If It Won't Grow, Let It Cost Nothing

By Georg Malahov

Shipped version 1.9.10 overnight. Made the demo fully simulated — so it shows how everything works without asking me for money on every run. Fixed the onboarding; it should walk people through properly now. Emails go out, tokens get counted, stats get collected. Lovely.

And it was those very stats that brought the bad news: almost 30 people left in the last three days. That's a lot. I used to have a daily inflow; now it's an outflow, and I don't fully understand how or why. Installs are down about half, if not worse. A few guesses: the first cohort is leaving — the people I once brought in with ads who, after the recent updates, can no longer record calls for free; plus mandatory registration; plus I changed the listing card, and the Chrome Web Store seems to demote a listing in recommendations for a while after an update; plus it's August and vacation season. Which of these is true — I don't know.

The logical move would be to go get traffic. I did — and failed. The idea was simple: find existing comparison articles that list extensions like mine and buy a spot or a link there. Couldn't find a single suitable one. Signed up on AlternativeTo — and landed 144th in the alternatives list for my category. Nobody will ever find me there. An evening spent, zero result.

So I turned the other way. If growth won't come yet, I'll work on costs — so the project at least stops burning money while it looks for its audience. Sat down with Claude to go through what I'm actually paying for. And ran a risky experiment while I was at it: gave the agent access to Hetzner through the browser, over SSH, and to GitHub — essentially handed over nearly my entire server infrastructure. It broke nothing and wrote a report with a cleanup plan. My experiments keep getting more dangerous, and when exactly that comes back to bite me is an open question.

We did the math. Paid Supabase — about 32 euros; I moved the project to the free tier and covered the domain lost in the move with the simplest possible proxy through my own server. The Miro board where I ran a now-finished client project moved from the Business plan to Starter — another eighteen euros a month off. And a switch to a different Hetzner plan: twice the CPU and 6 euros cheaper. Though the server I need isn't available in Nürnberg anywhere yet — yesterday nowhere at all, today it's in Finland — so I'm waiting and will probably set up availability monitoring. All told — just under 700 euros a year saved. All that's left of Call Copilot's costs is tokens for Deepgram, the transcription service, and I have a stockpile of those.

My one paying user at 12.99 is alive and well: on the renewal date Stripe had scheduled, the payment went through — he didn't cancel the subscription, renewed for a second month, and the fixed token-crediting bug did its job. No new paying users have appeared after all these changes, and so far nobody has canceled.

I've reframed for myself what counts as success — it's no longer about growth. The project should cost me practically zero and just sit there running on its own, without my involvement and without a monthly bill. If that holds, I can calmly spend the next three to six months on something I've never done: learning free traffic. Build up a Reddit account, where self-promotion is supposed to stay under a tenth of your posts, publish the extension on Firefox and Edge as well, write articles. And if there's still no traffic in six months — then costs were never the problem.

For now I'm calling the product temporarily done and waiting for signals from the market. A year's savings is roughly the price of an Oura Ring — that's a result too.