How to actually reduce callbacks (and measure first-time fix)

Every contractor knows callbacks cost money. Far fewer can tell you which callbacks, caused by what, on whose jobs — and without that, the usual response is a toolbox talk about "being thorough", which changes nothing because it does not address a specific cause.

The work below is unglamorous: define the term, record the cause, then fix causes in order of frequency. It is also the only version of this that produces a number that moves.

First, define it so two people count the same thing

Ask three managers what a callback is and you will get three answers. Until the definition is written down, the metric is noise. A workable one:

Every part of that carries weight. Same equipment excludes the second system in the house. Same complaint excludes an unrelated failure two weeks later. A set window — 30 days is common, some shops use 90 for installs — stops the definition sliding. And at no charge is what separates a callback from a legitimate new job, because if the customer paid, the market has told you it was new work.

Write it in one sentence, put it where dispatch can see it, and make the flagging decision belong to one person rather than to whoever books the call.

Second, record the cause — this is the part most shops skip

A count tells you that you have a problem. A cause code tells you which problem. Keep the list short enough that a dispatcher will actually use it under pressure — six or seven codes, not twenty:

• Misdiagnosis — the fault found on the first visit was not the actual fault.
• Incomplete repair — right diagnosis, work not finished or not verified.
• Part failure — the part fitted was defective or the wrong part for the unit.
• Commissioning — charge, airflow, settings or programming left wrong.
• Second fault — a real separate failure that surfaced after the visit.
• Customer expectation — nothing was broken; the outcome was not what they expected.

That last one matters more than it looks. A meaningful share of "callbacks" at most shops are communication failures, not technical ones, and they need a completely different fix — better job-completion explanations, not more training on refrigerant circuits. Lumping them in with misdiagnosis hides both problems.

Third, fix causes in the order they occur

Once you have two or three months of coded data, the ranking is usually a surprise. Attack from the top.

Misdiagnosis

This concentrates in two places: newer technicians, and equipment the shop rarely sees. Both are answerable with access rather than with training — the tech usually knows how to diagnose, they just do not have the fault table, the sequence of operation, or the note from the last visit in front of them.

Check how long it takes one of your techs to find the fault-code table for a unit they have never worked on. If the honest answer is fifteen minutes of phone browsing on a roof in August, they will guess instead, and the guess is your callback.

Commissioning

The classic version is charging by feel — "beer can cold" — instead of by superheat and subcooling. Trainers in this industry have been saying so for decades and it persists, because measuring takes longer and nobody checks.

The fix is a verification step that produces a number, and a requirement that the number gets written on the ticket. Not because the office will audit every one, but because a value that has to be recorded is a value that has to be taken.

Part failure and incomplete repair

These are usually inventory and process problems rather than skill problems. A tech who leaves without finishing generally left because the part was wrong or missing. Track which parts cause return trips and the pattern is normally a small number of SKUs — often because the model number was misread from a nameplate, which is a fixable data problem, not a competence problem.

What to measure alongside it

Callback rate on its own can be gamed — a shop can improve it by taking fewer difficult jobs. Pair it with first-time fix rate: the share of jobs closed without any return visit. Definitions vary; the useful thing is not the industry benchmark but your own trend line, measured the same way each month.

Two more worth watching, because they usually move before the callback number does:

• Callbacks by technician, normalised for job count and job type — the raw count unfairly punishes whoever gets the hard work.
• Time from dispatch to diagnosis. When this rises, techs are hunting for information, and misdiagnosis tends to follow.

What not to do

Do not publish a callback leaderboard. The reliable result is under-reporting: techs stop flagging returns, dispatch reclassifies them as new jobs, and the number improves while the problem does not. You lose the data you needed and the trust required to get it back.

Do not attach pay to it either, for the same reason and faster.