Zera Zheng runs supply chain resilience products at Maersk, and this week she shared a research finding that belongs on every roadmap wall. When prospective apartment residents are asked whether they want a gym, around 80 percent say yes. Around 20 percent ever use it.
The building still pays for the gym. The equipment gets bought, the space gets cleaned, the cost lands in everyone's rent. Meanwhile the thing that actually moves how residents feel about the place, in the same research, is far less glamorous: how reliably problems get fixed when something breaks.
Swap residents for users and the gym for your feature backlog, and the number barely changes.
Why do people ask for things they will never use?
Because answering a question costs nothing and using a product costs effort. The person who answers your survey is the person they would like to be. The person who uses your product is the person they are. The aspirational self wants the gym, the language course, the analytics dashboard. The actual self wants the elevator to work.
Nobody is lying to you. A yes in a survey is a sincere opinion about a hypothetical life. That is exactly why it cannot carry the weight founders put on it.
What does the empty gym look like in software?
Two small examples crossed our research desk the same morning as the gym stat. An indie builder running a sound effects library for video editors published his numbers: the sound everyone would name in a survey was being outdownloaded three to one by one almost nobody requested. His download logs knew his users better than his users did.
A founder of a data product described his funnel in a forum thread: people sign up and never touch their free credits, or spend the credits and never place a paid order. Every signup was a stated yes. The unused credits are the empty gym, counted in rows instead of treadmills.
How do you tell a gym request from a real one?
Put a cost next to the yes. Not necessarily money: effort, time, information and reputation all work. Someone who describes the spreadsheet they maintain today to work around the missing feature is showing you behavior. Someone who agrees your feature sounds useful is showing you politeness.
The evidence ladder is short. What people say sits at the bottom. What they already do sits in the middle. What they pay for sits at the top. A feature request lives on the bottom rung until you find the behavior that backs it, in your own usage data or in the daily routine of a real person who matches your buyer.
And the inverse matters just as much. The unglamorous feature nobody requests, the fix-it-fast equivalent of your product, rarely shows up in surveys because it is assumed. Usage data and real conversations with real users are the only places it appears.
Key takeaways
- Stated preference inflates because saying yes is free. In the rental research, 80 percent wanted a gym and about 20 percent used it.
- Usage data contradicts surveys constantly: downloads, logins and spent credits are revealed preference.
- Price every request in effort or money before it enters the roadmap. Behavior backs it or it stays a hypothesis.
- The features that retain people are often the ones nobody requests, because they are assumed. Only behavior and real conversations surface them.
Why this today: the gym numbers were shared this week by a product leader at Maersk, and the same morning brought the sound library downloads and the untouched free credits. Three datasets, one lesson, zero coordination.
The survey builds the gym. Attendance decides if it mattered.