Retention Cohort Analysis

A cohort retention chart displays retention data in a triangular matrix. Rows represent acquisition cohorts (January cohort, February cohort, etc.); columns represent time periods since acquisition (Week 1, Week 4, Month 3, Month 6, Month 12). Each cell shows the percentage of the original cohort that was active in that time period. Reading patterns: Diagonal comparison: comparing March 2024 Month-3 retention to March 2025 Month-3 retention shows whether the product retains customers better now than a year ago — controlling for cohort age. Row shape: the shape of a cohort's retention curve (how steeply it drops in the first 4 weeks, and at what percentage it flattens) reveals whether the product has a "retained core" of engaged users or continuously declines. A retention curve that drops to 5% in month 3 and stays there is radically different from one that drops to 35% in month 3 and holds — the former doesn't have a sustainable engagement model; the latter has found a healthy retained core. Column comparison: comparing all cohorts at month 6 identifies trends in 6-month retention over time — improving month 6 cohort retention signals that product improvements are working.

Behavioral cohort analysis groups users not by acquisition date but by a specific action they took (or didn't take) in the product, then compares the retention curves of users who performed the action vs. those who didn't. Classic methodology (popularized by Facebook's "10 friends in 7 days" discovery): define a cohort of users who performed Action X in week 1 and compare their 90-day retention to users who didn't perform Action X in week 1. If the Action X cohort retains at 55% vs. 22% for non-Action X users at 90 days, Action X is a strong retention predictor. Important caution: this analysis identifies correlation, not causation. Users who performed Action X may have been inherently more motivated or better-fit customers who would have retained regardless. To establish causation, run an A/B test that nudges a random subset of users to complete Action X and compares their retention to an un-nudged control group. If the nudged group retains better, the action has a causal effect on retention. Product Ops facilitates the behavioral cohort analysis as a standard component of the product analytics quarterly review, producing the candidate retention-predicting actions for A/B test validation.

B2B SaaS cohort analysis has five important differences from B2C consumer apps. Unit of analysis: B2B retention is measured at the account level (is the company still a customer?), not the user level (individual users leaving a company still counts as the company retaining). Survival analysis: B2B SaaS uses daily or monthly active usage as the retention measure (not app open rate) — an account that logs in at least once per month is "retained" even if individual usage varies. Contract alignment: annual B2B contracts create artificial retention plateaus — accounts don't churn mid-contract, creating a spike in churn events at contract anniversary dates. Cohort retention curves in B2B show an annual sawtooth pattern rather than the smooth curves typical in B2C. Cohort size: B2B cohorts are much smaller (a healthy enterprise SaaS may acquire 25–50 new enterprise accounts per month vs. thousands of consumer users) — requiring longer time windows for statistically meaningful analysis. Expansion effect: B2B cohort analysis should track not just survival rate but ARR per cohort over time — a cohort where 85% of accounts survive but ARR has grown 150% (through expansion) is dramatically different from a cohort where 98% survive but ARR is flat.