--- title: Cohort Analysis and Retention Curves — RGM Training url: https://realgrowthmatters.com/training/marketing-analytics/cohort-analysis-and-retention-curves/ updated: 2026-06-10 source_html: https://realgrowthmatters.com/training/marketing-analytics/cohort-analysis-and-retention-curves/ --- [Home](../../../index.html) › [Training](../../index.html) › [Marketing Analytics](../index.html) › Cohort Analysis and Retention Curves RGM° · Training # Cohort Analysis and Retention Curves The most under-used analytical method. Cohort types, retention curves, the smile signal, segmentation, and the techniques to operationalize. ### What you will learn 1. [Why cohort analysis is the most under-used analytical method](#why) 2. [Cohorts defined](#concept) 3. [Types of cohorts](#types) 4. [Retention curves: classic patterns](#retention-curves) 5. [The smile curve and product-market fit signal](#smile) 6. [Cohort analysis techniques](#analysis) 7. [Cohort segmentation](#segmentation) 8. [Tools for cohort analysis](#tools) 9. [Advanced playbook](#advanced) 10. [Common mistakes](#mistakes) 11. [Operating checklist](#checklist) ## Why cohort analysis matters Aggregate metrics lie. They tell you how the average customer behaves — which describes nobody. Cohort analysis reveals what's actually happening: are newer customers more or less retained than older? Are customers from specific channels more valuable? Is product-market fit improving or declining? The discipline: most growth questions are cohort questions. Lifetime value is cohort-based. Retention is cohort-based. Channel effectiveness is cohort-based. Without cohort literacy, you can't answer these questions properly. ## Cohorts defined A cohort is a group of customers who share a common starting event — typically signup date, purchase date, or activation event — tracked over time as a unit. - Customers who signed up in January 2024 form one cohort. - Customers who made first purchase via paid social form another cohort. - Customers who upgraded to premium tier in Q2 form a third cohort. - Cohorts are tracked over time: how do they retain, how do they expand, how do they churn? ## Types of cohorts - **Time-based.** Signup or purchase period (monthly, quarterly). - **Source-based.** Acquisition channel (paid social, organic, referral). - **Segment-based.** Customer segment (SMB, enterprise; persona; geo). - **Behavior-based.** First-week behavior; activation milestone; feature usage. - **Event-based.** Discount-acquired vs full-price; promotional event customers. ## Retention curves The standard cohort visualization. X-axis = time since cohort start; Y-axis = % retained. ### Classic patterns - **Smile curve.** Retention drops, then rises. Suggests product becoming more valuable over time for retained users. Strong signal. - **Flattening curve.** Drops then stabilizes at floor. Indicates loyal-core formation. - **Continuously declining.** No floor; no product-market fit signal. - **Step changes.** Sudden drops indicate specific event (price change, product issue). ## The smile curve and PMF The smile curve (retention rising in later periods) is a strong signal of product-market fit: - Suggests retained users find increasing value. - Often indicates network effects or habit formation. - Few products show genuine smiles; most show flattening or declining curves. - Casey Winters and others have written about smile curves as PMF indicator. ## Cohort analysis techniques - **Cohort retention tables.** Rows = cohorts; columns = periods since start; cells = retention %. - **Comparison across cohorts.** Are newer cohorts better/worse than older? - **Cohort revenue curves.** Revenue per cohort over time; LTV emerges. - **Cohort cumulative revenue.** Running total per cohort; LTV at observation point. - **Cohort triangle.** Shape revealing patterns: triangular fills with newer cohorts at right edge. - **Cohort heatmap.** Color-coded retention; patterns jump out. ## Cohort segmentation - Same time-cohort split by source: do paid social customers retain differently than organic? - Same time-cohort split by behavior: do customers who hit activation milestone retain better? - Same time-cohort split by segment: do SMB customers retain differently than enterprise? - Each segmentation reveals different actionable insight. ## Tools for cohort analysis - **Product analytics platforms.** Amplitude, Mixpanel, Heap, PostHog — built-in cohort tables. - **BI tools.** Looker, Tableau, Mode — custom cohort dashboards on warehouse data. - **SQL.** Cohort queries doable directly in SQL with window functions. - **Custom Python.** pandas, lifelines for advanced cohort analysis and survival. ## Advanced playbook - **Cohort dashboards as living tools.** Refresh automatically; team references regularly. - **Multiple cohort dimensions.** Time, source, segment, behavior simultaneously available. - **Comparative cohort analysis.** "Are newer cohorts better than 12 months ago?" as standing question. - **Cohort-based LTV.** LTV calculated by source cohort to inform channel investment. - **Cohort retention forecasting.** Project future retention based on early cohort behavior. - **Cohort revenue cohort triangles.** Visual revenue accumulation. - **Behavioral cohort triggers.** Activation milestone behavior identifies high-retention cohorts. - **Cross-cohort comparison normalization.** Adjust for cohort size and tenure when comparing. - **Cohort-based product decisions.** Features that improve early-cohort retention prioritized. - **Cohort-based marketing decisions.** Channels producing best-retaining cohorts get budget. ## Common mistakes - Aggregate metrics only; cohorts ignored. - Single-dimension cohorts; missed insights from multi-dimensional analysis. - Cohort dashboards built but never reviewed. - No source-based cohorts; channel quality differences missed. - Behavioral cohorts not built; activation patterns invisible. - Cohort comparison without normalization; misleading. - LTV calculated without cohort structure; based on averages. - Cohort decisions made on partial data (incomplete cohorts). - No retention forecasting; future LTV unguided. - Cohort patterns visible but not acted on. - Quarterly review of cohorts skipped. - Tools chosen without cohort capability. ## Operating checklist - Cohort dashboards built and refreshed - Time-based cohorts (monthly minimum) - Source-based cohorts (by channel) - Behavior-based cohorts (activation, milestone) - Segment-based cohorts (persona, tier) - Cohort retention curves visualized - Cohort revenue / LTV calculations - Quarterly cohort review - Cohort comparison normalization - Cohort patterns informing product / marketing decisions - Retention forecasting from early cohort behavior - Cohort capability included in tool selection ## Sources and further reading - Andrew Chen — cohort retention and PMF essays - Casey Winters, Reforge — retention frameworks - Brian Balfour, Reforge — retention models - Lenny Rachitsky newsletter retention coverage - Andrew Chen, "The Cold Start Problem" - Amplitude Retention research - Mixpanel cohort analysis tutorials - David Skok, For Entrepreneurs — SaaS retention - Patrick Campbell, ProfitWell — retention research - Daniel McCarthy, Theta — customer-based valuation - Sequoia Capital growth retention research - Reforge retention curriculum --- Part of the [Marketing Analytics](../index.html) series.