Reforge — Measuring and Analyzing Retention
Summary
Getting your retention metric right is foundational. Getting it wrong silently poisons every decision downstream. This consolidates Reforge’s framework for defining, measuring, and visualizing retention.
Defining Your Retention Metric
Start from the use case, not from what competitors measure. Five questions to answer:
- What is the problem, in the user’s words?
- Who has the problem?
- What is the core reason the user chooses your product?
- What is their alternative?
- How frequently do they encounter the problem?
The answers give you the core action (what “active” means) and the natural frequency (daily? weekly? monthly?). Your retention metric is: “Did the user perform [core action] within [natural frequency window]?”
Choosing the Right Core Action
To validate your hypothesis, use a two-step process:
- Form groups of users who did the candidate action for four successive periods (e.g., four successive weeks if weekly frequency)
- Create cohort charts for each action hypothesis and compare — the action that produces the flattest, highest retention curve is your core action
Three common mistakes in defining the metric:
- Combining actions — Teams combine multiple actions into one metric, which creates confusion and lets the team optimize for whichever action is easiest to move, not the one that matters
- Optimizing for revenue — Common in SaaS. Revenue is an output, not an input. Revenue retention masks the underlying behavioral problem.
- Wrong frequency — Borrowing from competitors or blog posts instead of understanding your product’s natural cadence
Retention Cohorts
A retention cohort takes a pool of users who signed up in a specific time period and analyzes how many remain active at later periods. Two visualization formats:
- Cohort Chart (heat map) — Shows retention for individual cohorts or averaged across cohorts. Helps disaggregate new user growth from existing user churn. Normalizes for temporal effects (seasonal changes, campaigns, shifts in acquisition quality).
- Retention Curve (line graph) — Shows retention trajectory for individual or averaged cohorts.
Three Retention Curve Patterns
- Trend to zero — The curve never flattens. You will eventually lose 100% of the cohort. Stacking these cohorts produces the “cake graph” that eventually shark-fins back to zero. This is the worst pattern.
- Flattish — The slope is very small, still trending toward zero but slowly. With engagement and resurrection optimizations, you can likely get this to flat. This is “okay.”
- Flat — The curve flattens at some percentage. You permanently retain a portion of every cohort. This is the foundation of growth — stacking flat-curve cohorts produces consistent up-and-to-the-right growth even without increasing acquisition.
The mental model: a flat retention curve is the single most important thing to achieve before investing heavily in acquisition. Without it, acquisition spending is filling a leaky bucket.
Multiple Use Cases
When your product serves multiple use cases with different frequencies, focus on the dominant use case (80%+ of usage). Don’t overcomplicate by defining multiple retention metrics unless the use cases are truly distinct and roughly equal in size.
The Use Case Frequency Spectrum divides products into the Habit Zone (monthly or more frequent) and the Forgettable Zone (less than monthly). Products in the Forgettable Zone must reacquire users each cycle or transition them to a more frequent use case.
Relevance to projects:
- 01-projects/squarely-puzzles/index — The core action should be “completed a puzzle” not “opened the app.” Need to validate by comparing cohort curves for each action hypothesis. Natural frequency hypothesis: daily for engaged users, but need data.
- 01-projects/data-marketplace/index — Multiple use cases (data buyer, data seller, data browser) likely need the dominant use case identified first. Revenue retention is the wrong metric — behavioral retention (searched for data? downloaded a dataset?) is the input.
- 01-projects/newsletter/index — Core action: “opened email” vs. “clicked a link.” Need to test which produces flatter cohort curves. Natural frequency is weekly (publication cadence).
Connects to 06-reference/2026-04-03-reforge-retention-is-the-output (retention as output of activation/engagement/resurrection), 06-reference/2026-04-03-reforge-engagement-activation (engagement as the depth input to retention), and 06-reference/2026-04-03-saas-metrics-that-matter (retention as foundation of LTV).
Open Questions
- For 01-projects/squarely-puzzles/index, what does “flattish” look like for a puzzle game? What’s a realistic long-term retention percentage to aim for in casual gaming?
- Should the 01-projects/data-marketplace/index even attempt to define a single retention metric, or is it fundamentally multi-use-case?
- How do you build cohort charts when you’re pre-launch and have no data? What assumptions are safe to borrow from comparable products?