Note · 18 Feb 2026

The quiet week problem in B2B apps

Consumer products can often treat a blank week as a warning. Workplace products cannot. A finance admin who is silent until month-end is not a lost lover. She is doing the job the product was built around, on a calendar the score refuses to see.

App Engagement Scoring that averages activity into seven-day buckets will punish that admin and reward a colleague who clicks around because they are lost. The lost colleague looks “engaged.” The competent one looks like churn. Sales then receives a list of “at risk” accounts that are simply waiting for invoices.

A quieter model keeps a cadence prior: expected gaps by role, by tenant, by billing cycle. Silence inside the prior is scored as rest. Silence outside the prior, after a history of weekly work, is scored as a question. Questions belong in a research queue, not in an automated win-back email.

New tenants break this. With no history, the prior is a guess. We teach people to exclude those tenants from the mature score and to report them separately. Mixing them in is how a launch cohort drags the mean into a story nobody can use.

If your executives want a single traffic light anyway, write the paragraph first. The light is optional. The paragraph is the product.