How People Actually Press the Button

By Georg Malahov

I've spent the first week of August deep in Call Copilot: learned to send users emails, hung a whole pile of extra analytics on it, and started watching what people actually do inside the extension. I almost wish I hadn't.

Here's the picture. A person installs the extension, opens it, presses the record button. Once, twice, three, five times in a row. Recording doesn't start — most often because they haven't given the browser microphone access. I do show an error message, but nobody reads it: the person just presses again, gets frustrated, and uninstalls the extension. That's it, end of funnel. The second story is even more painful: newcomers click "close sidebar" thinking that stops the recording, while the recording calmly keeps running in the background. Tokens burn, recordings get abandoned, the demo gets abandoned. By the feel of it, people poke at the buttons almost at random, and my "perfect" flow, the one I so carefully drew in my head, is completed by literally a handful.

The conclusion from all this is harsh and very simple: I can't quickly convey the product's value and I can't let people try it. Not "I convey it poorly" — I literally can't, because the path to the first "oh, it works" is too long and too full of holes.

And the users aren't to blame here. I broke the onboarding myself — no argument there. I stopped giving away lots of free recordings and piled on restrictions — because I didn't want to be abused. A free demo isn't free for me; every minute of transcription costs money. So I kept inventing the cheapest possible schemes for myself: one demo recording hardwired into the extension instead of a real one, and replaying that same recording; a couple more tricks in the same vein. Economical — yes. But for someone coming in for the first time, that's not a demo, it's an obstacle course.

The transcript dashboard. The yellow banner at the top is that very wall: free calls are used up, registration required from here on.

Yesterday I walked through my own onboarding end to end. Something pops up the whole time, it's unclear where to click. That walkthrough of my own was the trigger — today I'm redoing the onboarding and the demo flow. The main thing: I'm making the demo fully independent of the transcription service (Deepgram), so that a first-time visitor can actually try the product before registering, not after. And separately I need to handle the case where someone closes the sidebar convinced they've stopped the recording.

In parallel I'm running the numbers. Installs are at 6–8 a day right now; I want to get that up to at least twenty — that's close to a hundred a week and at minimum one sale. I do have a way to widen the funnel without extra work: publish a copy of the extension on Firefox and Edge with practically no changes. I'm not launching ads yet — I've tried, it costs money, and on a broken flow it seems pointless; better to be patient. Meanwhile I'm cutting costs: of the big items only Supabase is left, 35–39 euros a month, which currently doesn't pay for itself, because new paying users aren't showing up each month.

I improved a lot this week, and there's almost nothing to show for it — not in the stats, not anywhere. I do have an explanation: with constant changes, stable statistics simply never get a chance to accumulate. First everything has to be stabilized, verified to work as intended, and then left alone.

The nearest check is tomorrow: the subscription of the last remaining active paying client renews. On renewal the credits have to be granted correctly — there used to be a bug there, and after a renewal fewer credits were granted than were due. I'll check. After that I'm giving myself a couple of months to reach 1–3 sales a month and to get the economics into the black — any positive number, however laughable. If two months from now installs are still stuck at eight a day and sales are at zero, then the onboarding isn't the problem.

For now, off to finish the demo that has to work without Deepgram.