Making Everything Paid

By Georg Malahov

I took a look at the Call Copilot bills. Over the last month transcription cost about a hundred and eighty dollars, and three quarters of that was burned by free and anonymous users. Transcription is billed by the minute of recognized audio, so the bill grows exactly with how much people talked.

Then I sat down to see where exactly the money was going. The picture turned out simple and cheerless. It goes fast, because there are a lot of hour-long and forty-minute calls: people don't bring in a short five-minute stand-up — they bring a full-length call, sometimes two in a row. There's no malice in it; they're using the product exactly the way I designed it.

My biggest failure in this stretch isn't technical: I kept the free trial plan as an entry funnel, when in fact it's a pure expense line. Not "cheap marketing," not "an investment in growth" — just a bill I pay myself, every month, for people who have no intention of paying. I'd been calling it a normal cost of acquisition for so long that I stopped noticing there was no acquisition happening. A free user doesn't turn into a paying one on his own if the free tier is more than enough for him.

Because here's the second number. On Saturday, May 23, I got the fourth purchase in the project's history. Two euros. It's almost funny — I've nearly stopped celebrating. The pace is about one or two purchases a month, the amounts tiny. And next to those two euros, a hundred and eighty dollars of monthly costs look especially expressive.

So here's the decision I've made: I'm closing public free access. The simplest thing to do right now is to shut down the free trial plan entirely and keep only the paid one, or cut the free version way, way down so it stops burning my tokens. You use what you paid for. Those who already pay keep using it as they were — I'm not touching them.

Although "just shut it down" is easy for me to say. I already have live paying users, which means every move now has to be made carefully. On top of that, the plan is to move Call Copilot into the hub and off Supabase, and migrating people with active subscriptions is a separate headache I don't think I've fully pictured yet. There's a lot of work, and none of it is about beautiful code — it's about not breaking anything for the few people who have already put money in.

Checking the decision is simple: after the free plan is closed, the transcription bill should become comparable to what I'm being paid — not equal, but at least the same order of magnitude. If costs drop while purchases stay at their dreary one or two a month, then at least I've stopped paying extra for other people's hour-long calls. But if costs drop to zero together with the purchases, it means the free users weren't an expense — they were everything I had, and then I'll have to admit that, for money, nobody wants the product yet.

For now I'm off into the analytics: what share of users is abusing free access, who burns the most, and how that looks in the overall numbers.