The Seasonal Customer Who Disappears and Returns

Why this matters

A seasonal customer is absent for most of the year by design, and any lapse rule that counts months since the last job will misread them twice: it flags them as gone when they are simply between seasons, and it clears them as current when they have genuinely left. The second error is the expensive one, because a shop that thinks a customer is fine does not call, and the customer quietly gets the work done by whoever answered their search.

If a meaningful share of your book is seasonal, then "months since last job" is not a lapse signal at all in that segment. It is a measurement of what time of year you ran the report. This card is about the clock those customers actually run on and how to read absence against it.

Two clocks, and putting each customer on the right one

The interval clock measures months since the last completed job and compares them against that customer's own median gap. It works for demand that is failure-driven or wear-driven, where the trigger is elapsed running time rather than a date.

The season clock measures position in the calendar year and compares it against the band of weeks that customer has historically used. It works for demand that is anchored to weather, to a use season, or to an annual event: pre-season readiness, post-season shutdown, an opening or closing, an annual inspection tied to a renewal date.

The test for which clock a customer belongs on is not what trade you are in. It is whether their own job dates cluster in the calendar. Convert each of their completed jobs to a week of the year and look at the spread. If their jobs sit inside a band of a few weeks year after year, they are on the season clock regardless of what the rest of your book does. If their jobs are scattered across the calendar, they are on the interval clock even if your trade is famously seasonal.

Most shops have both, mixed together in one list, running under a single lapse rule. That single rule is the problem.

The four shapes of seasonal absence

Only one of these is a loss, and they look identical in a months-since column.

Between seasons. Absent for most of a year, returns in their band. Not lapsed, not at risk, and should not appear on any call list before their band opens.

Missed a season. Their band opened and closed with no job. This is the real warning, and it is equivalent to a full missed cycle on the interval clock, not to a couple of late months. One missed season is the moment to act; waiting for a second is waiting a full year.

Season shifted. They still call annually but the band moved, usually because their own schedule changed, a property changed hands, or a budget cycle moved. Not a loss, but your anchor is now wrong and the reminder will land at the wrong time forever unless you re-derive the band.

Property seasonally vacant. A second property, a seasonal business, a rental let only part of the year. Their absence is occupancy, not preference, and the relevant date is when the property is occupied again, which may be knowable from the customer directly rather than inferred from job history.

Building the season-anchored due window

1. Convert every completed job for that customer to a week of the year.
2. Set aside off-season jobs that were clearly reactive: an emergency, a failure, a one-off unrelated to the annual work. Keep them in the record; exclude them from the anchor.
3. Take the median week of the remaining jobs. That is the anchor.
4. Take the spread of those weeks. The band is the anchor plus and minus the spread, with a minimum of about two weeks either side because customers do not schedule to the day.
5. Write the band and the anchor to the record, along with the count of seasonal jobs it was derived from. Two jobs is a guess. Four or more is a band you can act on.

Step 2 is the one people skip and it is the whole trick. An off-season emergency is not evidence about when this customer wants their annual work, and letting it into the median drags the anchor toward the middle of the calendar, which is precisely the time of year they never call.

Worked example: seven years, one emergency, one silent loss

A customer with seven completed jobs. Converted to week of the year, the annual work landed in weeks 16, 18, 15, 17, and 16 across five consecutive years. In the fifth year there was also a job in week 43, a failure call in the opposite half of the calendar.

Excluding the week 43 emergency, the five seasonal jobs sit at weeks 15 through 18. The median is week 16, the spread is roughly two weeks either side, so the band is weeks 14 through 20. That is the customer's true annual window, derived from five events, which is enough to act on.

Now watch the failure. The shop's lapse rule is a flat 14 months since the last completed job. The customer's most recent job is the week 43 emergency in year five.

At week 16 of year six, the moment this customer is due, the months-since-last-job figure runs from year five week 43 to year six week 16. That is 9 weeks to year end plus 16 weeks, or 25 weeks, which is under 6 months. The 14-month rule reports them as recently serviced. No reminder goes out. The customer, having not heard from anyone, books their annual work with whoever came up first when they looked.

The shop does not find out. At week 20 of year six the band closes with no job, and nothing in the system marks that as an event because nothing in the system knows the band exists.

The flat rule finally fires much later. From year five week 43 to year seven week 30 is 9 weeks plus 52 weeks plus 30 weeks, which is 91 weeks, or about 21 months. So the customer surfaces on a lapsed list roughly 21 months after their last job, and roughly 15 months after the season they were actually lost in. By then a competitor has done the work twice.

What the season clock would have done instead. At week 14 of year six the record enters its band and becomes due. A reminder goes out. If the band closes at week 20 with no booking, the record flips to missed-a-season on that date, and it lands at the top of the call list in week 21 of year six, not in year seven. The gap between the two approaches, in this one example, is about 15 months of not knowing, and it is entirely produced by an emergency call resetting a clock that was never measuring the right thing.

The record change that prevents it. Keep two dates, not one: last completed job of any type, and last completed job of the recurring seasonal type. The season clock reads only the second. The first is still useful for equipment history and for the interval clock, but it must not be allowed to satisfy a seasonal due window.

Why your lapse rate depends on what month you ran the report

Take a book of 260 repeat customers where 96 are on a spring band. Run the lapse report in early spring, before the band opens: those 96 are roughly eleven months out from last year's work, so a flat rule flags most of them and the shop reports a large lapsed population. Run the same report on the same customers in early summer, after the band has closed: the ones who came back are days old and none of them flag.

Same 260 customers, same behavior, two very different lapse numbers, entirely because of the report date. Any shop tracking a lapse rate month over month across a seasonal book is watching a sawtooth and interpreting it as performance.

The fix is a third state. A seasonal customer is pending from the start of their band until it closes plus a short grace, and pending records are excluded from both the active and the lapsed counts. They resolve to one or the other when the band closes, which means each seasonal customer contributes to your lapse rate exactly once a year, at a date determined by them rather than by you.

Without a pending state, the honest alternative is to report the seasonal segment separately and only at one fixed point per year, after every band has closed.

What changes the answer

• The customer has fewer than three seasonal jobs on record. You do not have a band, you have a guess. Use the shop-wide band for that service type and mark confidence low until a third data point arrives.
• The band is wide rather than tight. A customer whose jobs land anywhere across a three-month stretch is not tightly seasonal; treat them on the interval clock with a seasonal preference noted, and do not build a narrow due window they were never going to hit.
• Weather ran unusually early or late. Whole-band shifts across your entire seasonal segment in one year are a weather fact, not a retention fact. Check the segment before concluding anything about an individual customer, and do not re-anchor a band on one anomalous year.
• A property manager or commercial account. The band is usually a budget or contract cycle rather than weather, which makes it more precise, and it moves when their fiscal year or contract date moves rather than drifting. Ask rather than infer.
• Seasonal vacancy. If the property is unoccupied part of the year, the band is set by occupancy and the customer can tell you the dates directly. One question during the closing visit beats three years of inference.

How to verify you got this right

• Pull five seasonal records and confirm the anchor was derived with reactive off-season jobs excluded. If an emergency is in the median, the band is wrong.
• Confirm that a record's seasonal due window is driven by the last seasonal job date and cannot be satisfied by an unrelated off-season job. This is the single failure that costs whole customers.
• Check that your lapse reporting either has a pending state or reports the seasonal segment separately at one fixed annual point. If neither, your lapse trend is a calendar artifact.
• Confirm that a band closing with no job creates a dated event on the record. If a missed season leaves no trace, nobody will act on it until the flat rule fires a year later.
• Count how many of your seasonal records were derived from four or more seasonal jobs. If most were built on two, your bands are guesses and should be labeled that way rather than trusted.

References

• See related: How to Use Service History to Predict the Next Call
• See related: The Dormant Customer Definition Worth Setting
• See related: How to Track Whether a Customer Actually Came Back
• Trade-standard practice for pre-season and post-season planned service scheduling