Cohort analysis tools and cohort analysis software for SaaS retention

A cohort chart is a triangle. Each row is one cohort, defined by when those users started. Each column is time since that start, so column zero is the starting period and column four is four weeks or months later. Each cell is the share of that cohort still active in that period, which is why the triangle narrows: recent cohorts have not lived long enough to fill the later columns.

Read it three ways. Across a row shows how one cohort decays as it ages. Down a column compares cohorts at the same age, which is the honest test of whether your changes worked, since it holds age constant. Diagonally follows a calendar date across every cohort at once, which is how you spot an outage, a pricing change or a seasonal effect that hit everyone on the same day.

The shape matters more than the number. A healthy cohort curve drops steeply at first, then flattens into a horizontal line, and that flat section is the share of users who found durable value. A curve that keeps sliding toward zero with no flattening means nobody is sticking, and no amount of top of funnel work will fix it.

Absolute benchmarks depend heavily on the model. B2B SaaS sold annually looks nothing like a consumer app measured on weekly actives, and comparing yourself to a published median usually misleads. The comparison that pays is internal: this month's cohort against the same cohort age three months ago. If the newer line sits above the older one at week four, whatever you shipped worked.

A funnel measures a sequence inside one session or one short window: how many people who started onboarding finished it. Cohort analysis measures the same population over a long period: how many of the people who finished onboarding in March were still here in June. Funnels answer where users drop out of a flow, cohorts answer whether they came back at all.

They work best together and in that order. Use a funnel to find the step losing people, ship the fix, then watch the cohorts that signed up afterward to confirm the fix actually held past week one. Plenty of onboarding changes lift funnel completion and do nothing to retention, which is the exact failure a funnel alone cannot show you.

Start by defining the return action honestly. Retention means someone came back and did the thing your product is for, not that a background tab pinged your servers. For most B2B products that is a core workflow action, not a login. Then pick the cohort key, usually signup month for a first pass, and the period length, weekly for self-serve and monthly for annual contracts.

Build the triangle, then read down the columns to find the cohort that broke ranks. From there the work is diagnostic: what changed for that group, what did they have in common, and what did they tell you on the way out. That last question is where most teams stall, because the answer sits in support tickets and cancellation replies that never made it into the analytics tool.

For flexible behavioral cohorts at a mid to large SaaS company, Amplitude and Mixpanel are the two to shortlist, and the choice usually comes down to pricing model and which interface your team prefers. If you need to ask cohort questions about events nobody instrumented in advance, Heap's retroactive autocapture is the one genuine differentiator in the category. Engineering-led teams often land on PostHog because analytics, replay and flags arrive on one bill.

If your team already has a cohort chart and still argues about what caused the dip, adding a second charting tool will not settle it. That is the gap UserInsight fills, by attaching the feedback, tickets and reviews from the same users to the cohort that produced them.