Cohort and Retention Analytics

Parent: data analysis · researched 2026-05-30T22:47:15.389Z· 38 sources · 11 concepts · skill da-34-cohort-retention-analytics

Retention answers the single most important growth question: do users who join

Overview

Retention answers the single most important growth question: do users who join [source]
keep coming back? Acquisition without retention is a leaky bucket - you pour [source]
users in the top and they fall out the bottom, so growth stalls no matter how [source]
much you spend. This skill covers the math and methods for measuring retention [source]
across cohorts (groups of users grouped by a shared start or behavior), [source]
reading retention curves, doing growth accounting (decomposing user and [source]
revenue change into its parts), and the SaaS revenue-retention metrics [source]
(NRR/GRR). It is the methods layer beneath product-led and SaaS growth. [source]
Scope boundary: this skill is the measurement of who stays and by how much. [source]
For funnels/activation/North-Star use da-21-product-analytics; for [source]
probabilistic CLV (BG/NBD, gamma-gamma) use da-23-customer-lifetime-value; [source]
for hazard-rate/survival modeling use da-24-survival-analysis; for [source]
experiment-driven lift use da-12-ab-testing-causal-inference. [source]

1. Acquisition vs behavioral cohorts

Acquisition (time) cohort - users grouped by when they first signed up [source]
or activated (same day/week/month). Answers when users churn and lets you [source]
compare cohort quality over time. It does not tell you why [source]
(Amplitude, Cohort Retention Analysis, 2024). [source]
Behavioral cohort - users grouped by what they did (completed [source]
onboarding, used a key feature, invited a teammate), independent of join [source]
date. Answers which behaviors correlate with retention - the input to [source]
finding your activation/aha moment [source]
(Amplitude, Guide to Behavioral Cohorting, 2024; [source]
Amplitude Docs, Behavioral Cohorts). [source]
Workflow: use acquisition cohorts to detect a retention problem, behavioral [source]
cohorts to diagnose and fix it (find the behavior that separates retained [source]
(Chameleon, Cohort Analysis 101, 2024). [source]

2. Retention-curve shapes

A retention curve plots % of a cohort still active against periods-since-start. [source]
Canonical shapes (Amplitude, Retention Curve; [source]
Churnkey, Retention Curves, 2024; [source]
Product Growth, Retention Curves Guide): [source]
Declining - slopes to zero; no group finds lasting value → no PMF. [source]
Flattening - steep early drop, then levels off at a retention floor [source]
(the long-term stable %). A flattening curve is the classic *signal of [source]
product/market fit*: a stable set of users is hooked. [source]
Smile - drops, flattens, then rises as churned users resurrect (often [source]
via network effects or re-engagement). The aspirational shape (Slack, Airbnb). [source]
Dead-on-arrival - near-vertical drop to ~0 by period 2; users tried it [source]
once and never returned. [source]
The retention floor (where the curve flattens) is your real long-term [source]
retention. Improving the floor (curve flattens higher) compounds far more [source]
than improving early-period retention that still decays to the same floor. [source]

3. Retention definitions: N-day, unbounded, bracket, rolling

Choosing the definition changes the numbers dramatically - always state which [source]
you use (Amplitude, 3 Ways to Measure Retention, 2024; [source]
Mixpanel Docs, Retention; [source]
Amplitude, N-Day Retention for Mobile Games): [source]
N-day (classic / bounded) - % of cohort active on exactly day N. Strict; [source]
best for daily-use products. Day-2 retention = 50% means 50% came back [source]
specifically on day 2. [source]
Unbounded ("rolling" in Mixpanel's loose sense) - % active on day N *or [source]
any day after*. Always ≥ N-day. Good for infrequent-use products. Note: this [source]
is not a true moving average despite the "rolling" label. [source]
Bracket / range - % active within a custom window (Day 0; Days 1–7; Days [source]
8–14). A flexible generalization of N-day; matches a product's natural cadence. [source]
Rolling retention (classic survival sense) - % active on day N or later, [source]
used to estimate a survival/lifetime curve; conceptually same as unbounded. [source]
Bounded vs unbounded asymmetry: bounded undercounts weekly/monthly-cadence [source]
products; unbounded inflates if you never re-baseline. Match the metric to the [source]
product's expected frequency. [source]

4. Retention ↔ engagement

Retention is binary (active or not in a period) and is the output; [source]
engagement depth (frequency × breadth of actions) is the leading indicator. [source]
Deeper engagement → habit → retention → sustainable growth - Reforge frames [source]
retention/engagement as "the power plant of the growth model" [source]
(Reforge, Retention is the Silent Killer; [source]
Reforge, Growth Loops are the New Funnels; [source]
Conor Dewey, Reforge Recap: Engagement + Retention). [source]
Practical move: don't just track the retained/churned flag - track how deeply [source]
retained users engage, because depth predicts long-term value and is the lever [source]
you pull to raise the retention floor. [source]

5. Growth accounting (users)

Decompose period-over-period active users into additive components. The [source]
fundamental identity (Social Capital / Jonathan Hsu, Diligence Part 1, 2017; [source]
Amplitude, Growth Accounting, 2024): [source]
new - first-ever active this period. [source]
retained - active last period AND this period (carried over). [source]
resurrected - active in some past period, inactive last period, active now. [source]
churned - active last period, inactive now (enters as a negative). [source]
User Quick Ratio (QR) = (new + resurrected) / churned. Users gained per user [source]
lost. QR > 1 means growing; rule of thumb QR ≥ ~1.5 is healthy [source]
(Hsu, Diligence Part 1, 2017; [source]
The SaaS CFO, SaaS Quick Ratio, 2024). [source]

6. Growth accounting (revenue / MRR)

Same identity applied to dollars (Social Capital / Hsu, Diligence Part 2, 2017; [source]
Lenny Rachitsky, Bottom-Up SaaS Metrics): [source]
expansion - existing customers paying more (upsell/seats). [source]
contraction - existing customers paying less (downgrade) but not zero. [source]
churned - dropped to zero. [source]
SaaS Quick Ratio = (new MRR + expansion MRR) / (churned MRR + contraction [source]
MRR). Mamoon Hamid (Social Capital) popularized a target of QR ≥ 4 for [source]
early-stage SaaS - $4 of growth for every $1 lost [source]
(The SaaS CFO, 2024; [source]
Cobloom, SaaS Quick Ratio). [source]

7. DAU/WAU/MAU and stickiness

DAU / WAU / MAU - unique active users in a 1-day / 7-day / 30-day window [source]
(Mixpanel, MAU Benchmarks, 2026; [source]
Gainsight, DAU/MAU Guide). [source]
Stickiness ratio = DAU/MAU (× 100) ≈ how many days/month an average [source]
monthly user shows up; DAU/MAU = 20% ≈ 6 days/month. Use WAU/MAU for [source]
products not meant for daily use [source]
(Statsig, Understanding DAU/MAU). [source]
Benchmarks: daily-habit/social aim for DAU/MAU > 50%; B2B SaaS averages ~30%; [source]
<20% can be fine for infrequent products [source]
(CleverTap, DAU vs MAU, 2024). [source]

8. Churn rate vs retention rate (and the asymmetry)

Single-period complements: Retention = 1 − Churn [source]
(Churnkey, Churn vs Retention; [source]
Orb, Churn vs Retention; [source]
Maxio, Retention vs Churn). [source]
The compounding asymmetry: churn compounds multiplicatively - 5%/mo churn [source]
≈ 0.95¹² ≈ 54% retained after a year, NOT 1 − (5%×12). Always state the period [source]
and never linearly annualize. [source]
Customer (logo) churn ≠ revenue churn - a small customer and a whale count the [source]
same in logo churn but very differently in revenue churn. Track both. [source]

9. Revenue / dollar retention — NRR & GRR

Cohort the revenue of a customer group and measure it a year later [source]
(Drivetrain, GRR; [source]
Orb, NRR vs GRR; [source]
SaaS Capital, Good Retention Rate, 2025): [source]
GRR measures pure leak prevention (best-in-class 90–100%). [source]
NRR measures retention plus expansion - NRR > 100% means a cohort grows [source]
in revenue with zero new logos (the "negative net churn" holy grail). [source]
Benchmarks (private B2B SaaS, ~2024): median NRR ~100–106%; enterprise (>$100K [source]
ACV) ~118%; SMB (<$25K ACV) ~97%; best-in-class NRR > 120–130%; median GRR [source]
~88–90% (Optifai, B2B NRR Benchmarks; [source]
Ordway, NRR Guide). [source]

10. Sean Ellis test & power-user curve

Sean Ellis (40%) test - survey: *"How would you feel if you could no [source]
longer use [product]?"* If ≥ 40% say "very disappointed," you likely have [source]
(FitSignal, Sean Ellis 40% Test; [source]
LearningLoop, Sean Ellis Score; [source]
StartupArchive, Sean Ellis on PMF). [source]
Power-user curve (L28 / L30) - histogram of users by active days in the [source]
month (1 of 30 … 30 of 30), coined by the Facebook growth team ("Ln" = active [source]
n of last 30). A right-skewed "smile" with a heavy right tail signals strong [source]
engagement a single DAU/MAU average hides [source]
(a16z / Andrew Chen, Power User Curve, 2018; [source]
andrewchen.com). [source]
Use together: Sean Ellis = attitudinal PMF; flattening curve + heavy-tailed [source]
power-user curve + healthy stickiness = behavioral PMF. [source]

11. Building cohort tables in SQL

Canonical three-step pattern (Holistics, Cohort Retention with SQL; [source]
Cube, Cohort Retention Recipe; [source]
O'Reilly, SQL for Data Analysis ch.4): [source]
Assign each user a cohort (their first-activity period). [source]
Compute period offset for every activity (period − cohort_period). [source]
Pivot/aggregate counts per (cohort, offset) and divide by cohort size. [source]
For zero-activity periods (gaps), build a date spine with [source]
generate_series/recursive CTE, LEFT JOIN activity, and COALESCE(...,0) [source]
so missing months render as 0 rather than vanishing [source]
(Holistics, 2024). [source]
Self-joins read more clearly for N-day retention; window functions are [source]
terser but harder to review. Pick legibility for shared analytics code. [source]

Tools / Frameworks

Amplitude / Mixpanel - built-in N-day/unbounded/bracket retention, [source]
behavioral cohorts, stickiness, power-user curves. Read the docs for the [source]
exact retention definition each uses before comparing dashboards. [source]
SQL warehouse (BigQuery / Snowflake / Postgres) - DATE_TRUNC, [source]
DATE_DIFF/DATE_PART, generate_series, window functions; the portable [source]
ground truth behind any BI tool. [source]
Reforge / Lenny's Newsletter / a16z (Andrew Chen) - growth loops, [source]
retention/engagement engine, power-user curve. [source]
Social Capital "8-ball" growth accounting - the canonical [source]
new/resurrected/churned decomposition and quick ratio. [source]
BI layer (Looker/Cube/Metabase) - cohort retention as a reusable model; [source]
Cube ships a retention recipe. [source]

Methodology

Define the active event explicitly (login? key action? value moment?). [source]
Everything downstream depends on this. [source]
Pick the retention definition (N-day vs unbounded vs bracket) to match [source]
product usage frequency. State it on every chart. [source]
Build acquisition cohorts, plot curves, find the retention floor. [source]
Segment by behavioral cohort to find the activation behavior that lifts [source]
Run growth accounting (users and MRR) to see whether growth is new-driven [source]
or retention-driven; compute the quick ratio. [source]
Layer revenue retention (NRR/GRR) for monetized products. [source]
Validate PMF with flattening curve + power-user tail + Sean Ellis test. [source]

Practical Patterns

Lead every retention chart with the **definition + active-event + cohort [source]
granularity**; otherwise numbers are uncomparable. [source]
Optimize the retention floor (curve shape), not just Day-1 - a higher [source]
Report NRR and GRR together: NRR can mask churn that expansion papers [source]
over; GRR exposes the underlying leak. [source]
Use WAU/MAU (not DAU/MAU) for weekly-cadence products so stickiness isn't [source]
Decompose growth with the 8-ball / growth-accounting view in every [source]
business review so "we grew 10%" reveals new vs resurrected vs reduced churn. [source]

Anti-Patterns

Linearly annualizing churn (5%/mo ≠ 60%/yr). Compound it. [source]
Comparing N-day to unbounded numbers as if equivalent - unbounded is [source]
Reporting only NRR and hiding gross churn behind expansion. [source]
A single DAU/MAU average masking a bimodal power-user split - show the [source]
Day-1-retention obsession while the curve still decays to the same floor. [source]
Ignoring the survivorship of recent cohorts - the newest cohort has no [source]
long-tail data yet; don't compare its Day-30 to an old cohort's before 30 days [source]

Troubleshooting

Retention "improved" suspiciously → check whether the active-event [source]
definition or the retention type (bounded↔unbounded) changed. [source]
Recent cohorts look worse → likely right-censoring, not real decline; [source]
only compare offsets that have fully matured for all cohorts shown. [source]
NRR > 100% but business feels shaky → inspect GRR and logo churn; [source]
expansion from a few whales can hide broad SMB churn. [source]
Curve never flattens → no PMF in that segment; re-segment by behavioral [source]
cohort to find a sub-population that does flatten. [source]
SQL retention has gaps/jumps → you're missing zero-activity periods; add a [source]
date spine + COALESCE. [source]
Stickiness looks terrible → wrong window; switch DAU/MAU → WAU/MAU for [source]

References

Amplitude - Cohort Retention Analysis (2024): https://amplitude.com/blog/cohorts-to-improve-your-retention [source]
Amplitude - Guide to Behavioral Cohorting (2024): https://amplitude.com/blog/guide-to-behavioral-cohorting [source]
Amplitude Docs - Behavioral Cohorts: https://amplitude.com/docs/analytics/behavioral-cohorts [source]
Amplitude - Retention Curve: https://amplitude.com/explore/analytics/retention-curve [source]
Amplitude - 3 Ways to Measure Retention (2024): https://medium.com/@amplitudeHQ/3-ways-to-measure-user-retention-2af5e4e82a45 [source]
Amplitude - N-Day Retention for Mobile Games: https://amplitude.com/blog/n-day-retention-for-mobile-games [source]
Amplitude - Growth Accounting (2024): https://amplitude.com/blog/growth-accounting [source]
Mixpanel Docs - Retention: https://docs.mixpanel.com/docs/analysis/reports/retention [source]
Mixpanel - MAU Definition & 2026 Benchmarks: https://mixpanel.com/blog/mau/ [source]
Churnkey - Retention Curves (2024): https://churnkey.co/blog/retention-curves/ [source]
Churnkey - Churn Rate vs Retention Rate: https://churnkey.co/blog/churn-rate-vs-retention-rate/ [source]
Product Growth - Retention Curves Guide: https://productgrowth.in/resources/guides/retention-curves-guide/ [source]
Chameleon - Cohort Analysis 101 (2024): https://www.chameleon.io/blog/cohort-analysis [source]
Reforge - Retention is the Silent Killer: https://www.reforge.com/blog/retention-engagement-growth-silent-killer [source]
Reforge - Growth Loops are the New Funnels: https://www.reforge.com/blog/growth-loops [source]
Conor Dewey - Reforge Recap: Engagement + Retention: https://www.conordewey.com/blog/reforge-engagement-retention [source]
Social Capital / Jonathan Hsu - Diligence Part 1: Accounting for User Growth (2017): https://medium.com/swlh/diligence-at-social-capital-part-1-accounting-for-user-growth-4a8a449fddfc [source]
Social Capital / Jonathan Hsu - Diligence Part 2: Accounting for Revenue Growth (2017): https://medium.com/swlh/diligence-at-social-capital-part-2-accounting-for-revenue-growth-551fa07dd972 [source]
Lenny Rachitsky - Most Important Bottom-Up SaaS Metrics: https://www.lennysnewsletter.com/p/the-most-important-bottom-up-saas-69d [source]
The SaaS CFO - SaaS Quick Ratio (2024): https://www.thesaascfo.com/saas-quick-ratio/ [source]
Cobloom - SaaS Quick Ratio: https://www.cobloom.com/blog/saas-quick-ratio-how-to-measure-your-startups-revenue-health [source]
Gainsight - DAU/MAU Guide: https://www.gainsight.com/essential-guide/product-management-metrics/dau-mau/ [source]
Statsig - Understanding DAU/MAU: https://www.statsig.com/perspectives/understanding-daumau-key-metrics-for-product-success [source]
CleverTap - DAU vs MAU (2024): https://clevertap.com/blog/dau-vs-mau-app-stickiness-metrics/ [source]
Orb - Churn vs Retention Rate: https://www.withorb.com/blog/churn-rate-vs-retention-rate [source]
Orb - NRR vs GRR: https://www.withorb.com/blog/nrr-vs-grr [source]
Maxio - Retention vs Churn: https://www.maxio.com/saaspedia/retention-rate-vs-churn-rate [source]
Drivetrain - Gross Revenue Retention: https://www.drivetrain.ai/strategic-finance-glossary/what-is-gross-revenue-retention-formula-benchmarks [source]
SaaS Capital - Good Retention Rate (2025): https://www.saas-capital.com/blog-posts/what-is-a-good-retention-rate-for-a-private-saas-company/ [source]
Optifai - B2B SaaS NRR Benchmarks: https://optif.ai/learn/questions/b2b-saas-net-revenue-retention-benchmark/ [source]
Ordway - NRR Guide: https://ordwaylabs.com/resources/guides/net-revenue-retention-guide/ [source]
FitSignal - Sean Ellis 40% Test: https://www.fitsignal.com/blog/sean-ellis-40-percent-test [source]
LearningLoop - Sean Ellis Score: https://learningloop.io/glossary/sean-ellis-score [source]
StartupArchive - Sean Ellis on PMF: https://www.startuparchive.org/p/sean-ellis-on-how-to-tell-if-you-have-product-market-fit [source]
a16z / Andrew Chen - The Power User Curve (2018): https://a16z.com/the-power-user-curve-the-best-way-to-understand-your-most-engaged-users/ [source]
andrewchen.com - The Power User Curve: https://andrewchen.com/power-user-curve/ [source]
Holistics - Calculate Cohort Retention with SQL (2024): https://www.holistics.io/blog/calculate-cohort-retention-analysis-with-sql/ [source]
Cube - Cohort Retention Recipe: https://cube.dev/docs/product/data-modeling/recipes/cohort-retention [source]
O'Reilly - SQL for Data Analysis, ch.4 Cohort Analysis: https://www.oreilly.com/library/view/sql-for-data/9781492088776/ch04.html [source]

Children

Acquisition vs behavioral cohorts (frontier)
Retention curve shapes (smile/flattening/dead-on-arrival) (frontier)
N-day vs unbounded vs bracket vs rolling retention (frontier)
Retention and engagement relationship (frontier)
Growth accounting (users) (frontier)
Growth accounting (revenue/MRR) and quick ratio (frontier)
DAU/WAU/MAU and stickiness ratio (frontier)
Churn rate vs retention rate asymmetry (frontier)
Net and gross revenue retention (NRR/GRR) (frontier)
Sean Ellis test and power-user curve (L28/L30) (frontier)
Building cohort retention tables in SQL (frontier)

Frontier under this node: Acquisition vs behavioral cohorts, Building cohort retention tables in SQL, Churn rate vs retention rate asymmetry, DAU/WAU/MAU and stickiness ratio, Growth accounting (revenue/MRR) and quick ratio, Growth accounting (users), N-day vs unbounded vs bracket vs rolling retention, Net and gross revenue retention (NRR/GRR), Retention and engagement relationship, Retention curve shapes (smile/flattening/dead-on-arrival), Sean Ellis test and power-user curve (L28/L30)

← the whole tree · 3D view· how to read this page