Why the exit button moves after the second visit

The cancel link was grey text at the bottom last month. Now it is grey text inside a submenu. Nobody redesigned the page in the ordinary sense; a test ran, one arrangement produced fewer cancellations, and the arrangement that produced fewer cancellations was kept.

What everyone sees

Interface drift is read as churn from redesigns or as an accumulation of small mistakes. Neither describes the process. Continuous experimentation is a well-documented engineering practice with a clear selection criterion, and the criterion is the metric the team is accountable for — not the user’s ease of leaving.

What is actually happening

Kohavi, Longbotham, Sommerfield and Henne describe controlled online experimentation as standard practice, with large organisations running thousands of concurrent tests and shipping whichever variant improves the target metric, often by margins invisible to any individual user. Mathur and colleagues’ crawl of shopping sites documents what accumulates when the target metric conflicts with the user’s intent: obstruction patterns that make cancellation, unsubscription and account deletion measurably harder while remaining technically available. Put together, no designer needs to intend hostility. A selection process that optimises retention, applied repeatedly to the same page, will find and keep whatever makes leaving slightly harder, because that is what the fitness function rewards.

Why it stays hidden

The process hides because each increment is small and each is defensible in isolation. No single change is worth complaining about, the previous version is not available for comparison, and the user has no memory of layouts precise enough to notice drift. Selection without a designer looks like it was never decided.