Cohort Analysis

What is a cohort analysis?

In a cohort analysis, customers are grouped by the time at which they arrived, and each group is tracked separately over time. Instead of asking “How many customers do we have?”, the question is “How are the customers who arrived in March doing in their twelfth month?”.

Why the average is not enough

A company with a strongly growing number of new customers automatically has many young customers in its base. Young customers cancel less often because they have only just bought. The overall churn rate therefore looks good even though nothing has improved. As soon as growth slows, the figure tips. Cohort analysis makes this effect visible before it becomes a problem.

Example

A provider wins 100 customers every month.

• January cohort: after 6 months 71 customers remain, after 12 months 62.
• July cohort: after 6 months 81 customers remain.

In the same month of life, the July group is ten points above the January group. Something between January and July worked, be it a better onboarding, a narrower target group or a higher price that keeps the wrong customers away. This is exactly the question that can now be investigated. From the overall average it would never have become visible.

What the shape of the curve reveals

What matters is not the starting value but whether the curve flattens. A retention curve that turns into a horizontal line after six months means: whoever survives the first months stays for the long term. A curve that keeps falling steadily means that there is no lasting customer base, no matter how good the initial figures look.

The second view: revenue instead of heads

The same calculation with revenue instead of the number of customers is even more telling. If the January group has only 62% of its customers after 24 months but 115% of the original revenue, then the remaining customers grow by more than the departed ones cost. This is the basis for a higher valuation multiple.

Common mistakes

1. Periods too short. Six months is not enough to recognise the shape of the curve. At least 18 months, better 24.
2. Cohorts cut too small. With 8 customers a month, every movement is chance. Better to group by quarter in that case.
3. Mixing segments. Customers from paid advertising and from referrals behave differently. In a combined cohort, the effects cancel each other out.
4. Showing only the good cohorts. In due diligence the complete table is requested. Gaps look worse than weak values.

Frequently asked questions

By which characteristic are customers grouped?

The standard is the month of joining. Groupings by channel, tariff and segment are also useful, because this is where the levers are found.

How does this differ from the churn rate?

The churn rate is a snapshot across all customers. Cohort analysis is the film over the lifetime of a group. Both figures can move in opposite directions without either of them being wrong.

Is a tool needed for this?

No. Up to about 2,000 customers, a spreadsheet with contract start, contract end and monthly revenue per customer is enough. More important than the tool is that these three fields are maintained cleanly.

The Roemer Capital view

The cohort table is one of the first documents requested in due diligence, and one of the few that cannot be built retrospectively if the data were not maintained. We therefore set it up in the data room from the start, not only when an investor asks for it.