What retention rate measures, and why the subtraction matters
Retention rate answers one question: of the customers you had at the start of the period, what share were still customers at the end? Everything hangs on separating that from growth, which is why the formula subtracts new customers from the closing count before dividing.
Consider a business that starts a month with 100 customers, ends it with 105, and acquired 20 along the way. The headline looks like 5% growth. The retention picture is quite different: only 105 − 20 = 85 of the original hundred survived, so retention is 85% and churn is 15%. Without the subtraction the metric would read 105%, and it would keep reading well above 100% for as long as acquisition outran losses — which is exactly the period during which a retention problem is most expensive to ignore, because you are pouring acquisition spend into a base that leaks.
Retention and churn are complements: churn is 100% minus retention, for the same period and the same denominator. They are not two measurements; they are one measurement expressed twice. What matters is that both use the opening cohort as the denominator. Some teams compute churn as customers lost divided by the average of the opening and closing counts, which is a defensible convention for a fast-growing base but is not the same number, and mixing the two across periods produces a trend that is partly an artefact of the method.
For transactional businesses without subscriptions there is a second, complementary measure: repeat purchase rate, the share of buyers in a period who bought more than once. It answers a different question — not "did they stay" but "did they come back" — and it is often the more actionable of the two in retail and e-commerce, where nobody formally cancels anything.
Working through the formula and its annualisation
Retained customers = end − new. This assumes every customer who leaves does so from the opening cohort, and that nobody acquired during the period also leaves during it. Both assumptions are approximations. In a fast-growing month with high early-life churn, some of the new customers will have gone by the closing date, so the subtraction slightly overstates the number of original customers retained. If your acquisition is large relative to your base, measure the cohort directly instead — tag the opening customers by ID and count how many are still active — which is exact and only slightly more work.
Rate = retained ÷ start. The denominator is the opening count and never the closing count. Using the closing count when the base is growing flatters the rate, and using it when the base is shrinking understates it, so the error changes sign with your growth and cannot be corrected by a constant adjustment.
Annualised retention = rateperiods per year. Compounding is the step people get wrong most often. A 95% monthly retention rate is not 95% annual and it is not 40% annual either — it is 0.9512 = 54.0%. The intuition to hold on to is that survival multiplies rather than adding, so small monthly differences produce enormous annual ones: 99% monthly leaves 88.6% of the cohort after a year, while 95% leaves 54.0%. That four-point monthly difference costs you a third of the base.
The reciprocal of churn gives the average customer lifetime in periods. At 2% monthly churn the average life is 1 ÷ 0.02 = 50 months. That figure is the bridge to customer lifetime value, which is essentially the monthly gross profit multiplied by that lifetime — and to the CAC payback calculator, which asks whether that life is long enough to repay what the customer cost to acquire.
Worked example: a month with 1,000 opening customers
You begin the month with 1,000 customers, end it with 1,150, and acquired 250 during the month. Of the 900 customers who placed at least one order, 320 placed two or more.
- Retained. 1,150 − 250 = 900 customers from the original cohort.
- Retention rate. 900 ÷ 1,000 × 100 = 90.00%.
- Churn rate. 100% − 90% = 10.00%, which is 100 customers lost out of the opening 1,000.
- Average customer life. 1 ÷ 0.10 = 10 months, if this rate persists.
- Annualised retention. 0.9012 = 0.2824, or 28.24%. Of the opening thousand, roughly 282 would remain after a year at this rate.
- Repeat purchase rate. 320 ÷ 900 × 100 = 35.56%.
Step 5 is the one that changes decisions. The month looks healthy — the customer count rose by 150 — but the base is turning over completely in well under two years, and at 250 acquisitions a month you are replacing 100 lost customers before you add a single net new one. That is 40% of your acquisition budget spent standing still. Improving retention from 90% to 93% cuts monthly losses from 100 to 70 and lifts annualised survival from 28.24% to 0.9312 = 41.86%, which is 41.86 ÷ 28.24 − 1 = a 48% relative improvement in the share of a cohort still present at the year mark.
How to read your retention rate
Read it against the period first. A 90% rate is excellent annually, ordinary quarterly and poor monthly, because the compounding differs by a factor of twelve. Any retention figure quoted without its period is uninterpretable, and this is the most common way the metric is miscommunicated between teams.
Then read it against your business model. Contractual businesses with annual commitments should see very high period retention simply because customers cannot leave mid-term; the meaningful measurement is at renewal. Month-to-month subscriptions expose the true rate every month. Transactional retail has no formal relationship at all, so the retention framing is a modelling choice — you have to define a lapse window, and the rate you report depends heavily on whether you chose 90 days or a year.
Then look at the shape rather than the level. A cohort's retention rate almost always improves with age, because the customers most likely to leave leave first, so the survivors are progressively more committed. The projection in this calculator's table applies the current period rate unchanged and is therefore conservative for a real cohort — treat it as a floor for the later months. If your rate is not improving with cohort age, that is a stronger signal than the level itself, and it usually points to a product problem rather than a marketing one.
Finally, distinguish customer retention from revenue retention. Losing 10% of your customers is a very different event depending on whether they were your smallest or your largest, and expansion revenue inside the survivors can leave revenue flat while the customer count falls. Read this metric next to net revenue retention, which weights every account by what it pays.
What a monthly retention rate means over a year
| Monthly retention | Monthly churn | Share of cohort left after 12 months | Average customer life |
|---|---|---|---|
| 99% | 1% | 88.6% | 100 months |
| 98% | 2% | 78.5% | 50 months |
| 97% | 3% | 69.4% | 33.3 months |
| 96% | 4% | 61.3% | 25 months |
| 95% | 5% | 54.0% | 20 months |
| 90% | 10% | 28.2% | 10 months |
| 85% | 15% | 14.2% | 6.7 months |
The average-life column assumes the rate stays constant, which real cohorts beat because early leavers are removed from the surviving population. Use it to compare scenarios, not as a forecast of any individual cohort.
Measurement traps
- Forgetting to subtract new customers. The rate then measures growth, and it will read above 100% precisely when acquisition is covering a retention problem.
- Counting reactivated customers as retained. A lapsed customer who returns is a re-acquisition, not a retention. Counting them as retained pushes the rate above 100% and hides real losses.
- Mixing period lengths. Monthly, quarterly and annual rates are not comparable without compounding. Convert with the exponent before you put two figures side by side.
- Changing the denominator between periods. Opening count and average count are both defensible conventions; switching between them creates a trend that is an artefact of the method.
- Blending segments with different behaviour. Enterprise and self-serve cohorts often differ by twenty points or more. The blended rate describes neither, and it moves whenever the mix moves.
- Treating a customer count as a revenue proxy. Losing your ten largest accounts and your ten smallest give identical customer retention and completely different revenue outcomes.
Where retention sits among loyalty and revenue metrics
Retention rate is the count-based member of a family. Churn rate is its complement and reports the same information from the other side. Net revenue retention weights each account by its spend and adds expansion, so it can exceed 100% when upsell outpaces losses. Net promoter score measures stated intent rather than behaviour, and is best used as a leading indicator that you validate against realised retention rather than as a substitute for it.
The economic reason to care is compounding. Retention enters lifetime value through the average customer life, which is the reciprocal of churn, so improvements in retention have a convex effect: moving from 5% to 4% monthly churn extends average life from 20 months to 25 — a 25% gain — while moving from 2% to 1% extends it from 50 months to 100, a doubling. The same one-point improvement is worth four times as much at the low-churn end. This is why mature subscription businesses fight over fractions of a point that would look trivial to a company at 10%.
For e-commerce, the repeat-purchase framing is usually the more practical entry point, because there is no cancellation event to measure. Track the share of a month's buyers who had bought before, and the share of a cohort that buys again within a defined window, and pair them with average order value to see whether returning buyers are also bigger buyers. When both rise together, lifetime value rises faster than either metric alone suggests, because the two multiply.
Whatever the model, measure cohorts rather than aggregates wherever you can. An aggregate rate blends cohorts of different ages and different acquisition sources, so it can drift for reasons that have nothing to do with how well you serve customers — a large intake of low-intent customers from one campaign will depress the aggregate for months. Cohort tables isolate that, and they are the only way to see whether the product is actually getting stickier.