10% Activation, Zero Payments — I'm Turning Off the Ads

By Georg Malahov

For the last two weeks of February I wasn't working on the product — I was working on finally seeing the numbers. I built an n8n workflow that updates user stats on its own: who came back, who burned through the free limit, who actually recorded something real, and who installed and left. I bolted on ad-stats collection and install attribution too. Then I generated myself a local admin panel to see it all in one place and send emails from there. Essentially I put together a small CRM for a project that doesn't have a single payment yet.

While I was building it, something unexpected hit. On Sunday I got abused hard: in a single day someone cranked through forty hours of transcription and almost twenty-five dollars of my limit. The hole was in how the extension gets its token for Deepgram — the transcription service everything runs through for me: anyone could request a token. I closed it. And immediately shipped 1.7.11 with that protection — and cut off my existing users with my own hands: the new check expected an Install ID, and old versions simply don't send one. Classic: you fix a hole and break the workflow for people who did nothing wrong.

The second surprise was even more galling. I'd been counting installs wrong. I count a function call that fires on install — but it also fires when the extension syncs to a second device of the same person. So the number I'd been staring at for weeks, the one I'd been happy about, was inflated. I had to recount everything.

Then comes the arithmetic all of this was built for. Eighty to ninety percent of users see the paywall. Activation, if you count it as two recordings (meaning that after the demo, a person voluntarily made a full recording on their own), is ten percent. If you count one recording, it's thirty, but the first number gives the real picture. At a payment conversion of even one percent, the ads don't pay off even in the most optimistic scenario — and my traffic mostly comes from countries where people simply won't pay for a transcription tool like this. By that point I had already cut the ad budget down to about six euros a day. Over that same weekend I went through about eighty euros on development tokens.

All this time I'd been iterating on the paywall: a thirty-day guarantee, Stripe coupons, discounts, a countdown timer. Careful, pretty, and completely beside the point. A paywall fixes the step from seeing it to paying, but my hole is one floor up — people get to it, they just don't use the product. I didn't lower the price: lowering the price at ten percent activation just means selling something that doesn't work, cheaper.

And the most galling failure of the month: the emails. I never finished the templates, never finished the automation, never sent a single email to users. The tool is built and has never been used even once.

As of Monday the ads are off. Not because I'm in a bad mood, but because at these numbers paid traffic means buying people who don't activate. I'll test the decision with a month without ads. If organic installs don't collapse to zero and at least one real payment comes in — the channel exists, and it's worth working on SEO further. If it's zero there too — the problem isn't traffic, and what needs rebuilding isn't the ads but the thing people are supposed to pay for in the first place.

For now I'm off to finish those damn email templates. I'm also thinking about a free tier for people who run everything locally — but that's another story.