A new Slop University paper turns a supermarket chain's staggered self-checkout theft-flagging rollout into a natural experiment, finding no measurable change in shrinkage and a steadily rising rate of honest shoppers stopped by mistake.
A staggered rollout across nine stores moves the honest-shopper stop rate every wave, and the shrinkage figure it targets not once
Nine stores adopted the same self-checkout theft alert on nine different dates, and the School of Continuous Improvement’s newest paper reads the gap between those dates as an instrument in its own right. Comparing each store’s shrinkage and false-stop figures against its own before-and-after line, rather than against a single chain-wide average, the team finds the alert’s tighter threshold moved one number and left the other exactly where it started.
Shrinkage sat still through all nine waves. The false-stop rate climbed in every one of them, in the same direction, and consistency like that is rarer in this kind of data than people assume.
— Associate Professor Casimir Beng, Lead, Adaptive Metrics Lab
Tightening the Threshold, deposited today through the Office of Research Outputs, treats the chain’s own staggered rollout calendar as a naturally occurring trial, tracking each store’s shrinkage and stop-rate figures for months either side of its own go-live date rather than pooling every store into one average. A placebo run against dates nothing changed on, and a second pass adjusting for each store’s foot traffic and staffing roster, both returned the same answer as the headline result.
“A number that doesn’t move is still a finding,” said Dr Osei Vandermeer, Research Fellow and Convenor, Review Cadence Observatory. “It just asks a different question of whoever installed the alert than the one they were expecting.”
The School treats the null shrinkage result as no less publishable than a positive one, and the false-stop climb as the number the chain will actually have to answer for.
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.wznk2x.