Advanced Analytics

Cohort Analysis

Aggregate metrics often look clean while the underlying business is changing for better or worse. NRR can look stable even when every new cohort is retaining worse than the one before it. Cohort analysis matters because it separates those realities instead of averaging them into one number.

Language: Read this article in Russian

COH

Cohort Analysis: why aggregate metrics hide the truth and cohorts expose it

Definition and base logic

A cohort is a group of customers that share a common starting point, most often the month of first payment. Cohort analysis tracks what happens to that group over time as it ages.

Cohort Analysis = Tracking a defined customer group across months of life

This is fundamentally different from aggregate metrics. Aggregate churn tells you how many customers left the active base. Cohort analysis tells you which generation of customers is behaving better or worse, and at what point in the lifecycle the problem begins.

That is why cohorts are diagnostic rather than merely descriptive. They reveal whether the product, onboarding, targeting, pricing, or expansion engine is actually improving over time.

How cohort analysis is structured

Three useful cohort types

  • Time cohorts: customers grouped by first payment month or quarter.
  • Behavior cohorts: customers grouped by whether they completed a key action, such as connecting an integration.
  • Attribute cohorts: customers grouped by channel, plan, geography, or company size.

Time cohorts answer whether the product is improving. Behavior cohorts help identify the real activation event. Attribute cohorts reveal which segments and channels create better long-term economics.

Logo Retention and Revenue Retention are different analyses

Logo Retention(month t) = Active customers in month t / Customers in Month 0 × 100
Revenue Retention(month t) = Cohort MRR in month t / Cohort MRR in Month 0 × 100

Logo Retention answers how many customers remain. Revenue Retention answers how much value remains, including expansion. A mature SaaS business can lose logos while revenue retention still rises above 100% because the retained customers expand.

How to build the cohort table correctly

  • Define an active customer consistently, usually as a paid customer with MRR above zero at month-end.
  • Start the cohort at first payment, not signup, to avoid trial distortion.
  • Use rows for cohort month and columns for months of life: Month 0, Month 1, Month 2, and so on.
  • Always divide retention by the original cohort size or original cohort MRR in Month 0.

This sounds basic, but small definition errors make the whole analysis misleading.

The shape of the curve is as important as the value

A healthy retention curve usually drops early and then flattens. A continuously declining line means the product is not becoming sticky. A dramatic Month 1 drop with a stable plateau usually signals an onboarding or expectations problem rather than a deep product problem.

Revenue retention can also form a J-curve, falling early and later climbing above 100% as expansion begins to dominate churn.

Why heatmaps help

Cohort heatmaps make patterns visible instantly. Improving diagonals show better onboarding or product quality in newer cohorts. Darkening later columns show stronger long-term retention or expansion. A deteriorating diagonal is an early warning that aggregate metrics may hide.

Related metrics and strategic interpretation

Cohort LTV is more realistic than formula LTV

Cohort LTV(month T) = Σ(Cohort MRR × Gross Margin from Month 0 to Month T) / Customers in Month 0

Cohort LTV is based on real retention and expansion behavior rather than on the strong assumption that churn is constant forever. That makes it much more useful for serious unit-economics decisions.

Cohort half-life

Half-life = First month where Logo Retention falls to 50% or below

Half-life is a compact way to compare cohort durability. It should usually increase over time if the product and targeting are improving.

Area Under the Curve

AUC(12 months) = Retention M1 + Retention M2 + ... + Retention M12

AUC approximates how many customer-months of life the cohort produced during the first year. It is a strong summary metric when you want one number without losing the lifecycle perspective entirely.

Behavior cohorts reveal the real activation event

One of the highest-value uses of cohort analysis is splitting customers by whether they completed a likely activation action in the first few days. If customers who connect an integration in week one retain dramatically better at Month 3 or Month 6, that action is probably close to the real aha moment.

Channel cohorts change how CAC should be read

Organic, referral, paid search, outbound, and event cohorts often have very different long-term retention curves. That means CAC alone is not enough. Channel LTV and cohort retention must be read together before scaling or cutting a channel.

Cohort NRR often follows a J-curve

Many SaaS cohorts start below 100% because of early churn, then approach 100%, and later move above 100% once expansion starts to dominate. The speed at which cohorts cross 100% is a strong signal of expansion maturity.

Common cohort analysis mistakes

  • Starting the cohort at signup instead of first payment. Trials distort Month 1 retention badly.
  • Dividing by the previous month instead of Month 0. That turns cohort retention into ordinary monthly retention.
  • Comparing cohorts at different ages. Month 6 for one cohort is not comparable to Month 12 for another.
  • Ignoring cohort size. Small cohorts are noisy and can create false narratives.
  • Ignoring seasonality. Compare January with January when possible, not automatically with December.
  • Counting reactivation as uninterrupted retention. Reactivated customers should be treated separately.
  • Looking only at logo retention. Revenue retention can tell a very different story about business quality.

Worked example and diagnosis

Mid-market SaaS cohort example:

  • January cohort M1 Logo Retention = 89%
  • April cohort M1 Logo Retention = 83%
  • July cohort M1 Logo Retention = 72%
  • January cohort M3 Revenue Retention = 91%
  • May cohort M3 Revenue Retention = 78%

The pattern is not random noise. Newer cohorts are deteriorating consistently, and revenue retention is falling alongside logo retention. That points to a real acquisition-quality or onboarding problem rather than a temporary reporting artifact.

Diagnosis: newer customers are worse-fit or are reaching value more slowly.

Main suspects: weaker paid channels, expanded ICP, degraded onboarding, or stronger competitor pressure.

Best next step: break the damaged cohorts by channel, segment, and activation behavior before changing budget or product priorities.

How Dnoise handles cohort analysis

Dnoise builds time cohorts automatically from Stripe first-successful-payment data. Trial-only customers are excluded by default, and both Logo Retention and Revenue Retention tables are generated in parallel.

The platform also supports behavioral cohorts when activation events are sent through API or automation integrations. Heatmaps, inter-cohort improvement, and cohort-based LTV make it easier to see whether changes are driven by onboarding, retention, channel mix, or expansion quality.

See cohort behavior in the demo

Why serious retention work starts with cohorts

Dnoise helps teams see whether newer customers are actually getting stronger or whether aggregate SaaS metrics are hiding a slow decline.

Why serious retention work starts with cohorts

Dnoise helps teams see whether newer customers are actually getting stronger or whether aggregate SaaS metrics are hiding a slow decline.