Retention Cohort Analysis in 2026: Reading Your Repeat-Purchase Curve Correctly
TL;DR
How to build and read a repeat-purchase cohort table in 2026, and the specific curve shapes that reveal a genuine retention problem vs normal category behaviour.
Ask a brand what their repeat purchase rate is and most can give you a single number - "we're at 25%." That single number hides more than it reveals. A cohort table, breaking that number down by when customers first purchased, tells a very different and far more useful story. I am Naman Khetawat, and here is how to build and actually read one.
The citable answer: a retention cohort table tracks the percentage of customers from each acquisition month who make a repeat purchase within 30, 60, and 90 days, and reading it correctly means comparing the curve shape across cohorts over time - a flattening or declining curve across newer cohorts signals a genuine retention problem, while a single low blended number could simply reflect a large recent cohort still early in its purchase cycle. Here is how to build and interpret it.
Why a Single Blended Repeat Rate Misleads
A blended repeat purchase rate averages together customers who have had 12 months to make a second purchase with customers who joined last week and have barely had time. If a brand had a large acquisition spike last month, the blended rate will look artificially low simply because a large chunk of recent customers have not had time to repurchase yet - not because retention is actually declining. A cohort table separates customers by acquisition period specifically to avoid this distortion.
Building the Cohort Table
| Acquisition month | 30-day repeat % | 60-day repeat % | 90-day repeat % |
|---|---|---|---|
| Jan cohort | 8% | 15% | 22% |
| Feb cohort | 7% | 14% | 21% |
| Mar cohort | 6% | 11% | — (too recent to measure yet) |
Reading this table means comparing each cohort's trajectory at the same point in its lifecycle - is the March cohort's 30-day rate meaningfully behind January's, or roughly in line? A genuine retention problem shows up as newer cohorts consistently underperforming older cohorts at the same age, not as a single month's blended number looking lower than expected.
What Different Curve Shapes Actually Mean
- Flat curve across cohorts: Stable retention. Not necessarily good or bad in isolation, but consistent - useful as a baseline for testing new retention initiatives against.
- Declining curve across newer cohorts: A genuine retention problem worth investigating - something about recent customer experience, product, or onboarding has changed for the worse.
- Improving curve across newer cohorts: Evidence that retention initiatives (new flows, product changes) are working, if the timing lines up with when those changes launched.
Diagnosing the Cause Behind a Declining Curve
Once a cohort table reveals a genuine decline, the next step is correlating it against what changed around the same time - a product change, an acquisition channel shift bringing in lower-intent customers, a pricing change, or a support process degradation. A common finding is that a declining cohort curve traces back to an acquisition channel shift (a new, high-volume but lower-intent traffic source) rather than anything about the product or retention flows themselves - the newer cohort simply contains a different type of customer than earlier cohorts did.
This diagnostic process is central to every retention and remarketing engagement, because a cohort table alone identifies that a problem exists, but correlating it against acquisition and product changes is what identifies why.
How Often to Rebuild the Cohort Table
Monthly rebuilds are typical for most D2C brands - frequent enough to catch emerging trends, infrequent enough that each cohort has meaningful time to develop before being judged. Rebuilding weekly usually produces too much noise from small sample sizes per cohort to draw reliable conclusions.
A Real Example
A skincare brand's blended repeat rate had dropped from 24% to 18% over two months, triggering concern about a possible product quality issue. The cohort table showed the decline was concentrated entirely in cohorts acquired after a new, cheaper-CPA ad channel had been added - customers from the original channels were retaining at the same historical rate. The real issue was the new channel bringing in lower-intent customers at a cheaper acquisition cost but a much worse retention profile, not a product or fulfilment problem. Reallocating budget away from that channel resolved the apparent retention decline without any product change.
FAQ
What is a retention cohort table?
A table tracking repeat purchase rate for customers grouped by their acquisition month, measured at consistent intervals (30, 60, 90 days). It reveals whether retention is genuinely changing over time, rather than being distorted by a blended average across cohorts at different lifecycle stages.
Why does a single blended repeat purchase rate mislead?
It averages customers who have had ample time to repurchase with recently acquired customers who haven't. A large recent acquisition spike can make the blended number look artificially low without reflecting any real change in retention behaviour.
What does a declining cohort curve usually indicate?
A genuine retention problem worth investigating, but the cause is not always product-related - a common cause is an acquisition channel shift bringing in customers with a different, lower-retention profile than prior channels.
Build a Cohort Analysis That Reveals the Real Story
If you're tracking a single blended repeat purchase number, a cohort breakdown is usually the fastest way to find out what's actually happening underneath it. Book a call with Balistro and we will build your cohort analysis.
