Cohort

How cohort analysis works

The classic output is a retention grid: rows are acquisition periods, columns are periods since acquisition, cells show what share of each cohort was still active — or how much it had spent — after one month, two months, three. Read down a column and you see whether newer cohorts behave better than older ones; that single view answers questions a blended average cannot, like whether last quarter’s discount campaign bought customers who reorder or customers who churn. The same structure feeds customer lifetime value: LTV is only trustworthy when computed per cohort, because a blended figure mixes loyal 2023 customers with untested ones acquired last month.

What per-user cohorts require

To place a visitor in a cohort and find them again months later, an analytics tool must recognise the same individual across sessions — a persistent identifier stored in the browser or derived from the device. That is per-user tracking, with everything it implies in the EU: consent under ePrivacy, GDPR obligations, and coverage limited to whoever accepted the banner. It is the same requirement that underpins multi-touch attribution, and it fails the same way: with 40-60% of EU visitors rejecting consent, browser-based cohorts are built from the consenting minority, and their retention curves describe that minority only.

What SealMetrics does and does not do

SealMetrics does not build per-user cohorts, deliberately. Its architecture is anonymous, aggregate event measurement: no persistent visitor identifier is ever created, so there is no mechanism for recognising an individual across sessions — the precondition for behavioural cohort tracking. What it provides instead is aggregate comparison over time on 100% of traffic: acquisition, conversions and revenue by channel and period, segmentable and unbiased by consent status. For purchase-based cohorts — the kind most retention and LTV work actually uses — the natural home is your order database, where customer identity already persists lawfully; SealMetrics’ role is supplying the complete channel-level acquisition data those cohorts are joined against. If per-user behavioural cohorts inside the analytics tool are a hard requirement, SealMetrics is the wrong tool for that job, and it is designed to be.

What cohort analysis does not tell you

Cohorts show that groups differ, not why — attributing a retention gap to the campaign, the season or the product mix requires controlled testing. Small cohorts produce noisy curves that invite overreading, and cohorts built on incomplete browser data inherit its bias: a retention change can reflect a shift in who accepts consent banners rather than in who stays a customer.