MetricRaven

Analysis

What is cohort analysis?

Cohort analysis groups customers who share a starting point, such as the month they signed up or the channel that brought them, and tracks each group's behavior over time. It shows whether retention and revenue are improving, which an overall average hides.

Updated

What cohort analysis is

A cohort is a group of people who started at the same time or in the same way. Cohort analysis follows each group separately, month by month, instead of mixing everyone into one number.

Averages lie when your customer base is changing. Say your overall churn rate holds at 5% a month. That could mean nothing changed. Or it could mean your old customers have become very loyal while every new month's sign-ups leave faster and faster. A cohort table shows which one it is.

  • Acquisition cohorts group people by when they started: "everyone who subscribed in January". They answer "are newer customers staying longer?"
  • Behavioral or segment cohorts group people by what they did or where they came from: customers who arrived from search, from ads, or who used a feature in their first week. They answer "which kind of customer is worth more?"

Retention formula

Retention in month N = customers from the cohort still active in month N ÷ customers who started in the cohort

Revenue retention uses money instead of people: the cohort's revenue in month N divided by its revenue in month 0. With upgrades it can go above 100%, which is the idea behind net revenue retention.

Worked example: a monthly retention table

Each row is the month customers first paid. Each column is months since then. Each cell is the share of that cohort still paying.

CohortCustomersMonth 0Month 1Month 2Month 3
January200100%78%66%60%
February250100%80%70%64%
March180100%85%76%—
April300100%71%——

Read it three ways:

  • Across a row: how one cohort decays. January lost 22% after the first month, then 12 points, then 6. Most churn happens early, and the curve flattens.
  • Down a column: whether you are improving. Month-1 retention went 78% → 80% → 85%, so something in February or March worked, perhaps better onboarding.
  • The odd row out: April dropped to 71%. April also had the most customers. If a big ad push ran in April, those customers were a worse fit, and the campaign's ROAS on the first payment overstated its value.

Now the money. If each customer pays $30 a month, the January cohort's first four months brought 200 × $30 × (1 + 0.78 + 0.66 + 0.60) = $18,240, or $91.20 per customer. That per-customer figure, followed further out, is how you estimate customer lifetime value by cohort.

What cohort analysis is good for

  • Judging channels on retention, not just acquisition. Customers from search may stay twice as long as customers from a discount campaign. Compare cohorts by first channel.
  • Checking product changes. If a new onboarding flow shipped on March 1, the March cohort should retain better than February's.
  • Forecasting revenue. Stable cohort curves let you project how much this month's sign-ups will be worth in a year.
  • Setting acquisition budgets. A cohort's cumulative revenue per customer tells you how much customer acquisition cost you can afford, and how many months it takes to earn it back.

How to run a cohort analysis

  1. Pick the starting event. First payment for revenue retention; first visit or sign-up for product retention.
  2. Pick the grouping. Calendar month is standard. Use weeks for fast-moving consumer products, quarters for slow B2B sales.
  3. Define "active". Still paying, logged in, or made a repeat purchase. Write it down and keep it.
  4. Build the table. One row per cohort, one column per period since start, each cell the share still active.
  5. Compare like with like. The newest cohorts have only a few columns; don't compare their month 1 with an old cohort's month 12.

Common mistakes

  • Cohorts that are too small. A cohort of 12 customers swings 8 points when one person leaves. Group by quarter or by channel until each cohort has enough people.
  • Mixing calendar months with months since start. "Retention in June" and "retention in month 3" are different questions.
  • Counting trials as customers. Free trials that never paid make month-0 retention look worse than it is. Start cohorts at the first payment.
  • Ignoring reactivations. A customer who leaves in month 2 and returns in month 5 can make retention go up. Decide whether that counts.
  • Reading cohorts as cause. A better cohort coincides with a change; it doesn't prove the change worked if other things changed too.

How MetricRaven handles cohort analysis

MetricRaven doesn't draw a retention cohort grid. It gives you the pieces to compare acquisition cohorts by where customers came from and when they arrived, with revenue from your payment provider:

  • Channels compared under Insights → Details → Path to purchase groups visitors by the channel of their first-ever pageview and shows purchase rate, revenue, median days and visits to buy, how many came back, and how many of those paid (buyer insights).
  • New vs returning visitors: new visitors have no tracked pageview before the period, returning visitors do. Chart both on the traffic chart (filters and segments).
  • Revenue by first touch: every revenue figure credits the customer's first tracked visit, so filtering by channel, campaign or landing page and a date range gives that acquisition cohort's revenue, renewals included.
  • MRR movements on the Revenue tab split monthly changes into new, expansion, contraction and churned MRR (Revenue dashboard).
  • Lifetime value shows the average, median and per-month revenue per customer, by plan.

To build a classic grid, export the breakdowns you need as CSV and pivot them in a spreadsheet, or pull revenue and customers by period from the Reports API.

Questions

What is an example of cohort analysis?
Group customers by the month they first paid, then track the share still paying 1, 2 and 3 months later. If January keeps 60% after three months and March keeps 70%, newer customers are staying longer.
What is the difference between cohort analysis and segmentation?
Segmentation splits users by a trait at one point in time. Cohort analysis follows a group from a shared starting point over time, so you can see how its behavior changes month by month.
How many customers should a cohort have?
Enough that one person leaving doesn't swing the numbers. With fewer than about 50 people per cohort, a single cancellation moves retention by 2 points or more, so group by quarter or combine channels.
Does MetricRaven have cohort analysis?
Not as a retention grid. It compares acquisition cohorts by first channel in Insights → Details → Path to purchase, shows new vs returning visitors and MRR movements, and credits revenue to each customer's first visit, which you can filter by channel and date.
Full glossary

Compare customers by the channel that found them

MetricRaven credits every payment and renewal to the visit that first brought the customer, and compares channels on purchase rate, revenue and time to buy.

14-day free trial, no card required.