What is cohort retention analysis?

Cohort retention groups customers by the month (or week) they were acquired and tracks what percentage remain active in each subsequent month. The result is a triangular heat-grid: rows are acquisition cohorts, columns are months-since-acquisition. Mixpanel and Amplitude popularised the visual; the math goes back to actuarial science.

What is a cohort cliff?

A cohort cliff is the month (or week) where retention drops sharply between two consecutive periods - often M0 → M1 (post-trial), or M3 → M4 (box-2 cancellation in DTC). The tool detects cliffs automatically by flagging period-over-period drops of more than 15 percentage points.

What is the difference between cohort retention and overall churn rate?

Overall churn averages across all customers active in a month. Cohort retention isolates customers acquired in a specific month, so you see how product changes affect new acquisitions vs legacy customers. If overall churn looks flat but cohort M1 is improving, the product is getting better even though aggregate metrics hide it.

What is the Mixpanel/Amplitude cohort heat-grid convention?

Rows = acquisition month, columns = months-since-acquisition (M0, M1, M2...). Cell value = retention percentage. The diagonal moves through time. Both Mixpanel (Cohort Analysis view, since 2014) and Amplitude (Retention chart, since 2015) use the same layout, popularised in product analytics by Sean Ellis and the growth community.

What is a good M1 retention for SaaS?

OpenView 2024: B2B SaaS median M1 retention 85-92%, top-quartile 92%+. DTC subscription: 60-70%. Consumer apps: 35-45%. The wider gap between B2B and consumer reflects activation difficulty and switching cost.

What is a flattening curve vs a smile curve?

A flattening retention curve shows steady decline that levels off (typical B2B SaaS). A smile curve dips early then rises back up - this is net retention including expansion (upsell, cross-sell, seat growth). Bessemer best-in-class enterprise SaaS shows smile curves with NRR over 130% by M24.

How does the tool detect the cohort cliff?

For each cohort row, the tool computes the period-over-period drop (M_n - M_n+1). The largest single drop is flagged as the cohort cliff. If the drop exceeds 15 percentage points, the tool highlights it in the cell with a red border.

How is N-cohort average retention calculated?

For each month-since-acquisition column, the tool averages retention across all cohort rows that have data for that month. The result is a single 'representative cohort' curve, useful when individual cohorts are noisy or sparse.

What does a heat-grid colour mean?

The tool uses a graduated scale: dark green for 90%+ retention, green for 75-90%, lime for 60-75%, amber for 40-60%, orange for 25-40%, red for under 25%. Eyeballing the colour gradient across columns shows where in the customer journey retention collapses.

How many cohorts should I compare?

Practitioner convention: 8-12 monthly cohorts (one year of acquisition data). Fewer than 6 makes seasonal comparisons unreliable; more than 12 is hard to visually scan. The tool ships with 8 by default, expandable to 12.

What is the difference between user-week and revenue-week cohorts?

User-week cohorts count active users; revenue-week cohorts count surviving revenue. They diverge sharply for products with expansion: a cohort can lose 30% of users but retain 110% of original revenue (Bessemer NRR pattern). The tool defaults to user retention; multiply by expansion factor for revenue.

How do I diagnose a cohort cliff?

Five Amplitude/Mixpanel-recommended diagnostic steps: (1) check activation events completed in M0; (2) check time-to-first-value; (3) check NPS / CSAT at the cliff month; (4) check feature adoption by cliff-month; (5) survey churned users for cancellation reason. Most cliffs trace back to a missing activation event.

Cohort Analytics

Cohort Retention Heat-Grid

Months-acquired × months-since-acquisition heat-grid that visualises customer retention across cohorts and auto-detects the cohort cliff (the month where retention collapses). Methodology mirrors Mixpanel Cohort Analysis (2014) and Amplitude Retention (2015), with benchmark presets from Bessemer Cloud Index, OpenView, Recurly, Amplitude Product Benchmarks, and a16z.

Methodology

Mixpanel + Amplitude

Cliff threshold

15pp drop

Industry presets

5 verticals

Default horizon

12 months

Quick Conversion

FROM (Monthly retention %)

TO (Annual retention %)36.77

Formula: Annual = Monthly^12

Retention pattern presets

Bessemer Cloud Index Q4 2025: best-in-class B2B SaaS shows a flattening retention curve after month 3 (the 'smile' curve when expansion is included), netting >100% NRR by month 12.

Inputs

Cohort size

Cohorts

Horizon (mo)

Retention curve (proportion at each month)

M0100%

M192%

M288%

M385%

M483%

M582%

M681%

M780%

M879%

M979%

M1078%

M1178%

Live cliffs detected: 0

Cohort heat-grid

rows = acquisition cohort • cols = months since

Cohort	M0	M1	M2	M3	M4	M5	M6	M7
C-07	100	96	92	87	81	78	78	80
C-06	105	94	86	81	80	82	84
C-05	99	88	84	84	86	86
C-04	95	91	91	89	86
C-03	102	97	91	85
C-02	104	92	85
C-01	97	87
C-00	97

Retention %:<1010-2525-4040-6060-7575-9090+

Formula Card

Retention(M_n) = active_users_in_M_n / cohort_size_at_M_0Cliff(M_n) = (Retention(M_n-1) - Retention(M_n)) ≥ 15ppAnnual retention = (monthly_retention)^12Column average = mean(Retention(M_n) across cohorts with data for M_n)NRR (with expansion) = (Retention(M_12) × ARPU_M12) / (1 × ARPU_M0)

Worked: cohort of 100 users, retention vector [100, 92, 88, 85, 83, 82, 81, 80, 79, 79, 78, 78]. M1 retention 92%, M12 78%. No cliff exceeds 15pp - Bessemer best-in-class B2B SaaS shape.

Saved snapshots

No snapshots yet.

How to read the heat-grid

Each row is a cohort acquired in a specific month. Top row = most recent cohort.
Each column is months-since-acquisition (M0, M1, ...). M0 is always 100% by definition.
Colour scales from dark green (90%+) to dark red (under 10%). Eyeball the gradient across columns.
Red-outlined cells are cohort cliffs (period-over-period drop more than 15pp). Diagnose by checking what happened that month in that cohort's product experience.
Save snapshot weekly or monthly; compare M3 column over time to see whether product changes are moving the retention needle.

A Brief History of Cohort Retention

In 2026, a product manager at a Series B SaaS opens her Monday cohort review and immediately spots a 15pp drop in M1 retention for the September cohort versus August. That single diagonal cell on the heat-grid triggers a sprint planning session. Cohort retention analysis as a discipline was pulled into mainstream tech from actuarial science (life-table analysis) by Sean Ellis (growth hacking, 2010), Andrew Chen, and Brian Balfour, then operationalised by Mixpanel (2014) and Amplitude (2015).

Mixpanel's Cohort Analysis view (launched 2014) and Amplitude's Retention chart (launched 2015) standardised the heat-grid format used in this tool: acquisition cohort on rows, months-since on columns, retention percentage in each cell with colour gradient. The cohort-of-cohorts comparison let product teams see whether changes shipped in week N improved retention versus week N-1, week N-4, week N-52.

The Bessemer Cloud Index has tracked SaaS retention quarterly since 2014. Best-in-class B2B SaaS shows a flattening curve (steady decline that levels off by M3-M6) or a smile curve (early dip recovers via expansion, netting more than 100% NRR by M12). The 'smile' is the holy grail and the basis of Bessemer's NRR-over-130% best-in-class threshold.

OpenView 2024 SaaS Benchmarks add SMB-specific data: median M1 retention for self-serve SMB SaaS is 75-82%, with a sharp M0 → M1 cliff caused by post-trial dropout. Top-quartile companies close the cliff to under 12pp through PLG activation experiments (Sean Ellis 40% rule: 40% of users say they would be 'very disappointed' without the product = strong product-market fit).

On the consumer side, Amplitude's 2024 Product Benchmarks report covers thousands of apps: M1 consumer-app retention median 38-45%, M3 around 22-28%, M12 plateau 5-8%. The dispersion is enormous - top-decile apps hold M3 over 50% (Duolingo, Strava-class engagement), bottom-decile cliff to single digits. The heat-grid here lets you stress-test where your cohort sits.

DTC subscriptions have their own cliff pattern. Recharge's 2025 data (Stuart Brand, Recharge CMO) shows the canonical 'box 2-3' cancellation cliff: M0 retention 100%, M1 around 65-70%, M2 50%, M3 around 40%, then long flat tail of loyalists. The Recharge preset in this tool encodes that shape. Optimisation tools: pre-shipment dunning, second-box discount, content/community in M2-M3.

On the marketplace side, a16z Marketplace 100 (2024 update by Andrew Chen, Olivia Moore) shows asymmetric retention: supply-side typically 40-55% steady state at M12, demand-side often higher post-liquidity. The asymmetry drives marketplace investment priorities — fix the worse side first. The tool here lets you compare two cohort tables side by side.

Cohort Retention Heat-Grid — FAQ

Have more questions? Contact us

Trusted by product analysts and growth teams

4.9

Based on 5,290 reviews

“Our M0 → M1 drop was 58%. The cliff detector flagged it instantly. Built a 7-day onboarding nudge series, cut the cliff to 38% in a quarter. Stakeholders finally understood what 'cohort' meant.”

Wren Sutherland

Product Analyst, Consumer App

April 20, 2026

“Used the Bessemer preset as a benchmark for board-prep. Our M3 retention was 7pp behind best-in-class - good talking point for the activation roadmap pitch.”

Anish Kapadia

Head of Growth, B2B SaaS

March 30, 2026

“Box 2 cancellation was killing us. The heat-grid showed a 30% M2 drop. We added a $5 'second-box' incentive; pushed M2 retention from 52% to 71%. Tool was the diagnostic.”

Maria Solano

DTC Founder

April 5, 2026

“Comparing supply-side and demand-side cohorts in the same grid was the unlock. Demand retention flattened nicely; supply had a brutal M2 cliff. Resourced the supply experience team accordingly.”

Hiroshi Tanaka

Marketplace Operator

April 25, 2026

Love using our calculator?

Related business-metrics tools

Churn Predictor

5-axis at-risk cohort radar.

CAC Payback

Clock dial + channel breakdown.

Gross Margin

Contribution donut + COGS sensitivity.

All Business Metrics

Index of every founder calculator.

Learn More

Dive deeper with our expert guides and tutorials related to Cohort Retention Heat-Grid

Loading articles...