Cohort retention analysis for Indian ecommerce brands

Ask most brands how retention is doing and you get a single number: the share of customers who bought more than once. It is the most quoted metric in ecommerce and one of the least informative.

The problem is not that it is wrong. It is that it moves for reasons that have nothing to do with retention.

Why the blended number misleads

Your repeat rate mixes together customers who joined two years ago and customers who joined last week. The first group has had many opportunities to return. The second has had almost none.

Now grow. Each month brings a larger group of brand new customers with, by definition, no repeat purchases yet. They enter the denominator immediately and the numerator only later.

The blended rate falls. Not because retention worsened, but because you acquired successfully. Brands regularly respond to this by cutting acquisition spend, which fixes the metric and damages the business.

The reverse is equally misleading. Stop acquiring and your repeat rate climbs beautifully as the existing base matures. That looks like improvement and is the opposite.

What a cohort actually is

Group customers by the month of their first purchase. That group is now fixed and nobody joins or leaves it.

Then track it forward: of the people who first bought in March, what share bought again in their first month, their second, their third. Do the same for April, May, June.

Because the group never changes, the curve reflects genuine behaviour rather than the mix of who happens to be in your customer base this week.

Now the useful comparison becomes available: does the June cohort behave better at month three than the March cohort did at month three. That single comparison tells you whether the business is improving, which the blended number cannot.

Read the shape, not the level

Every cohort curve falls steeply at the start. Most first time buyers do not return, in every category, everywhere. That is not a problem to be solved so much as a fact to be planned around.

What matters is what happens after the drop. If the curve flattens, you have a core of customers who have made you part of their routine. That base persists and each new cohort adds to it, which is what compounding growth actually looks like from the inside.

If the curve keeps sliding toward zero, you do not have a retained base. Every rupee of revenue must be bought again, and growth is entirely a function of how much acquisition you can afford.

Two brands can show the same blended repeat rate while one has a flattening curve and the other does not. They are not remotely the same business.

Split the cohorts or miss the point

A single blended cohort curve is a large improvement on a single blended number, and it still averages away the two most actionable facts.

Split by acquisition channel. Customers won through a heavy discount frequently retain very differently from those who arrived organically or through a recommendation. If they do, your acquisition cost comparison across channels is wrong, because you were comparing the cost of a customer without accounting for how long they stay.

Split by first product. In most catalogues, which item somebody buys first strongly predicts whether they return. If one product produces markedly better retention, that is not a merchandising curiosity, it is where your acquisition spend should point.

Start now, even if the numbers are small

Brands often postpone this because they think they lack volume. If monthly cohorts are too small to be stable, use quarterly ones. The method still works and the shape still tells you something.

What you cannot do is reconstruct history you never recorded. Keeping a clean record of first purchase date, acquisition channel and first product from today costs nothing and is the single most valuable analytical asset a consumer brand accumulates.

The brands that struggle with this later are almost never short of customers. They are short of the record that would have let them learn from them.