When to kill a feature you already shipped

Cutting something before it ships is a decision on paper. Nobody has used it. There is no changelog entry to explain away, no person who might notice, no version of events where you admit the first instinct was wrong in public. Killing something that is already live is a different kind of decision entirely, and none of the tests I wrote about for trimming an MVP actually help with it, because they all assume the thing hasn't happened yet.

The feature was a confidence slider

About six weeks after launch I added a small control under each of Appray's eight rounds of questions: a slider from one to five asking how sure you were of the answer you just gave. The reasoning felt sound at the time. A Build verdict built on five confident answers should probably count for more than one built on five answers where someone was mostly guessing, and the slider was a cheap way to capture that difference and feed it into the verdict.

It took about a week and a half to build, including the little animation on the handle that I was, if I am honest, disproportionately pleased with.

What actually happened

People left it alone. I added basic local logging so I could see slider positions without seeing anything personal about the answers themselves, and after the first month, 91 percent of sliders were sitting exactly where they had opened, dead centre. Not because everyone felt exactly medium confident about everything. Because a slider under a question you are already concentrating on is one more decision nobody asked for, and the easiest way through it is to not touch it.

The verdict logic barely changed either. I had built the weighting conservatively, so a room full of default answers moved the outcome by a few percentage points at most, which is another way of saying the whole feature was mostly decorative.

Why I didn't remove it immediately

This is the part I find more interesting than the slider itself. I knew within three weeks that it wasn't doing anything. I did not remove it for another two.

Some of that was ordinary sunk cost, the kind that has a name because everyone falls for it the same way. I had spent real hours on it, written a paragraph about it in the FAQ, and there is a specific reluctance to undoing something you were proud of building, separate from any argument about whether it works. Some of it was more specific to shipping publicly. Cutting an idea from a list is invisible. Removing a feature from a live app shows up in a changelog that people can read, and some part of me treated that as evidence I had gotten something wrong in front of an audience, rather than as the ordinary, boring maintenance it actually was.

Neither of those is a real reason. Both of them are exactly the kind of reasoning Appray exists to catch when someone else is doing it to themselves about a feature they haven't built yet. It is considerably less funny when it's your own reasoning about a feature you already shipped.

What finally moved me

Not a metric. I had the 91 percent number for two weeks and it changed nothing, because a number I already expected doesn't feel like new information, it just feels like a number.

What moved me was an email. Someone wrote in asking what the slider was actually for and whether they were doing it wrong by leaving it in the middle. That was the third question like it I had gotten. Three people confused by something is a small sample, but it told me something the usage log couldn't: the feature wasn't neutral. It was quietly asking a small percentage of people to stop and worry about a control that did almost nothing, in an app whose entire pitch is that it respects your time. A feature nobody uses is a shrug. A feature that occasionally makes someone stop and second-guess themselves for no reason is a small tax, and I had been charging it for five weeks after I already knew.

The part I still got wrong

I removed the slider in the next update and said nothing about it anywhere. No changelog line, no note in the FAQ, nothing. My reasoning was that a feature nobody used didn't need an announcement to explain its absence.

That was its own small mistake. Two of the same three people who had asked about the slider wrote back a few weeks later asking if I had removed it on purpose or if something was broken. Killing a shipped feature quietly isn't the same as never building it. Somebody already has a mental model that includes it, however few people that is, and a silent removal reads as a bug report waiting to happen rather than a decision. One line in the changelog would have cost me four minutes and saved two confused emails. I write that line now, every time, even when I am fairly sure nobody will read it.

The MVP tests taught me what to leave out before I start. This one taught me the separate, slower skill of admitting that something I already shipped was the wrong call, on a timeline nobody was forcing on me, which turns out to be a different muscle entirely.