· Armen Ter-Oganezov
Seven apps on the App Store: what made money and what did not
I have seven apps on the App Store today and an eighth waiting for review. They fall into two groups that could not be more different, and for a long time I did not see it.
The two groups
The first group is AI tools: a photo generator, a video generator, a music generator. They run on paid APIs, they need a server, and they sell coins and subscriptions. Two of them shipped in June 2024, the music one in January 2025, and for two years they got most of my attention.
The second group is utilities: a sleep timer for the Mac, a cut list optimizer for woodworkers, a home inventory app for insurance claims, an envelope budget. Native apps, one purchase, no account, no server. The sleep timer is the odd one out. It was my very first app, from May 2024, and it took me two years to see that it was the model for everything else. The other three utilities came out in August and September 2026.
What the numbers said
In August I sat down with the sales reports from App Store Connect and looked at one number per app: how much an app keeps from one first download, after Apple’s cut and after the API bills.
The AI apps win on downloads by a mile. Together they bring in roughly forty times as many first downloads as the sleep timer.
The sleep timer wins on money per download, and it is not close. One download of the sleep timer is worth about fourteen downloads of the photo generator and about ninety downloads of the music generator. The video generator loses money on every download, because the API bill for a free trial is bigger than what the few paying users bring.
There is a second pattern in the same reports. A large share of the photo generator’s installs come from one country where the app ranks well but almost nobody pays. In the United States the app is invisible for the big search terms. So the traffic is real, but it is the wrong traffic.
What I changed
I stopped treating downloads as the score. A utility that gets twenty downloads a week and keeps most of the price beats an AI app that gets three hundred and keeps cents.
Every new app since then follows the same rules: it does one job, it works offline, it costs money once, and it needs no server that I have to pay for while the user is asleep. SawKit, Roomkeep and Fundkeep are all built that way. CamDial, the webcam settings app for the Mac, is the same idea and is in review as I write this.
The AI apps still bring people in, and the photo generator has the only real rating count in the whole portfolio. But they get maintenance now, not new features, and their store pages are being rewritten for the searches where people actually pay.
What I would tell myself two years ago
Ship the boring utility first. It will teach you what a paying customer looks like, and you will need that picture before you spend a year on the exciting app.