Cohort-Based vs Sitewide-Average Retention Rate
Sitewide-average retention is a blended, lagging number. A cohort table shows you which acquisition month is actually bleeding — often months before revenue notices.
Cohort-Based vs Sitewide-Average Retention Rate
Sitewide-average retention blends all buyers into one number; cohort-based retention tracks each acquisition group separately over time.
Sitewide-average retention is a single percentage: of everyone who bought in a window, what share came back. It's easy to compute, easy to report, and structurally unable to tell you which customers are actually leaving. The number moves slowly because loyal buyers acquired two years ago mask the behaviour of buyers acquired last quarter.
Cohort-based retention slices the same customer base by acquisition month (or week, or campaign) and follows each cohort down its own curve. The output is a table, not a KPI — one row per cohort, one column per month-since-first-order. Reading that table is how you spot a bleeding Q3 cohort behind an apparently healthy 34% sitewide average.
The two numbers answer different questions. Sitewide-average retention answers 'is the store, on aggregate, keeping customers?' Cohort retention answers 'is what we did last quarter working?' If you only report the first, you'll congratulate yourself in April on a healthy 34% while your February acquisition cohort is quietly repeating at half the rate of last year's.
This matters most for stores that scaled paid acquisition, ran a big promo, changed price, or shifted product mix in the last 6-12 months. Every one of those events creates a cohort with a different retention shape — and every one of them gets flattened into the sitewide average alongside your loyal 2022 buyers.
Same store, same month: how a bleeding Q3 cohort disappears into the sitewide average
| Acquisition cohort | Cohort size | M1 repeat rate | M3 repeat rate | M6 repeat rate | Contribution to sitewide avg |
|---|---|---|---|---|---|
| Legacy (pre-2023) | 18,400 | 42% | 38% | 35% | Heavy — anchors the blend |
| Q1 2024 (organic-led) | 4,200 | 31% | 26% | 22% | Moderate |
| Q2 2024 (mixed) | 5,100 | 29% | 24% | 20% | Moderate |
| Q3 2024 (paid + discount push) | 6,800 | 18% | 12% | 9% | Diluted by legacy weight |
| Sitewide blended average | 34,500 | 31% | 27% | 24% | Looks fine |
The Q3 cohort in that table is repeating at 9% by month 6 — roughly a third of what the older cohorts do. But because legacy buyers make up more than half the base and repeat at 35%, the blended sitewide number lands at a comfortable 24%. Anyone reading only the top-line KPI would conclude retention is 'flat', when in reality the newest and largest recent cohort is bleeding.
Why the sitewide average is structurally a lagging indicator
The blended rate has a demographic problem: it's weighted by whichever cohort is largest, and the largest cohort in most stores is the accumulated legacy base. Loyal legacy cohorts make your sitewide retention rate a lagging indicator — deterioration in a new cohort has to become very large, or persist for many months, before it moves the aggregate needle.
By the time a bleeding cohort is big enough to drag the average down, you've usually spent another quarter of ad budget acquiring more customers with the same broken pattern. That's the operational cost of only watching the top-line number: you find out about a retention problem two quarters after you caused it, and after you've paid to reproduce it at scale.
The 'retention is fine' trap
If your sitewide retention has been stable for 6+ months while your CAC is climbing and paid volume is up, that stability is almost certainly a legacy-cohort mirage. Pull the cohort table before you take the flat trend line as good news.
How to actually read the cohort table
Read the table two ways. Down a column: are cohorts at the same age (say, month 3) retaining better or worse over time? Steady decline column-by-column means acquisition quality is degrading. Across a row: how does one cohort's curve shape compare to older cohorts at the same maturity? A sharp M1→M3 drop that older cohorts didn't have is your break-in-the-curve signal — often traceable to a specific price change, promo, or channel shift.
Two caveats before you sound the alarm. First, recent cohorts almost always look worse in month 1 because of truncation — they haven't had time to complete the repeat window yet. Second, small cohorts are noisy: a 200-customer row swinging 8 points month to month is often just low N, not a real signal. Setting a cohort-deterioration alert threshold that fires before revenue does means baking both of those into the rule.
Sitewide average vs cohort retention curves — the divergence the blend hides
Legacy cohort (pre-2023)
Q1 2024 cohort
Q3 2024 cohort (paid + discount)
Sitewide blended average
Cohort vs sitewide-average retention — FAQ
Average (sitewide) retention is one number blending every customer regardless of when they were acquired. Cohort retention groups customers by acquisition month and tracks each group's repeat behaviour separately, producing a table rather than a single KPI. The average tells you the aggregate state; the cohort table tells you which group is driving the change.
Because loyal legacy cohorts anchor the blend. A deteriorating new cohort has to be very large or very bad to shift the aggregate, and by the time it does, you've often reproduced the pattern for another quarter. Pull cohort rows for the last 6 acquisition months to see what the average is hiding.
Group orders by customer first-order month, then for each cohort compute the share of customers who placed another order in month 1, 2, 3 and so on. Shopify's built-in report gives a basic view; Metricuno's retention calculator pulls the same data from your Shopify order history and produces the full cohort table plus curve chart automatically.
Read down columns to compare cohorts at the same age (are new cohorts retaining worse at month 3 than old ones did at month 3?), and read across rows to see each cohort's curve shape. A sharp M1→M3 drop in a specific cohort that older cohorts didn't have usually maps to a specific event — a promo, price change, or channel shift that quarter.
Because they haven't had time to complete the repeat window. A cohort acquired last month has had one month to come back; a cohort acquired a year ago has had twelve. This is truncation, not deterioration. Only compare cohorts at the same maturity — month 3 of the new cohort against month 3 of older ones.
A rough rule: cohorts under 200-300 customers swing 5-10 points month to month from randomness alone. Look for a bad row that (a) has enough size, (b) shows the pattern across at least two consecutive month columns, and (c) diverges meaningfully from adjacent cohorts. One noisy point in a 150-customer row is not a five-alarm fire.
Discount-acquired cohorts typically show a decent M1 (they came back to use another code) but drop off sharply by M3-M6 compared to organic or full-price cohorts. Sitewide averages hide this because the M1 bump partially offsets the later collapse. Splitting the cohort table by acquisition source or first-order discount status makes the pattern obvious.
Yes — reading paid and organic cohort rows side by side is one of the most useful diagnostics on the table. Paid cohorts usually have lower and flatter curves; organic cohorts sit meaningfully higher. If paid retention gets close to organic retention, your paid channel is finding better-fit customers. If the gap widens, you're buying worse buyers.
It depends on your repeat cadence. Apparel and beauty typically use a 6-12 month window; consumables with 30-day replenishment use 90-day; considered-purchase categories may need 18-24 months. Choose a window at least 2-3× your median inter-order gap so you're not calling churn on customers who just haven't hit their next natural repurchase yet.
It depends on vertical: subscription beauty might hold 45%+ at M3, one-off apparel 25-30% at M3, high-consideration electronics 10-15%. Healthier curves flatten (retention stabilises after M3-M6) rather than collapsing linearly toward zero. A curve that keeps falling steeply through M6 usually means you're not building genuine repeat demand — you're just cycling through acquisition.
Get an AI expert review of your site
Paste your URL — Metricuno's AI runs the same heuristic checks a senior CRO consultant would, scoring your page and prioritising the fixes that'll move conversion fastest.