Cohort analysis

A single blended average hides what is happening. It mixes customers who joined under different product versions, prices and channels. A business can show steady overall retention while every new cohort is worse than the last, because a large base of loyal old customers props the average up. Cohorts show that slide early.

They also let a change be judged fairly. Compare the group that joined after an onboarding redesign with the group before it. If the newer curve sits higher, the redesign helped. If not, it did not, whatever the monthly total suggests. A cohort's curve also helps estimate Customer Lifetime, which shapes how much can be spent acquiring a similar customer.

• Start with monthly signup cohorts. Weekly cohorts are usually too small to read.
• Pick one metric, such as the share still paying, and keep it constant across cohorts.
• Track revenue per cohort next to customer count, since a group can keep most customers and still lose revenue through downgrades.
• When a cohort underperforms, ask what was different that month: price, product, channel or season.
• Test a fix on one group and credit it only if that group's curve moves.

A cohort is a group of customers who began at the same time or in the same way, such as everyone who signed up in March. Cohort analysis follows each group separately, month after month, and puts the groups side by side. The usual output is a table or chart with cohorts down the side and months since signup across the top.

• Reading a young cohort against an old one at different ages. Compare at the same month since signup.
• Cutting cohorts so thin that a handful of customers swings the result.