The Dormant Customer Definition Worth Setting

Why this matters

Every shop has a working definition of a dormant customer, and in most shops it is "I have not heard that name in a while." That definition cannot be delegated, cannot be scheduled, and cannot be measured, so the reactivation work either never happens or happens once a year in a panic on a list nobody trusts.

The cost of getting the definition wrong is not abstract. Set it too long and your highest-frequency customers, who are usually your most profitable ones, are gone before anyone notices they stopped calling. Set it too short and you spend your reactivation hours calling people who are behaving completely normally, which produces a bad response rate, and the bad response rate is then used as evidence that reactivation does not work.

What the definition has to do

Three jobs, and a definition that only does the first one is the one most shops have:

1. Separate silence that means nothing from silence that means something. A customer who calls every 18 months is not dormant at month 14.
2. Fire early enough to act. A flag that trips after the customer has already committed to somebody else is a report, not a trigger.
3. Produce a list small enough to work by hand. If the dormant list is a third of your database, nobody will call it, and the definition has failed regardless of how correct it is.

Those three pull against each other, which is why the answer is a multiple rather than a duration.

The number: use a multiple of their interval, not a fixed number of months

State the rule this way: a customer is dormant when the time since their last completed job exceeds twice their own median interval between jobs, with a shop-wide backstop for customers who do not have enough history to compute one.

Twice is the starting point. It is deliberately concrete, and you should tune it after one cycle rather than debating it beforehand. Here is the full set of defaults worth adopting on day one:

The median interval, not the average. One emergency call dropped into the middle of a planned rhythm pulls an average down hard and makes a customer look more frequent than they will ever actually be, which then makes them look overdue months before they are.

The residential floor has an exception worth stating in the same breath: on commercial accounts running a contracted frequency, the multiple should run tighter than 2.0 and the floor does not apply, because a missed cycle there is a contract-performance problem to raise this month, not a lifecycle signal to watch.

Why a single shop-wide duration fails

The common approach is one number for everybody: dormant at 18 months. Look at what that number means to three different customers.

• A customer whose natural rhythm is 6 months has been silent for three full cycles before the flag fires.
• A customer whose natural rhythm is 12 months has been silent for one and a half cycles, which is a reasonable trigger.
• A customer whose natural rhythm is 24 months gets flagged at 18 months, which is before they are even due. They are behaving normally and your list says they left.

So a single duration is simultaneously too slow for your frequent customers and too fast for your infrequent ones. It generates false alarms exactly where nothing is wrong and stays quiet exactly where something is. Since high-frequency customers tend to be the ones carrying the most annual volume, the errors land in the most expensive place available.

How to compute your own baseline in one afternoon

You need job dates and nothing else. Paper invoice copies work.

1. Pull every customer with three or more completed jobs. Two jobs give you one gap, which is not enough to call a rhythm.
2. For each, list the gaps in months between consecutive jobs, in order.
3. Take the median of that customer's gaps. That is their personal interval.
4. Sort all the personal intervals and look at the shape, not the average. You are looking for whether they cluster into two or three groups, which they almost always do, or spread evenly, which is rarer.
5. Set the backstop from the middle of the distribution, so customers who cannot be computed individually get a defensible default rather than a guessed one.
6. Recompute quarterly. A customer's interval changes as their equipment ages, and it usually shortens.

Customers with one or two completed jobs get the backstop. They are also the population where the definition matters least, because a first-time customer who never came back is a follow-up failure rather than a dormancy case, and it belongs to a different process.

Worked example: setting the line for a mixed-trade shop

A shop has 480 names in its list. Running step 1 above, 210 of them have three or more completed jobs and can have a personal interval computed. The remaining 270 fall to the backstop.

Sorting those 210 personal intervals produces three clear clusters rather than a smooth spread:

• 74 customers around a 6-month interval, the seasonal-service group.
• 88 customers around a 12-month interval, the annual-service group.
• 48 customers in a long tail around 24 months, the break-fix-only group.

74 plus 88 plus 48 is 210. The middle of that sorted list falls inside the 12-month cluster, so 12 months is the shop-wide median interval and a defensible basis for the backstop.

Now test the old rule against the new one. The shop had been using a flat 18-month dormancy line.

• For the 6-month group, 18 months is 3.0 times their interval. Three cycles of silence before anyone looks.
• For the 12-month group, 18 months is 1.5 times their interval. Reasonable, and this is the only group the flat rule ever served properly.
• For the 24-month group, 18 months is 0.75 times their interval. The flag fires while they are still inside their normal cycle.

Under the 2.0 multiple, the same three groups flag at 12 months, 24 months, and 48 months. That last number is why the 36-month hard cap exists: 48 months of silence is not a rhythm, it is a departure, and no multiple should be allowed to argue otherwise.

Here is what the change did to the actual list. The flat 18-month rule produced a dormant list of 96 names. Reading them against personal intervals, 41 of those 96 were inside their own normal cycle, so about 43 percent of that list was a false positive. The 2.0-multiple rule with the 36-month cap produced a list of 58 names, of which 9 sat inside their own interval because of rounding and thin history, about 16 percent false positives.

The new list is roughly 40 percent smaller than the old one and carries about a third as much false-positive share. When the shop worked it, 7 of the 58 booked a job, about 12 percent of the list contacted. The prior year's flat-rule list had booked 5 of 96, about 5 percent of the list contacted. That is more than double the response rate on a list that took about 40 percent less time to work.

Read the cause correctly, because the tempting conclusion is that the new script worked. The script was unchanged. What changed is that a larger share of the people being called had actually stopped calling, so a larger share of them had something to say yes to. The prior list was diluted with 41 people who were not gone, and those people cannot convert on a reactivation call because there is nothing to reactivate.

One more thing the numbers show that is easy to miss. Both lists included the 74-customer seasonal group, but the flat rule reached them at 18 months of silence and the multiple rule reached them at 12. Those 6 months are the most recoverable window in the whole model, because at 12 months the customer still remembers your tech's name.

Writing the definition down

The definition is only real once it exists as a sentence anyone in the office can apply without asking. Something in this shape:

"A customer is dormant when months since their last completed job exceed 2.0 times their median gap between jobs, or 24 months if they have fewer than three completed jobs, capped at 36 months in every case, and never less than 9 months for a residential account. A dormant customer moves to lost only when a reason is recorded: a stated departure, a confirmed move or property sale, or a clear no on a reactivation contact."

Post it where the list gets pulled. A definition that lives in the owner's head gets reinterpreted every time somebody else runs the report.

What would change these numbers

A trade with a genuinely long natural cycle. If most of your work is replacement-driven with a multi-year life, the 36-month cap will flag customers who are simply between systems. Raise the cap, but replace the lost signal with something else, such as an equipment-age list, rather than just extending the silence you tolerate.

A shop that just changed its service mix. If you added a recurring service last year, historical intervals describe the old business. Compute intervals from the last two years only until the new mix has enough history.

A property-driven customer base. Where the customer is a property rather than a person, a sale resets everything. The new owner is a first-time customer at an address you know well, which is a strong position, and treating them as a continuing dormant relationship gets the tone wrong on the first call.

Weather or a hard season. A regionally unusual mild winter or wet summer suppresses call volume across an entire cohort at once. If the dormant list doubles in one quarter and the increase is concentrated in one service type, suspect the season before you suspect the customers.

How to verify you got this right

The dormant list is workable in the hours you have. If one person can work the list in two or three sittings, the definition is sized right. If it needs a project, tighten the multiple or the cap until it does not, and accept that you are trading coverage for actually doing the work.

Sample ten names and read their history. If more than one or two of the ten are clearly inside their own rhythm, the intervals are being computed with averages instead of medians, or emergency calls are being counted as part of the rhythm.

The Slipping bucket is where most of the action is. If almost nobody is being caught at 1.5 times and everyone is arriving already dormant, the intermediate flag is not being read. That flag is worth more than the dormant list itself, because it fires while the relationship is still warm.

Response rate is tracked per list, not per year. Comparing this year's reactivation results to last year's tells you nothing if the definition changed in between. Record the definition used alongside the result.

References

• U.S. Small Business Administration (SBA), customer retention and small business marketing guidance
• Trade-standard practice for service-interval recall and maintenance scheduling programs
• See related: The Customer Lifecycle Stages a Service Shop Actually Has, The Customer Reactivation SOP, Reactivating Dormant Customers, How to Spot a Customer About to Go Quiet