Customer Retention Rate Formula: Why Ecommerce Needs a Different One

Search for the customer retention rate formula and every result gives you the same three variables. Take the customers you finished the period with, subtract the ones you acquired during it, divide by the customers you started with. That formula was written for businesses that sell contracts, and it quietly falls apart on a store where people buy whenever they feel like it.

It is worth knowing where the formula came from before you trust the number it gives you. It comes from subscription and account management, where a customer is a signed agreement with a start date and a renewal date. Ecommerce has neither. Nobody signs anything, nobody renews, and a shopper who bought once in March is not meaningfully "a customer you have" in April. The arithmetic still runs. It just answers a question you did not ask.

Below is the standard formula, the reason it produces a flattering and useless number for a direct-to-consumer brand, and the cohort version that actually tracks whether people came back. Start with the calculator: put the same brand through both and watch the two answers separate.

What the customer retention rate formula actually measures

The standard customer retention rate formula is consistent across the pages that rank for it. Zendesk, Salesforce and Gainsight all publish the same three variables:

CRR = [(E - N) / S] x 100

Every variable, defined

• E is the total number of customers at the end of the period.
• N is the number of new customers you acquired during that period.
• S is the total number of customers at the start of the period.
• The period is whatever window you chose, and it does more work than the other three variables combined.

Subtracting N is the only clever part. Without it, acquiring enough new customers would make retention look perfect while your existing base walked out of the door. Removing new customers from the ending count leaves you with the share of the original base that stayed.

A worked example where the formula works

Take a business selling annual software contracts. It starts the year with 200 accounts, signs 15 new ones, and finishes with 195. Apply the formula: (195 - 15) / 200 x 100 = 90%. It kept 180 of the original 200 accounts and lost 20.

That number is trustworthy because every variable is a fact rather than a choice. An account either had a live contract on 1 January or it did not. It either renewed or it churned. The start and end counts are not estimates and the period is set by the contract, not by the analyst. WallStreetPrep frames the metric this way, as a subscription and contract measure, which is exactly where it belongs.

Why the standard formula breaks on ecommerce data

Now run the same formula on a skincare brand on Shopify. Nothing in the calculation errors. Everything in the interpretation does.

Your lookback window decides your answer

To count "customers at the start of the period" you first have to decide what makes somebody a customer on a given date. There is no contract to check, so you pick a lookback window: anyone who ordered in the last 12 months, or 24 months, or ever. That choice is not a detail, it is the denominator, and it changes the answer more than your actual retention does.

Widen the window to "anyone who has ever ordered" and S grows every month while the numerator does not, so your retention rate declines forever regardless of how well you serve people. Narrow it to 90 days and S collapses, so the rate jumps. Two analysts at the same brand, both applying the formula correctly, can report retention 30 points apart and both be right. Neither number is comparable to last quarter's unless the window was identical, and in most brands nobody wrote the window down.

Still a customer is not the same as bought again

The deeper problem is what the formula treats as success. It counts a customer as retained when they are still on the list at the end of the period and were not newly acquired. In a subscription business that means they are still paying. In ecommerce it means nothing more than that they bought something once and have not been deleted.

A shopper who ordered a single candle 11 months ago, never returned, and is sitting in your customer table counts as retained under this formula. That is the opposite of retention. Retention in a store where nobody signs anything can only mean one thing: they came back and bought again. If your formula never asks whether a second order happened, it cannot measure retention no matter how carefully you compute it.

Why ecommerce sits at 30% on every benchmark table

This is where the two problems become visible. Industry benchmark tables consistently put ecommerce at the bottom. Triple Whale publishes a table of Statista figures giving media companies and professional services 84%, software 77%, banking 75%, hospitality 55%, and ecommerce 30%, the lowest row on the list. It publishes that table on the same page as the subscription formula, and never reconciles the two.

Ecommerce is not four times worse at keeping customers than a bank. It is being measured with a different instrument. The 77% for software is contractual renewal, a fact. The 30% for ecommerce is a repeat purchase rate wearing a retention label, and it is being compared with numbers that mean something else. Reading that table and concluding your category is bad at retention is the wrong lesson. The right one is that a cross-industry retention table cannot be compared row to row at all.

Gainsight's version of the same guide makes the mismatch clearer by omission: its benchmark table covers B2B SaaS, enterprise software, banking, telecom, streaming and quick-service restaurants, with a note telling you to annualise the monthly figures before comparing rows. The advice is sound and the ecommerce row is simply absent.

The cohort retention formula for ecommerce

Replace the question. Instead of asking what share of your customer list survived a period, ask what share of a specific group of buyers came back within a set time of their first order. That is a cohort, and it removes both faults at once.

The formula

Cohort repeat rate = (buyers in the cohort who placed another order within the window / total buyers in the cohort) x 100

A cohort is every customer whose first order landed in the same month. That definition is the whole trick. Membership is fixed forever the day someone first buys, so the denominator cannot drift, and there is no lookback window to argue about. The numerator counts an actual second order, so "retained" means what the word ought to mean.

Because each cohort is anchored to its own start date, cohorts are comparable to each other in a way that period retention rates never are. January's buyers at day 180 sit next to June's buyers at day 180 on the same footing. That comparison is what tells you whether the customers you are acquiring now are better or worse than the ones you acquired last year, which is the question behind most retention work. Running it properly is the subject of cohort analysis for ecommerce.

Set the window from your purchase cycle, not the calendar

The window is the one judgement call left, and quarters and years are usually the wrong answer. Set it from how long your product actually lasts. A coffee subscription refill runs out in about 30 days, so a 30 to 60 day window is the honest test. A supplement bottle covers 60 to 90 days. A mattress has a replacement cycle measured in years, so no repeat window under 24 months means anything and the metric to watch is referral rather than repurchase.

Get this wrong in either direction and the number misleads. Measure a 90 day consumable at 30 days and you will declare a retention problem that is really just the product still being in use. Measure it at 365 days and you will miss the customers who lapsed at day 120, long after you could have done anything about it.

One brand, three formulas, three different answers

Here is the same store, one calendar year, measured three ways. It began the year with 4,000 customers under a 12 month lookback, acquired 1,800 new ones, and ended with 5,200. It has 8,300 buyers in its all-time history. Of the 1,800 first-time buyers who formed that year's cohorts, 342 placed a second order within 180 days.

Only the third number survives contact with a follow-up question. Ask "which of these customers should we email" and 85% has no answer, because it never identified anybody. The cohort figure names the 1,458 first-time buyers who did not come back and dates their last order, which is a working list rather than a statistic. It is also the version that tells you whether spending more on acquisition is a good idea, since a 19% repeat rate means four in five acquired customers have to pay for themselves on the first order.

Retention rate, repeat purchase rate and churn rate

These three get used as synonyms and answer different questions. Picking the wrong one is how a retention review ends with everyone agreeing on a number nobody can act on.

Churn deserves a specific warning. Churn and retention only sum to 100% when both are built on the same definition of a customer, and in ecommerce they frequently are not: retention gets computed on a 12 month lookback while churn gets computed on people who lapsed past 90 days. Both get reported in the same deck and the pair is quietly incoherent. If you want the leading indicator rather than the arithmetic, it is more useful to spot at-risk customers before they leave than to total up the ones who already did.

What counts as a good customer retention rate in ecommerce

The honest answer is that the question is badly formed, and the reason is the table above. A published average is only a target if it was computed the way you compute yours, and almost none of them were.

The useful ranges, for cohort repeat rate inside a window matched to the purchase cycle, look nothing like the benchmark tables. For most non-consumable DTC categories a first-to-second order rate of 20% to 30% is ordinary, 30% to 40% is strong, and past 40% you are either in a consumable category or you have a genuine subscription habit. Consumables and refill-driven brands run considerably higher, which is why comparing a coffee brand with a furniture brand on the same number is meaningless.

Two things matter more than any of those ranges. The first is direction: whether the cohort you acquired this quarter is repurchasing better or worse than the one you acquired a year ago, measured at the same number of days after first order. The second is spread by acquisition source, because a blended repeat rate averages together channels that behave nothing alike, and the average hides which ones are worth more money.

Compare cohorts against your own history, not an industry average

Your own back catalogue is the only benchmark computed to your definitions. Line up the last eight or twelve monthly cohorts, read each at a fixed day count after first order, and the trend line answers the question a benchmark table cannot: is this getting better. A brand at 18% and climbing across four consecutive cohorts is in better shape than one sitting at 28% and sliding, whatever the published average says. That is also the comparison that survives a board meeting, because nobody can dispute the definition when both numbers came out of the same query.

How to calculate customer retention rate without a spreadsheet

The formula is arithmetic. The reason retention numbers are wrong in practice is almost never the formula, it is the customer records underneath it, and a spreadsheet cannot fix that.

To build a cohort you need every order tied to the right person, for every person, across the whole period. That is where it breaks. The same shopper who buys on a phone over cellular data, then again on a laptop at home, then a third time through a link in an email, can land in your data as three separate customers. Three first orders, three cohorts, three people who never repurchased. Your repeat rate is understated and the fix is not in the formula.

Platform reporting does not resolve this on its own, because a store sees a new cookie and records a new visitor. It is why the retention number in one tool disagrees with the retention number in another, and why brands stop trusting both. Identity has to be stitched before any of the arithmetic means anything.

Polar handles this with LifetimeID, a persistent identifier built from device, network and personal signals rather than a browser cookie, so repeat orders from the same human collapse back onto one customer across devices, networks and stores. On top of that, Polar's retention model is order-driven rather than subscription-driven by design: customers are grouped by the period of their first order, and the cohort table tracks repeat purchasing, orders, sales, margin, lifetime value and lifetime value against acquisition cost for each cohort over time. It supports both period bases the article has been arguing about, calendar periods where month zero is the acquisition month, and rolling windows measured from each customer's own first order date, so you can pick the one that matches your purchase cycle instead of the one your reporting tool assumed. Cohorts can be filtered by what a customer bought first or which channel acquired them, which is how the blended average gets broken apart. It is the same view that keeps retention and lifetime value in one place, rather than in two tools that disagree.

Five mistakes that make a retention number meaningless

1. Not writing down the window. If the lookback or the repeat window is not recorded next to the number, the number cannot be compared with anything, including itself last quarter.
2. Building cohorts from any order instead of the first. A customer must enter exactly one cohort, on their first purchase, forever. Group by "ordered in January" and repeat buyers land in every cohort they shop in, which inflates every row.
3. Mixing period lengths. Pulling S from a monthly view and E from a quarterly one produces a number that looks plausible and means nothing. Annualise before comparing anything monthly to anything yearly.
4. Comparing your rate with a cross-industry table. Covered above, and the most common. Those rows were computed with different instruments on different business models.
5. Reporting one blended rate. A single company-wide figure averages channels, products and geographies that behave differently. It moves for reasons nobody can explain, so it drives no decisions.

Where to start

Pick a repeat window that matches how long your product lasts. Group customers by the month of their first order and never move them. Measure what share of each cohort placed a second order inside that window, then read the last twelve cohorts as a trend rather than as one number. If the resulting figure is lower than the retention rate you have been reporting, that is the point: the old number was measuring your customer list, and this one is measuring your customers.

From there the useful next steps are splitting cohorts by acquisition channel and first product, which shows where good customers actually come from, and pairing repeat rate with lifetime value against acquisition cost, which turns retention from a reporting metric into a spending decision.

FAQ