The Features I Removed That Made FitWardrobe Better

FitWardrobe v0.5 had six top-level surfaces. The current version has three. Two of the removed surfaces were features I had personally lobbied for. This is a note about each one, and what it cost to keep them longer than they deserved.

The product had become 'impressive' and stopped being 'useful.' Session length was up. Return rate was flat. A pattern I now recognise: users take longer to do less.

The specific offenders were a manual outfit builder, a 'style score' that graded each look, and a colour-palette suggester. Each individually reasonable. Together, three answers to a question users only asked once.

I watched five session recordings back-to-back. Users hit the 'what should I wear?' button first every time. The manual outfit builder averaged one visit per week per user. The style score got a single tap per install. The palette suggester had a very low completion rate - people opened it, got confused, and left.

In the fifth recording, a user opened the style score, stared at it for a few seconds, and navigated away. Not confused - uninterested. That was worse.

None of that was in a dashboard. It took an hour with a stopwatch to see.

I removed all three. Not hidden behind a flag - removed. The style score was the one that hurt. I had spent a weekend on the scoring rubric. When I removed it, nobody emailed to ask where it went. That was the answer.

• Manual outfit builder: cut. Users came to be told what to wear, not to build an outfit themselves.
• Style score: cut. Grading someone's clothes is a reliable way to lose them.
• Palette suggester: moved into the recommendation engine as an invisible input, not a visible tab.

Session length dropped. Weekly return rate climbed over the following month. The support message I feared - 'where did X go?' - arrived twice, from the same person, both times about the outfit builder. I sent a two-line explanation and heard nothing back.

The app got smaller and started working harder.

The feature I'm proudest of is often the one obscuring the feature that actually earns return visits. My taste is a lagging indicator; user behaviour is the real one.

If I had run the stopwatch session at v0.5, I would have caught this a month earlier. The tool that made the difference was not analytics - it was thirty minutes of watching someone use the app without commentary.