The Repeat Visit Rate and the Seven Day Window It Uses

Why this matters

The repeat visit rate is the closest thing most shops have to a rework number they can read without opening a file. It is also a proxy, not a measurement: it counts second visits, and a second visit is only sometimes rework. The blind spot is the lookahead. A week-long window catches the comeback that happens while the customer is still annoyed, and misses the one that happens after the fault has had a chance to come back on its own schedule. That is not a random miss. It falls hardest on intermittent faults, which are the ones your best tech loses sleep over.

What the number is actually made of

Take the jobs completed inside your reporting window. For each one, ask whether the same customer - and, where you recorded one, the same piece of equipment - had another completed job within roughly seven days afterwards. The share of completed jobs that did is the rate.

Four details in that sentence decide everything the number can tell you.

The denominator is completed jobs, not customers. A customer with four visits in the quarter contributes four chances to be a repeat, not one. That is correct for a rework proxy and it is the opposite of a customer-loyalty measure, which counts each customer once. See related: Repeat Customer Rate Moves With the Window You Chose.

The anchor is the completion date of the first job, and the lookahead runs forward from it. A job completed on the last day of the quarter has almost no window inside the quarter, so the most recent week of work always looks cleaner than it is. Read the rate on a closed period, never on a period that is still running.

Both jobs have to be completed. A customer who called back on day four and could not be fitted in until day eleven produces no pair, because the second completion lands outside the window. The dispatcher's calendar decided that, not the technician's work.

The pair has to be two jobs. If the tech went back the next morning and logged it against the original job rather than creating a new one, there is one completed job and no pair at all. Shops that handle comebacks informally show almost no repeat visits for exactly that reason.

The blind spot, stated plainly

A seven-day window is a filter on fault physics. It catches anything that fails immediately: a connection that was never tight, a part fitted the wrong way round, a setting left where the tech put it to test something. It misses anything that needs a condition to come back before it can show itself - a load, a temperature, a duty cycle, a full tank, a weekend of use.

So the number is most accurate on the faults that are easiest to diagnose and least accurate on the faults that are hardest. That is the wrong way round, and it is the single most useful thing to know about it. A shop reading 5 percent of completed jobs and relaxing is relaxing about the half of its rework it has not looked at.

What a longer window catches, and what it wrongly lets in

Stretching the lookahead is not free. Every extra day adds real comebacks and also adds second visits that are not comebacks at all, and the second group grows faster than the first.

A measurement window and an attribution window are different tools and they should not carry the same number. This card is about a standing measurement: a short window keeps the figure a rework proxy you can read every month without classifying anything by hand, which is why it tightens toward seven days and why it stops being a rework read much past thirty. Costing a callback is the other question, and it wants the opposite: the costing card in the references defaults to 90 days, or the labor warranty if that is longer, precisely because a workmanship failure that shows up under load or a season change is still caused by that job even though no monthly rate should count it. Run both, name which one a figure came from whenever you quote it, and never compare one shop's 30-day rate to another shop's 90-day one.

Thirty days is the practical ceiling for a rework read, and it only works if somebody classifies the additions by hand rather than trusting the count.

Two kinds of shop should move that ceiling in. If most of your work sits under maintenance agreements, planned visits are dense enough that a 30-day recount is mostly contract scheduling wearing a comeback's clothes, and 14 days will give you a cleaner signal for the same classification effort. If you run multi-stage installs, where a single sold job is genuinely three visits, exclude that job type from the measure entirely and say so, rather than letting the classifier strike the same jobs every quarter. A measure that needs the same manual exception applied every period should have that exception written into its definition.

Anchoring on the call instead of the completion

There is a second definition worth knowing about, because it removes your dispatcher from your quality number. Instead of asking whether a second job was completed within seven days, ask whether the customer made contact within seven days - a call, a message, a portal request - about the same work.

The difference matters most in your busiest weeks, which is exactly when you want the number to be honest. In a slow month a comeback gets scheduled the next morning and both definitions agree. In a peak week the same comeback waits nine days for a slot, and the completion-anchored figure quietly improves while the customer sits with a fault. A shop that reads the completion-anchored number all summer can watch it fall through the busy season and conclude the crew tightened up.

The cost of the call-anchored version is that it needs the contact recorded and linked to the original job, which is a habit rather than a report. If you cannot get that habit, the workaround is to keep the completion-anchored number and read it alongside your average days from call to visit. When the second one climbs, the first one is understating by roughly the amount it climbed.

The equipment link, and what happens without one

When both jobs carry the same equipment record, the pair is tight: the same unit, twice, in a week. When they do not, the match collapses to customer-only, and a customer is not a machine. A landlord with nine doors generates unrelated calls constantly, and two of them in the same week read as a comeback.

If your records do not carry equipment on the job, you have two honest options. Run the number customer-only and treat every multi-unit or multi-property account as a known false positive to be struck by hand. Or restrict the whole measure to single-property residential accounts and say so on the face of the number. What you cannot do is quote a customer-only figure to a technician as their rework rate, because the properties with the most doors are usually the ones assigned to your most senior people.

Worked example: one quarter, recounted

A shop closes a quarter with 420 completed jobs, about 32 a week.

The standing figure. 21 of the 420 completed jobs had a second completed job for the same customer within seven days: 5.0 percent of completed jobs in the quarter.

Correction one, false positives from customer-only matching. Six of the 21 pairs sit at multi-unit properties. Reading the job notes, two of those six are different units, so they are two unrelated jobs and not a comeback. Corrected seven-day count: 19 of 420 completed jobs, 4.5 percent.

Correction two, the window. Re-running the same 420 completed jobs at a 30-day lookahead produces 25 additional pairs. Classified by hand: 9 are different work at the same property (three of them a different unit again), 4 are planned second visits for parts on order, and 12 are genuine comebacks on the original complaint.

Corrected 30-day rework: 19 plus 12 is 31 of 420 completed jobs, 7.4 percent. The standing figure of 5.0 percent of the same 420 completed jobs saw 19 of the 31 real comebacks, about 61 percent of them.

Where the missing 12 sat. Eight of the 12 late comebacks came from calls the office had logged as intermittent, meaning the customer described a fault that was not happening when they called. Those calls were 54 of the quarter's 420 completed jobs.

• Intermittent calls: 3 of 54 at seven days, 5.6 percent of that category's completed jobs. True at 30 days, 11 of 54, 20.4 percent of the same 54.
• Everything else: 16 of 366 at seven days, 4.4 percent of that category's completed jobs. True at 30 days, 20 of 366, 5.5 percent of the same 366.

On the standing number the two categories are 1.2 points apart and nobody argues about it. Recounted, the intermittent category's true rate is a little under four times the rest of the book's true rate, each measured against its own completed-job base. The gap the short window hid is 14.8 points on the 54 intermittent jobs and 1.1 points on the other 366.

The floor this read depends on. Do not rank a category off fewer than about 30 completed jobs in the window. At 30, a single pair moves the rate by more than three points, which is larger than most of the differences you would be ranking on. Both categories above clear it; a category running eight jobs a quarter gets rolled into a year before anyone reads it as a rank.

The quarterly hand cut worth running

Keep the seven-day figure as the standing number. It is fast, it is consistent with its own history, and a sudden move in it is a real signal. Then, once a quarter, on the last closed quarter only, run the same population at 30 days and classify every additional pair into three buckets: genuine comeback, different work, planned second visit.

At the volume above that is 25 pairs to read, which is well under half a day for one person with the job notes open. Record the classified 30-day figure next to the seven-day one so the pair builds its own history. Two quarters in, the ratio between them becomes the interesting number: if it drifts upward, your rework is moving later, which usually means it is moving toward the diagnostic end of the book rather than the workmanship end.

What the standing number can and cannot see

The last row is the proxy's hard floor and no window length fixes it. A rework measure built from your own completed jobs can only see the customers who gave you another chance. That is why the repeat visit rate is read alongside complaints and survey returns rather than instead of them, and why a flat repeat rate sitting next to a falling satisfaction score is a finding rather than a contradiction.

References

• See related: Callback Rate Management, for what counts as a callback and how to set the tracking up in the first place
• See related: How to Cost Callbacks and Rework Honestly, for the 90-day attribution window and why it differs from this one
• See related: Repeat Customer Rate Moves With the Window You Chose, for the loyalty measure that shares this shape and answers a different question
• See related: Mining Your Callbacks for What They Reveal About Your Systems
• See related: The Callback Root Cause Log That Pays for Itself