1, and no amount of improving the first three steps changes the fact that the last one is a human being deciding, alone, whether to hit send.

I am not going to close this by announcing the feature that fixes it, because I do not think a feature fixes it. A reminder does not make sending a message to a stranger feel less exposed. A better list does not either, not when 0.9 percent of the last list ever got opened. What I can do is stop treating "the tool found them" as if it were the same claim as "someone talked to them," because right now, for my own product, on my own numbers, it very much is not.

If you are building something similar and your own dashboard shows a healthy number of people surfaced, look one column over. Mine did too. The number that mattered was next to it, and it was a 1.