What I actually do with every feature request that comes in

People sometimes ask what my process looks like, expecting a board with columns and swimlanes. It's a text note, no labels, no priority score next to anything in it. I read it top to bottom most mornings, and most mornings nothing in it changes at all.

That sounds lazier than it is. The note isn't where decisions happen. It's where requests wait until they've earned one, and most of them never do.

Where the lines actually come from

Four sources, roughly in order of how much I trust them: App Store reviews, support emails, my own list of things I notice while using the app myself, and the odd mention on Reddit or Mastodon where someone's tagged Appray without writing to me directly. The nineteen Android emails I've written about before came in through the second channel. A confidence slider I built and later killed came from a pattern I only caught in the first.

None of it arrives labelled as a feature request. Most of it is a complaint, or one line at the end of an otherwise glowing review, or a screenshot with no text attached at all. Turning that into something I can act on is most of the actual work, and it's the part no tool does for you.

The board I built and abandoned

Early on I set up a proper board. Columns, drag and drop, a vote count on every card. I used it for about three weeks and then quietly stopped opening it, which is its own kind of answer.

The problem wasn't the tool, it was what the tool did to my head. The moment a request has a card, it looks like a decision has already half been made. Cards sitting in a column called "Considering" accumulate weight just by existing there. I caught myself defending things I'd never actually agreed to build, purely because deleting a card felt more final than deleting a line of text.

A plain note doesn't do that to me. Nothing in it looks like a commitment, so nothing in it costs anything to remove.

Three questions, not a priority number

When I sit down to work through the note, I'm not ranking anything. I ask three things about each line: who is actually asking, is this the same story from more than one person, and would I have thought of it myself if nobody had asked.

One person asking, on its own, almost never becomes a build. Not because one person doesn't matter, but because a single description of a problem can be a misunderstanding of how the app already works, a one-off situation nobody else will hit, or genuinely a job only that person has. I still answer every one of them myself. I just don't let a single voice, however clearly written, set direction on its own.

The same story from people who don't know each other is the strongest signal I have, stronger than any vote count a card would have shown me. The third question is the one I trust least: my own idea, arriving fully formed and urgent, with nobody outside prompting it. Those are usually the ones that are fun to build rather than needed, and the distinction is harder to feel in the moment than it sounds on paper.

The pattern I nearly missed twice

Here's where the plain note nearly failed me. For most of the summer I had a line that read "export, send to accountant," attributed to one email in June. One person, filed under watch, don't build, exactly where the system said it should sit.

It should have moved in July, when someone left an App Store review mentioning the same thing in different words, wanting a PDF they could forward on. I read the review, didn't connect it to the June email, and logged it as a separate, unrelated one-off. Two data points sitting in two different places in my head, not next to each other on a page.

It only became visible in August, when a third person emailed asking almost the exact question the June person had, close enough in wording that I actually searched my own inbox for it. Three requests, three different channels, eight weeks apart, and a system built to catch exactly this had still missed it, because I was reading each channel in isolation as it arrived instead of checking new lines against the old ones.

I built a basic export the following week. Four days of work. The fix to my own process took longer: I now keep every request in one place regardless of which channel it came in from, and I search the existing lines before deciding where a new one goes. Obvious in hindsight. It wasn't obvious while I was living inside three separate inboxes.

What I won't turn this into

People occasionally suggest a public roadmap or a voting board, and I keep saying no. Not because I don't want feedback, I clearly want quite a lot of it. A public vote changes what people write, though. Instead of describing what actually happened to them, they start campaigning for the outcome they've already decided on, and campaigning is a worse source of truth than a complaint written by someone with no idea I'm reading it as data.

The note stays private, plain, and small enough that I can hold all of it in my head at once. That last part matters more than any of the questions I ask about it. The day the list gets too long to read in one sitting is the day I'll need an actual system, and I'm not looking forward to it.