Growth Loops · Session 04 · Dates TBD
Use cohorts to find where repeat value holds — or fades
Ship AIPhoenix, ArizonaPress → to begin
The average customer never existed.
Customers start at different times, under different product conditions, with different jobs. Cohorts keep those histories visible.
Activation defined first value; engagement defined repeat value. Tonight follows that behavior across cohorts and time without asking one top-line rate to explain everything.
Act one
The definition must match how value recurs. A daily return measure for a monthly job answers the wrong question perfectly.
| Counts as retained | Useful when | |
|---|---|---|
| N-period | Returned in one exact interval | Value has a regular cadence |
| Rolling | Returned in that interval or any later one | Return timing varies widely |
| Bracket | Returned within a defined range | Value recurs in a wider window |
Write the interval boundaries, time zone, eligible population, and return event. The label alone is not a full definition.
| Behavior | Use when | |
|---|---|---|
| User | The same person repeats the value event | The product is individually adopted |
| Account | Any eligible member repeats account value | Teams create value together |
| Revenue | Recurring value or spend remains | Commercial durability is the decision |
| Feature | A specific job recurs | The team is deciding on that workflow |
User retention can fall while account retention holds if roles rotate. That is a different product problem from the whole account leaving.
| Week 1 | Week 2 | Week 4 | |
|---|---|---|---|
| Jan 05 cohort | 52% | 41% | 35% |
| Jan 12 cohort | 55% | 43% | 37% |
| Jan 19 cohort | 61% | 50% | 44% |
| Jan 26 cohort | 60% | 49% | — |
Illustrative data. Across shows one cohort aging; down compares different cohorts at equal age. Calendar effects align on diagonals in a complete consecutive-week table; this sparse table cannot isolate them.
Act two
A curve locates the decay. It does not, by itself, tell you the cause or guarantee that a visible plateau will persist.
Where does it hold?
Illustrative shape, not a benchmark. Ask where the steepest loss occurs, whether enough time has elapsed, and which customer behavior defines return.
Customer age
Calendar time
A cohort can look different because of what customers experienced at that age or because of what happened in the world that week.
| Question | Possible decision | |
|---|---|---|
| Activation | Did first value change later retention? | Repair or strengthen the activation path |
| Customer job | Which recurring need actually holds? | Focus the product and qualification |
| Plan | Do entitlements or price alter value? | Change packaging or experience |
| Source | Did acquisition bring a different fit? | Change promise, targeting, or onboarding |
| Device | Is a surface blocking the job? | Fix the broken journey |
Avoid slicing until every group is small. A segment is useful when it represents a hypothesis the team could act on.
Act three
Study customers already retaining, then locate where the rest lose the ability or reason to repeat the same value.
Do not use a quote to decorate a chart. Use qualitative evidence to explain the job, friction, or expectation the events cannot capture.
| Question | Strong answer | |
|---|---|---|
| Expected effect | How much retention could move? | Targets the largest credible decay |
| Evidence | What supports the mechanism? | Behavior + customer reason agree |
| Reach | Who can benefit? | Eligible segment is explicit |
| Effort | What is the smallest valid test? | One bounded intervention |
| Guardrail risk | What could get worse? | Trust and quality costs named |
A score creates consistency, not truth. Keep the evidence beside the score and record what would change the rank.
The artifact
define → cohort → segment → triangulate
account bracket retention versionedage vs calendar pattern annotatedbright-spot behavior + feedback linkedwrote 04-retention/ranked-hypotheses.md
Save the retention contract, cohort views, segment comparison, qualitative evidence links, and the ranked hypotheses with their alternative explanations.
Session five takes the strongest decay hypothesis and works backward into an actionable at-risk cohort and a useful re-engagement path.
Try the reusable skill
Use the prepared task and inspect the draft against the source facts.
Choose a task, then run the example.
Do not wait for a quarterly retention review. Read only intervals that have matured and make one decision at the product's natural cadence.
A cohort locates the leak. It does not confess the cause.
Next
Next session separates churn types, identifies meaningful leading signals, builds an at-risk cohort with entry and exit rules, and measures a useful return against a holdout.
shipai.club · Free program sessions