The Metric That Looks Fine Until It Isn't
Your revenue dashboard looks healthy. Monthly orders are up. Average order value is holding. Customer count is growing.
And then, quietly, somewhere in a spreadsheet nobody checks, 70% of last quarter's customers never came back.
Aggregate metrics are optimistic by design. They blend your best customers with your worst, your most recent cohorts with your oldest, your high-LTV buyers with your one-and-dones. They tell you what happened in total. They don't tell you what happened to the people who bought from you.
Cohort analysis does. It is the analytical lens that makes retention visible — and once you see your retention curve, you cannot unsee what it is telling you.
What a Cohort Is and Why It Changes Everything
A cohort is simply a group of customers who share a common starting point — typically the month or week of their first purchase. Instead of asking "how much revenue did we make in March," cohort analysis asks: "of the customers who first bought in March, what percentage came back in April, May, June?"
This distinction matters enormously. Aggregate revenue can grow while cohort retention deteriorates — because you are acquiring new customers fast enough to mask the loss of old ones. This is a dangerous trajectory. It looks like growth. It is actually a leaking bucket.
The retention curve is what you get when you plot cohort retention over time. The x-axis is time since first purchase (days or months). The y-axis is the percentage of that cohort still active (making purchases). Every ecommerce business has one. Most have never looked at it.
How to Read a Retention Curve
The shape of your retention curve is a diagnosis. It tells you what your retention system is — or is not — doing. There are three shapes every brand eventually recognises:
The Cliff
A steep drop of 70%+ within the first 30 days, continuing to fall with no sign of stabilisation. This is the default state for brands with no post-purchase system. Customers buy once and disappear. The curve never flattens because there is no mechanism bringing people back. Nearly every ecommerce brand starts here.
The Slope
A gradual decline over 90–120 days before flattening at a low level (5–10%). Some retention activity is happening — perhaps a basic email sequence, perhaps organic word of mouth — but it is passive. The brand is not actively re-engaging customers at the critical windows. Revenue is predictable but structurally limited.
The Plateau
A sharp initial drop (inevitable — not every first-time buyer becomes a repeat customer) that stabilises into a flat or slowly rising line after day 30–60. This is what a healthy retention system looks like. The plateau means a stable core of customers keeps returning, compounding revenue across cohorts without additional acquisition spend.
The goal is not to eliminate the initial drop. It is to find the floor — and raise it.
The Three Cohort Metrics That Actually Matter
Once you are looking at cohort data, three numbers give you the diagnostic picture:
1. 30-day cohort retention rate
What percentage of customers who first bought in a given month made a second purchase within 30 days. This is your immediate post-purchase signal. If it is below 10%, your post-purchase sequence is either missing or broken. If it is above 20%, you have something working worth doubling down on.
2. 90-day cohort retention rate
The standard health benchmark. For consumable categories, target 35–45%. For non-consumable, target 15–25%. This is the number most directly correlated with LTV and the number most directly moved by the systems covered in the previous posts in this series.
3. Cohort revenue curve (not just headcount)
Retention rate tells you who came back. Revenue per cohort tells you how much they spent when they did. A cohort with 30% retention but rising average order value on repeat purchases is more valuable than one with 40% retention and declining AOV. Track both.
How Cohort Analysis Connects Everything
This post is the fourth in a series, and deliberately so — because cohort analysis is the measurement layer that makes the other three strategies legible.
The 4th Option argued that retaining existing customers is more valuable than acquiring new ones in a high-CAC environment. Cohort analysis is how you prove that argument with your own data. Pull your cohort curve from 12 months ago and from today. If it has improved, the retention investment is showing up. If it has not, you know exactly where to look.
The Second Order Problem identified repeat purchase rate as the key metric. But aggregate RPR is a blunt instrument. Cohort RPR — broken down by acquisition channel, product category, and first purchase type — tells you which customers are returning and why. That granularity is where the real optimisation lives.
Post-Purchase Email Flows described a five-email sequence timed to specific days post-delivery. Cohort analysis lets you verify it is working: you should see a measurable bump in your 30-day and 90-day cohort retention rates for the cohorts who entered after you deployed the sequence. No bump means the sequence is live but not converting. A bump means you have found something worth scaling.
Starting Without a BI Tool
You do not need a business intelligence platform to start doing cohort analysis. You need:
- An export of your order history (customer ID, order date, order value)
- A spreadsheet with a COUNTIFS formula or a pivot table
- 90 days of data to make the first cohort meaningful
Klaviyo, Shopify Analytics, and Google Analytics 4 all have built-in cohort reports. For brands under $5M in revenue, the native tools are sufficient. The discipline is not in the tool — it is in looking at the output weekly and asking: is this curve getting better or worse?
Takeaway
Your retention curve is a record of every post-purchase decision your business has ever made — or failed to make. Its shape tells you whether you have a retention system or an acquisition dependency. Whether your customers are compounding assets or expensive one-time transactions.
The four posts in this series have laid out the problem, the metric, the tactical fix, and now the measurement layer. The final step is deceptively simple: pull your cohort data, look at the curve, and decide whether what you see is acceptable.
Most brands, when they see it for the first time, decide it is not.