Tracking performance in a service business: the five numbers I actually watch

There is a genre of business content I think of as dashboard theatre: screenshots of analytics suites glowing like mission control, implying that whoever owns them must be running a tight ship. In my experience the correlation runs the other way. The tightest operations I have seen ran on embarrassingly few numbers, watched obsessively; the loosest had dashboards nobody had logged into since the demo. In my maintenance business I watch five numbers. Here they are, why these five, and — just as important — how each one tries to lie to you.

The five

1. Response time. Not job completion time — the gap between a customer's enquiry and a human answering it. Whether the enquiry arrives by phone, WhatsApp or web form, the clock starts when they reach out and stops when a person (not an auto-reply) engages. Customers form their opinion of you in that gap, before anyone has picked up a tool. It is also the number most completely within your control: traffic is not your fault, but a message sitting unanswered all afternoon is.

2. First-time-fix rate. The share of jobs resolved on the first visit. This is the metric the customer actually feels, and internally it is a compound score of everything upstream: phone triage, dispatch matching, van stock. When it drifts down, the interesting question is never “which technician” but “which process” — are we diagnosing badly on the phone, sending the wrong skills, or carrying the wrong parts?

3. Callback rate. The share of jobs we have to return to because the fix did not hold. This is the true quality number, and it hides in politeness: many customers will not complain about a repair that half-worked, they will simply call someone else next time. So a callback that does surface represents customers who bothered — treat each one as several. A low callback rate earned honestly is the closest thing a service business has to a quality certificate.

4. Technician utilisation. Of a technician's paid day, how much is wrench time — actually working on jobs — versus drive time and dead time? In a city as spread out as Dubai, routing is the silent tax on the whole business; the same headcount can deliver very different capacity depending on how intelligently the day is sequenced. This is the economics number: the first three are about whether customers love you, this one is about whether you can afford to keep them loved.

5. Review velocity and score. How many reviews arrive per month, and what they average. I have written elsewhere that reviews are the P&L in disguise; here I will just note their role in the set: they are the lagging echo of the other four. Fix response time, first-time fix, callbacks and routing, and the reviews improve a month or two later without being asked. If the four leading numbers look fine but reviews are sagging, one of your four is being gamed — which brings us to the last section.

Why five and not fifty

Because a number you do not act on weekly is decoration. That is the whole argument. Every metric on the list must survive two questions: who owns it — a named person, not a department — and what threshold triggers a conversation. If callback rate crosses its line, a specific conversation happens that week, with a specific person, about specific jobs. No owner or no threshold means the metric is a spectator, and spectators multiply: dashboards accrete metrics the way lofts accrete boxes, each added for a reason nobody remembers, none ever removed.

Five numbers, each with an owner and a tripwire, will outperform fifty that are merely admired. The discipline is not in the measuring; it is in the refusing to measure.

The tooling truth

Here is the confession the software industry would rather I skipped: a shared spreadsheet, honestly maintained, beats an abandoned analytics suite by a distance. The value is not in the tool; it is in the honesty and the cadence. A sheet that five people update and one person interrogates every Monday is a performance system. A beautiful platform that nobody trusts or opens is furniture with a subscription fee.

Software earns its place only when the sheet starts failing — when the volume genuinely outgrows manual entry, when you catch transcription errors, when the Monday conversation stalls on “whose figure is right?” Those are real signals and worth acting on. But buy software to solve a failure you have actually observed, not to feel like the kind of company that has dashboards. The sheet first. The suite when the sheet breaks.

Every metric invites its own cheat

Now the uncomfortable part. The moment a number has consequences, it invites gaming — not necessarily through dishonesty, often just through people optimising exactly what you told them to optimise. Utilisation under pressure invites rushed jobs: wrench time looks magnificent, and the botched work comes back as callbacks. Response time invites the hollow first-touch: fire off a “we've received your enquiry” and let the customer wait anyway. First-time fix invites optimistic job closing: mark it resolved, and if the customer rings back, log it as a fresh job. Even review velocity can be gamed, by steering asks towards the obviously delighted.

None of this means the people are bad. It means the metric is doing what metrics do: shaping behaviour, including behaviour you did not intend. Assume the gaming; design for it.

Pairs keep each other honest

The design answer is to watch the numbers in pairs, so each metric's cheat shows up in its partner. Utilisation pairs with callback rate: rush the jobs to juice the first and the second rises to confess. Response time pairs with review sentiment: hollow auto-replies flatter the clock, but customers describe the actual experience in writing. First-time fix pairs with repeat-contact patterns: quietly reopened jobs betray the optimistic closer. Read any number alone and it will tell you what you want to hear; read it beside its partner and the two triangulate something close to the truth.

That, honestly, is the entire system. Five numbers, each owned, each with a tripwire, each shadowed by the partner that catches its favourite lie, reviewed every week by people empowered to change things. It fits on one sheet and it is duller than mission control. But the customer at the door does not experience your dashboard. They experience whether the phone was answered, whether the fix held, and whether anyone would answer for it if it did not — and those, not the glow of the monitors, are what the five numbers exist to protect.