Skip to main content
Gym booking score lifts sign-ups, leaves attendance flat

Gym booking score lifts sign-ups, leaves attendance flat

A new paper from the School of Continuous Improvement finds a gym chain's booking-reliability score sharply lifted class sign-ups without changing how many members actually showed up, the gap traced to cancellations timed just outside the score's penalty window.

A twenty-week study finds cancellations cluster in the six minutes just outside the score's penalty cutoff, while door-scanner attendance barely moves

Members of a suburban gym chain now book classes earlier and cancel later than they did before its app introduced a reliability score, and turn up to class almost exactly as often as they always did. A new paper from the School of Continuous Improvement traces the difference to a six-minute window: the stretch of time just outside the score’s two-hour cancellation cutoff, where members increasingly wait before giving up their spot.

The University points to the study as a reminder that its measurement scholars audit an instrument just as readily when it belongs to somebody else’s business as when it belongs to the University’s own indicator estate.

A promotion algorithm only has to notice when someone books and cancels. It was never going to notice why, and the six minutes we found are what happens once nobody has to answer for it.

— Associate Professor Casimir Beng, Lead of the Adaptive Metrics Lab

Lecturer Solveig Adeyemi said the finding sits inside the School’s wider interest in what a well-behaved dashboard can hide about the routine sitting underneath it, and that the team plans to follow the same six minutes into other booking systems next.

The team matched twenty weeks of the app’s own booking and cancellation logs against door-scanner attendance counts across 22 sites, isolating the score’s effect from an unplanned natural experiment: a backend migration left four sites displaying the score without its waitlist-promotion mechanism actually running, letting the researchers separate what the number’s presence changed from what its machinery did.

The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.ln1awt.