We Built What They Asked For. Nobody Used It.

Ask someone what they want for dinner and they'll say something healthy. Watch what they actually order and it's fries, every time. Nobody's lying. They're just answering two different questions.

We had a feature request that showed up on almost every call for two quarters straight. Customers asked for it by name. Sales asked for it. Support asked for it because they were tired of explaining why we didn't have it. We built it. It shipped clean, tested well, no incidents. Usage after ninety days was close to nothing.

What I'd missed, and had been missing for years without a name for it, was that people are decent at telling you what they want and bad at predicting what they'll actually do. I'd been running product decisions off the loudest list in the room instead of watching what people were already doing to work around the gap. There's a whole line of behavioral research on exactly this — people ask for one thing, choose another, and aren't being dishonest when they do it. I recognized it instead of learning it. I'd just never had the shorthand.

The part I don't love admitting: I liked that the request was popular. It made the decision easy to defend in the room. Nobody was going to argue with "customers have been asking for this for two quarters." That ease should have been a signal on its own. The stuff customers were quietly rigging together with spreadsheets and copy-paste, the actual workaround behavior, told us more about what they needed than the request did, and I looked at that data second instead of first.

What I do now, imperfectly, is treat the ask as one input and the workaround as the real one. If people are already solving a problem badly by hand, that's worth more than a feature request with fifty votes on it. I still don't have a clean rule for when the loud ask is the workaround and when it's noise. Sometimes it's both.

I know the counter to this. Sometimes customers ask for the same thing for two years because it's genuinely missing and no amount of watching behavior tells you that, because there's no workaround to watch — they've just given up and stopped trying.

How do you tell the difference between a real request and a request that's really something else?