How random clusters are mistaken for patterns

Three deals close on Thursday. By Friday morning the new sales script is “proven”. By lunch, somebody has ordered hoodies.

This is how randomness gets promoted to strategy. We imagine chance as a tasteful, even spread—the statistical equivalent of laying gravel. Real chance arrives in clumps. One city converts brilliantly, five support tickets mention the same button, three excellent candidates appear in a week. The human mind sees a pile of rice and immediately hires a detective.

Sometimes the pile matters. Usually the first explanation does not. If a team has sliced a campaign into twenty audiences, one of them ought to look freakishly good. If revenue bounces every week, one bounce will eventually coincide with the director’s new initiative. Dashboards make matters worse by rendering every twitch in expensive colours, as though the line were shouting for managerial intervention.

I want a higher bar than “that looks interesting”. What variation is normal? How many chances did this pattern have to appear? Did we choose the threshold before seeing the result? Does it repeat somewhere new? Act quickly when delay is dangerous, certainly. But where waiting is cheap, allow reality another throw of the rice. A story can be invented in seconds; evidence has the irritating habit of taking longer.

Behavioural principles

Behavioural ideas at play in this post

Short, plain-English explanations of the principles behind this post, with links to related books and examples in the archive.