Repeat Customer Rate Moves With the Window You Chose

Why this matters

Almost nobody needs two visits from your trade in the same week, so a weekly reading of this figure sits near zero for a perfectly loyal customer base. Stretch the window and the number climbs, with nothing whatever having changed about the customers. Owners compare this figure against last quarter's, against a trade publication's, or against the shop across town, and in every one of those comparisons the window is doing more work than the loyalty is. Read it without knowing that and you will congratulate yourself for a calendar change or panic over one.

Both halves, one window

Of the customers you served at least once inside the window, the share you served more than once inside that same window.

Both halves are period-scoped and both use the identical period. That sounds like a tidy construction and it is the source of every problem here, for one reason: a visit that falls outside the window removes a customer from the numerator while leaving them in the denominator, because one visit inside is enough to keep them counted as served. Shrink the window and you strip visits away faster than you strip customers away, so the ratio has to fall. It is not sensitive to the window. It is substantially a statement about the window.

One shop, three windows, one day

A shop running about 2,400 jobs a year across about 1,700 distinct customers. Same shop, same records, three windows computed on the same afternoon.

The job counts reconcile in each row. In the year, 1,263 customers had one job each and 437 averaged about 2.6 jobs, which is about 2,400 jobs. In the quarter, 452 customers had one job each and 68 averaged about 2.2 jobs, which is about 600 jobs, a quarter of the year. In the week, 40 customers had one job and 3 had two each, which is 46 jobs, about a fifty-second of the year.

Nothing changed between the rows except the length of the period. Any of the three figures could be reported as "our repeat customer rate" and all three would be arithmetically correct.

One more thing about the weekly row before anybody uses it. Its denominator is 43 customers, and the floor for reading a share over a customer list is 50, because this figure is read for a few points of movement between one window and the next, and at 43 a single customer moves it by 2.3 points - larger than most of the differences anybody would act on. The weekly figure fails that floor on its own denominator, so it is unreadable for a second and entirely separate reason before the window argument even starts.

Why the short window drops faster than you expect

The customer who makes this vivid is the one on a maintenance cycle: two visits a year, roughly six or seven months apart, a spring appointment and an autumn one. That customer is the minority here - 437 of the 1,700, which is exactly the 25.7 percent the annual row reports - and the other 1,263 were served once. In a twelve-month window that person is one customer, served twice, sitting in the numerator. Split the same year into quarters and that person is two different single-visit customers in two different quarters, in the denominator both times and in the numerator neither time.

That has a consequence people trip over constantly. You cannot add up window denominators and you cannot average window rates. The four quarterly denominators in the example sum to about 2,080 customers against an annual denominator of 1,700, because customers served in more than one quarter are counted once per quarter. And the average of the four quarterly figures is around 13 percent, against 25.7 percent for the year, both of them repeat rates each computed on its own window. They are not two estimates of one quantity that disagree. They are two different quantities, and the annual figure is not recoverable from the quarterly ones.

The numerator is not what "repeat customer" means in plain English

Ask an owner what a repeat customer is and they will describe someone who has used them before. The metric does not say that. It says someone served more than once inside this window, which gets two whole groups backwards.

A customer whose very first job with you was in March and who needed a return visit in April is counted as a repeat customer. Most owners would call that a new customer and a callback. A customer who has used you every year for nine years and needed you once this year is counted as not a repeat customer at all.

Put numbers on it in the example year. Of the 437 counted as repeats, 96 were customers whose first-ever job fell inside the same window. Of the 1,263 counted as single-visit customers, 311 had been served in at least one earlier year. Build the figure the way the phrase sounds instead, as customers served in the window who have any job history before it, and you get 341 plus 311, or 652 customers with prior history out of the same 1,700 served, which is 38.4 percent.

So the same customer base, in the same window, reads 25.7 percent on the in-window definition and 38.4 percent on the prior-history definition, both computed as a share of the identical 1,700 customers served. The gap is not an error in either. It is two different questions, and the second one is closer to what the phrase promises, so if you report the first you had better say which you mean.

Choose the window from your service interval, not from your calendar

The rule that makes this figure mean something: the window must be at least twice your trade's natural service interval, because a customer cannot register as a repeat unless the window is long enough to hold two of their visits.

The example shop's natural interval is about six months, two visits a year for a customer on a maintenance cycle, so twice that is twelve months and the annual window is the correct one. A round-based trade working a customer every week or fortnight has an interval measured in days, so twice it is about a month, and a monthly window is not only legitimate there but is more informative than an annual one, which would pin nearly every customer into the numerator and flatten the figure against its ceiling. A trade whose customers need them once every several years cannot use this figure at all on any window a business meeting happens in, and should be reading time between visits directly instead.

Quarterly reporting is where this goes wrong most often, because quarterly is the reporting rhythm most shops already have, so the window gets chosen by the meeting calendar rather than by the work. The example shop's quarterly figure of 13.1 percent is not a bad result. It is a result about quarters.

Why this figure cannot be compared across trades

Two shops in different trades quoting repeat rates are comparing service intervals. A cleaning round reporting 80 percent on a monthly window and a heating shop reporting 26 percent on an annual one tell you nothing about which has better customer relationships, and the comparison stays meaningless even after you put both on the same window, because putting both on a month would drive the heating shop near zero and putting both on a year would push the cleaning round against its ceiling.

The only cross-shop comparison worth making is between shops in the same trade, on windows scaled the same way to the same interval, with both stating which numerator definition they used. In practice that means the comparison worth making is against your own prior windows, of equal length, at the same point in the season.

The figure that answers the question you were actually asking

If what you want to know is whether customers come back, fix the cohort at an origin instead of scoping both halves to one window. Take everyone served in a stated year, then ask what share of exactly those people were served again in the following year. The denominator is frozen the day the origin year closes and does not move afterwards, which is the property the in-window figure lacks.

Run it on the example shop. Of the 1,700 customers served in the prior year, 598 were served again in the year measured, so 35.2 percent of that cohort returned. That reconciles with the other figure: of the 652 customers served in the measured year who had any prior history, 598 came from the immediately preceding year's cohort and the other 54 had last been seen two or more years earlier.

The two numbers answer different questions and both are worth having. The in-window figure tells you how much of this year's work went to people you saw more than once this year, which is a scheduling and capacity fact. The cohort return rate tells you whether last year's customers came back, which is the loyalty question, and unlike the in-window figure it gets worse when customers leave rather than staying silent about it.

Three ways to break this figure on purpose

Each of these collapses the metric into something else entirely, which is the fastest way to see what it is really made of.

Collapse the window to one day. The only way to be a repeat customer is a same-day return, so the figure becomes the shop's same-day return-trip rate. It moves when truck stock or diagnostic quality changes, and it does not move when a customer decides they like you. Nothing about the calculation is wrong. It has simply stopped being about loyalty.

Open the window to the whole life of the business. The numerator can then only ever rise, because a customer can move from one visit to two and can never move back - but the denominator rises with every first-time customer, so the figure drifts up in a quiet year and falls in a year of heavy acquisition. Computed the same way over this shop's eleven years of records it reads about 40 percent, roughly 2,300 of the 5,800 customers ever served having two or more visits, and next year's reading will say more about how many new customers walked in than about whether anybody came back. A number that moves with your intake rather than your retention is not a loyalty measurement.

Lose half your customers and serve the survivors twice. The denominator only contains customers you served inside the window, so a customer who left is not counted as a loss, they are simply absent. A shop that lost half its base and happened to see the remainder twice each would report a repeat rate close to 100 percent and a business in serious trouble. This figure cannot fall because customers left, only because the ones you did see came once, which means it is not a retention measure in either direction and should never be put in a sentence next to the word churn. The cohort return rate above is the one that would have collapsed in that scenario, because its denominator was fixed before the customers left.

References

• See related: Contract Attach Rate Is Measured Against Every Customer You Ever Had, which sets the 50-customer floor used above
• See related: The Difference Between a Repeat Customer and a Loyal One
• See related: First-Time to Repeat Customer Conversion, for the work that produces a second visit
• See related: Recurring Revenue Share and the Band Worth Holding