# Priors _The stat is real. It's not a prediction._ --- A customer has been with you for eight years. Embedded in the product, strong usage, the key contact fully trained on the platform. They've renewed without hesitation every year. This morning your account manager flags it: they've gone to tender. Your finance director pulls the data. Over the past five years, 75% of customers who went to tender ultimately churned. Three in four. Are they already out the door? --- If someone had asked you yesterday for their churn risk, you'd have said 5%. That can't have jumped to 75% overnight. You know there are ex-customers in that number with a very different profile, accounts that were already in trouble before they ever tendered. The 75% covers everyone who went to tender, sick and healthy alike. But the tender means something. This account isn't at 5% anymore either. So if it's not 5%, and it's not 75%, what is it? --- Take a hundred accounts that look like this one, all at 5% churn risk. Five would end up churning. Ninety-five would renew. Of those who churn, almost all go to tender first. Call it all five. But tendering isn't exclusive to churners. About 15% of customers who renew happily also go to tender at some point. Procurement cycles, new finance directors, compliance box-ticking. 15% of 95 is fourteen. | | Would churn | Would renew | |---|---|---| | Accounts like this one | 5 | 95 | | Go to tender | 5 | 14 | Nineteen tenders. Five from accounts that are genuinely leaving. Fourteen from accounts going through a procurement exercise and coming back. Five out of nineteen. About 25%. --- The tender moved this account from 5% to 25%. A fivefold jump, worth taking seriously. But 25% and 75% are different worlds. At 75% you're in rescue mode. Discount the renewal, fly out the exec team, make concessions before anyone has asked for them. At 25% you're alert but curious. Is this a genuine evaluation or a procurement obligation? Has a new finance director arrived who wants a process on record? Has something changed that the usage data didn't catch? Those questions only get asked if you haven't already panicked. That's Bayes' theorem in practice. ---