Measure returning use with an explicit entry cohort and return window

Choose a return that matters to the reader task

A useful question might be whether people who first use a reporting aid return to use it in a later period. Choose the actual permitted start and return events and a stable reporting unit. A later pageview may be a different event from completing the aid; neither should silently replace the other.

Use only properly authorized existing website/product observations. Anonymous browser/device identifiers do not guarantee one person across devices, and absent collection is not proof of no return. Do not fingerprint readers, identify LinkedIn viewers or invent a history merely to complete a retention table.

Freeze the cohort and distinguish first-entry modes

Write the entry dates and whether entry means first ever, first within the report window or recurring activity. PostHog documents these as different retention choices. A limited collection history may support first observed activity but cannot establish first ever use before that history existed.

PostHog also distinguishes static cohorts from lists that update with criteria. A current dynamic list can change membership and is not automatically the cohort used in an earlier report. Preserve the actual definition, snapshot time and exclusions; this guide creates no cohort or user profile in a service.

Match age, return criterion and complete observation

Exact-period retention asks whether a unit returned in that particular interval. On-or-after retention also includes later qualifying returns within the observed horizon. They can give different values from the same ledger. Keep the horizon and the rule in the label instead of calling both day-seven retention.

Periods still in progress are incomplete, not observed failures. Compare cohorts at the same age or clearly show the unobserved cells. Also keep the percentage base: original cohort size differs from a previous-period base. Exact-period returns can rise after a quieter period because the same people can return in several periods.

Resolve three constructed cohort readings

R1 — Constructed fixed cohort of eight units: three return in day-one’s declared bucket and four in day-seven’s bucket. Decision: report 37.5% and 50% against the same initial eight. Outcome: a possible pattern of periodic return, not an arithmetic error or proof that four new people entered the cohort.

R2 — Constructed maturity gap: an older cohort has a complete seven-day observation; a new cohort has only two days observed. Decision: leave the new day-seven value unavailable. Outcome: do not call its missing future window 0% retention or compare it as a completed failure.

R3 — Constructed rule mismatch: two of eight units return exactly in the declared day-seven bucket and two other units return only on day-nine within the fixed horizon. Decision: report 25% exact-period versus 50% on-or-after. Outcome: two correctly labeled questions, not contradictory measurements of one unchanged metric.

Keep identity, aggregation and usefulness separate

A unit can appear in more than one return period, so adding period counts does not count distinct retained people. When combining cohorts, preserve weighted versus simple averages and their bases. PostHog’s breakdown choices can also require matching property values at start and return; changing devices or labels may change the reported subset.

Returning activity is not proof of satisfaction, completed reading or payment. A person may return to fix an error; an infrequent but useful annual aid may not need weekly use. Interpret the behavior alongside the actual task rather than importing a universal retention target from another product.

Keep a reproducible entry-to-return record

The private worksheet records event definitions, permitted identity/unit, cohort snapshot, exclusions, period boundaries, maturity, return rule, base and aggregate calculation. Use only minimal allowed data; the constructed packet needs no real identifiers or analytics upload.

Current provider definitions were read on 1 November 2026. Cases are fictional and do not measure this blog’s actual retention. No actual account or cohort was created or changed.