Rubansi Vincent
Topline metrics are natural flatterers. Revenue can grow for six consecutive quarters while the underlying business quietly deteriorates, because acquisition spend is papering over customers who leave faster with every vintage. The instrument that exposes this, and the one I reach for first in any e-commerce or subscription engagement, is the cohort analysis: stop averaging across everyone, and start following groups of customers through time from the month they arrived.
Cohorts are to customer analytics what vintages are to a loan book. No credit officer would assess a portfolio by blending 2024 loans with last month's originations; yet businesses routinely assess "average retention" across customers acquired under entirely different products, prices, and channels. This article covers the mechanics, how to read the curves, and how to build a lifetime value number that would survive an accountant's review. Mine, for instance.
The cohort matrix: one table, most of the truth
Assign every customer to a cohort by first purchase or signup month. Then measure, for each cohort, activity in month 0, month 1, month 2, and so on. The result is the familiar triangular matrix: cohorts as rows, months-since-acquisition as columns, and a retention or revenue figure in each cell.
Two reading disciplines turn the triangle into insight:
- Read down the columns to compare vintages at the same age. If month-3 retention is 34% for the January cohort but 26% for April's, something changed: the product, the traffic mix, the onboarding, the discount that acquired them. This is the earliest reliable signal of quality drift you will ever get, months before it reaches the P&L.
- Read along the rows to see the shape of a customer lifetime: the steep early drop, then (in a healthy business) a flattening plateau of habitual customers. The plateau is the business; the cliff is the marketing bill.
"Growth tells you what you bought this month. Cohorts tell you what you get to keep."
Rubansi Vincent
Reading the curve: plateau, slope, and the smile you hope for
Plot retention against customer age and businesses sort themselves into three silhouettes. A curve that flattens at some stable percentage means you have a core of retained customers; the height of that plateau is arguably the single most important number in the company. A curve that decays toward zero without flattening means you are renting customers, and LTV math built on it is fiction with a discount rate. And occasionally a curve smiles, turning upward as dormant customers resurrect; common in replenishment categories, and a signal that reactivation deserves budget alongside acquisition.
Segment the curves before trusting them. Blended retention hides the fact that customers from paid social behave nothing like customers from referral, and that discount-acquired buyers churn at twice the rate of full-price ones. The operative question is never "what is our retention"; it is "which acquisition channels and offers produce customers whose curves flatten."
LTV, computed like an accountant rather than a pitch deck
Lifetime value is the most abused metric in commercial analytics, and the abuses are standard enough to enumerate. Revenue instead of contribution margin. Projected lifetimes stretched past any observed data. Blended averages across wildly different segments. A denominator-free existence, quoted without the CAC it is meant to justify.
The honest construction is layered and conservative:
- Start from contribution margin per cohort-month, not revenue: net of product cost, payment fees, shipping, returns, and the discounts that acquired the order. Margin is what a customer is worth; revenue is what they merely handled.
- Accumulate observed margin along the cohort row. For any cohort you can state, with no modeling at all, its actual value to date at each age. This empirical LTV-to-date is your bedrock; it cannot be argued with.
- Extend cautiously with the curve you observed, projecting the flattened tail forward over a stated horizon (12, 24, 36 months) rather than an infinite geometric series. State the horizon everywhere the number travels.
- Pair it with CAC and payback, always. The decision metric is rarely LTV alone; it is the LTV-to-CAC ratio by segment and, for cash-constrained businesses, the payback period: how many months until cumulative contribution margin covers the acquisition cost. A 4:1 ratio that pays back in 18 months can still sink a company that borrows at Kenyan interest rates.
"A customer lifetime value without a stated horizon, a margin basis, and a CAC beside it is not a metric. It is marketing about your marketing."
Rubansi Vincent
The operating loop: from triangle to budget
The point of this apparatus is a monthly decision, not a quarterly slide. The loop I install with clients is compact:
- The cohort matrix refreshes automatically: the SQL is a first-purchase window function and a group-by, feeding a Power BI page where the triangle, the curves by channel, and the payback table live together.
- Marketing reads payback by channel before setting next month's budget. Channels whose newest vintages pay back inside the target window earn more spend; channels whose curves are sagging get investigated, not defended.
- Product reads the early-retention columns after every meaningful release, because month-1 retention by cohort is the closest thing to a truth serum a roadmap will encounter.
- Finance gets the empirical LTV-to-date by vintage, which quietly becomes the basis for honest revenue forecasts and, in fundraising seasons, the diligence artifact that separates you from decks built on blended averages.
None of this requires exotic tooling. It requires the discipline of following vintages instead of averaging them, and a margin basis instead of a flattering one.
What changes when the curves run the meeting
The before-and-after on this work is behavioral. Before: growth debates run on opinion, the loudest channel wins budget, and "our customers love us" is supported by a blended average nobody can decompose. After: the January-versus-April column comparison settles the argument in one glance, a sagging vintage triggers a root-cause review within the month, and acquisition spend is allocated by observed payback rather than by last quarter's enthusiasm. The company starts buying customers the way an underwriter prices risk: by vintage, on evidence, with a stated appetite.
Make the shift to data-centric decision-making
If your growth reporting still runs on toplines and blended averages, you are flying a business on instruments that cannot detect its most common failure mode. Cohorts, retention curves, and margin-based LTV are not advanced analytics; they are the minimum honest accounting of what your marketing money buys.
Building this is squarely my work: the SQL that constructs the cohort spine, the Power BI model where triangles and payback tables refresh themselves, and the metric definitions that survive investor diligence. If you want to know what your customers are actually worth, vintage by vintage, tell me about your business and I will come back with an approach, a timeline, and a flat quote. Buy growth the way an investor would: with the curve in front of you.