Profit by Customer Cohort

A store-wide repeat rate is one of the least informative numbers in ecommerce. It mixes customers acquired last week with customers acquired three years ago, and the older ones — who have had far longer to buy again — dominate the result.

Cohorts fix that by holding age constant. Group customers by the month of their first order, follow each group forward, and every comparison becomes like-for-like.

Building the view

Three inputs, none of them exotic.

Cohort membership. Every customer whose first-ever order fell in a given month. Not first order within a window — first ever, or returning customers get counted as new and the repeat rate is understated.

Acquisition spend for that month, from billing across every account, including agency fees and invoice tax.

Contribution margin on every subsequent order those customers place, costed individually rather than at a store average — later orders often have a different mix, and sometimes carry loyalty discounts the first order did not.

Plot cumulative contribution against acquisition spend, and the shape of the curve tells you what you need.

What the curve shows

The starting point is whether first orders cover acquisition. In the example they did not — $28,851 of margin against $37,335 of spend, so the cohort began underwater by $8,484.

The payback point is where cumulative contribution crosses acquisition cost. Month four in that cohort.

The slope after payback is how much the cohort continues to contribute, and how quickly the curve flattens.

Read together, these answer the question that first-order margin alone cannot: was that acquisition sensible? On day one the answer looked like no. Over twelve months it produced $17.42 per customer.

Payback period is the constraint, not lifetime value

Lifetime value gets discussed more; payback governs what you can actually do.

A store funding growth from its own cash flow is limited by how long its money is tied up, not by the eventual return. Two cohorts each worth $100 per customer are very different if one repays acquisition in six weeks and the other takes nine months.

The practical rule: payback shorter than your cash conversion cycle means growth funds itself. Longer means every additional customer widens the working capital gap, and growing faster makes it worse rather than better. That is the specific mechanism behind stores that run out of cash while profitable on paper.

Comparing cohorts properly

The whole value is in comparison, and the comparison has one rule: compare cohorts at the same age.

January’s cohort at month six against June’s cohort at month six. Never January at month twelve against June at month two, which is the shape most reports default to and which guarantees the older cohort looks better.

What you are looking for is drift in like-aged cohorts. If each successive cohort reaches a lower cumulative contribution by month six, customer quality is deteriorating — usually because acquisition has scaled into less engaged audiences, or because discounting has attracted buyers who came for the offer.

That signal appears months before it reaches net profit, which is the main reason to maintain the view at all.

What distorts cohorts

Promotional months. A cohort acquired during an aggressive sale behaves nothing like its neighbours: lower first-order margin, lower repeat rate, higher return rate. Label those cohorts rather than wondering why they break the trend.

Duplicate customer records. A customer who checks out as a guest twice appears in two cohorts and never repeats in either, which understates retention across the board.

Seasonality. A November cohort acquired on gift purchases behaves differently from a February one. Comparing across a year is safer than comparing adjacent months.

Channel mix shifts. If the balance between paid and organic acquisition moved, the cohort’s composition changed, and the cohort-level number blends two populations.

Using cohorts to justify spending more

The most valuable thing a cohort view enables is also the riskiest, so it is worth being precise about the conditions.

If mature cohorts reliably contribute well beyond their first orders, you can rationally bid above first-order break-even — because the customer, not the order, is what you are buying. That is how a store with strong retention out-competes one without it for the same traffic.

Three things have to hold. The repeat pattern must be measured, from your own matured cohorts rather than from a category assumption. The payback period must be funded, since the gap between spending and recovery is real cash that has to come from somewhere. And recent cohorts must be tracking their predecessors at the same age, because a plan built on last year’s retention fails silently if this year’s customers behave differently.

The failure mode is a store that extends its ceiling on the strength of a two-year-old cohort curve while its recent cohorts have been quietly deteriorating for six months. The cohort view is what would have shown that; using it only to justify the ceiling and never to check it is the specific way this analysis gets misused.

The one report worth keeping

If cohort analysis is going to exist as a single view rather than a project, make it this: cumulative contribution margin per customer at month six, by cohort, with acquisition cost per customer on the same chart.

Two lines, one point per month. Where they cross is your payback. Whether the gap between them is widening or narrowing over successive cohorts is whether acquisition is getting better or worse.

Almost every strategic question about paid growth can be answered from that chart, and almost no store has it.