Field Service KPIs, SLAs and Contractor Management

There is a particular meeting I have sat in many times. The monthly performance review with a service provider. The dashboard is green. First-time fix is up, response compliance is at ninety-something percent, PPM completion is on target. And the building owner sitting across the table is unhappy, because the tenants are still complaining, the same air handling unit has been attended four times this quarter, and nobody in the room can reconcile the green dashboard with the lived experience of the estate. That gap is not usually caused by dishonesty. It is caused by a measurement system that was designed to be reportable rather than to be true, and by a contractor governance model that inspects the report instead of the work.

The message up front: every field service KPI creates an incentive, and every incentive gets optimised. You cannot prevent that, and you should stop trying. What you can do is pair each metric with a counterweight that makes the shortcut visible, so that a technician or a contractor improving the number honestly and one improving it dishonestly produce visibly different data patterns. Measurement design is mostly about designing those pairs. Contractor governance is mostly about making sure the data behind them is real.

1. Be honest about what the measurement is for

Before choosing a single metric, answer one question: is this KPI set for managing the work, for governing a contract, or for reporting upward? Those three purposes want different things, and most estates try to serve all three with one dashboard, which is why the dashboard ends up serving none of them well.

Managing the work needs metrics that are fast, granular and operationally actionable. How old is my oldest open job. Which trade is falling behind. Which site has jobs sitting unassigned. These are daily and weekly numbers, and their accuracy matters less than their timeliness, because you are using them to intervene today.

Governing a contract needs metrics that are defensible, defined in writing, auditable and stable over time. Both parties must be able to recompute the number from raw records and get the same answer. If the definition can be argued about at the review meeting, it is not a contract metric, it is an opinion with a decimal point.

Reporting upward needs very few numbers, framed in the language of the business: risk, cost, statutory exposure, tenant experience. A board does not need mean time to respond by priority by site. It needs to know whether the estate is getting safer or less safe, and whether the money is buying anything.

The practical recommendation I make is to build one measurement layer, the raw event data in the maintenance system, and then produce three deliberately different views from it. Trying to make a single scorecard do all three jobs is the root cause of the bloated, ignored KPI pack that most estates are carrying.

2. The KPI set that actually drives behaviour

I would keep a core field service set deliberately small. Seven or eight metrics, each chosen because it changes a decision, and each paired with a counterweight. Everything else is diagnostic, pulled when a core metric moves, not reported monthly.

The set I would start from:

• First-time fix rate: the proportion of jobs resolved on the first visit with no return required. The single most contested definition in field service, for reasons worth a section of its own.
• Mean time to respond: elapsed time from job creation (or notification) to a competent person being on site and engaged with the fault.
• Mean time to resolve: elapsed time from job creation to the fault being rectified and the asset returned to service.
• Schedule adherence: the proportion of work executed in the window it was planned for, by the resource it was planned for.
• PPM completion on time: planned preventive tasks completed inside their compliance window, not merely completed.
• Repeat visit rate: jobs on the same asset, for a related symptom, inside a defined window.
• Technician utilisation and wrench time: two different things that get confused constantly.
• Backlog age profile: not the size of the backlog, its age distribution and its trend.

Notice what is not on the list: cost per work order, jobs closed per technician per day, and customer satisfaction score as a standalone. All three are useful diagnostics and all three are badly behaved as headline KPIs, because each one has a trivially available shortcut that degrades service while improving the number. I will come back to them.

3. What each metric tells you, how it is gamed, and what to pair it with

This table is the heart of the article. Print it, take it into your next KPI design workshop, and refuse to accept any metric into the core set until you can fill in the fourth column.

The pattern across all eight rows is the same. The counterweight is almost never another headline metric. It is a structural check: a frozen snapshot, an audit trail, an evidence requirement, a second clock. Add the check and you do not need to accuse anyone of anything, because the data stops supporting the shortcut.

4. First-time fix: the metric everyone defines differently

First-time fix deserves its own section because it is simultaneously the most useful field service metric and the most inconsistently defined. I have seen four definitions in use, often within one organisation, and they produce materially different numbers from the same underlying work.

• One visit, fault rectified. The strictest reading. The technician attended once and the fault was fixed. Anything requiring a return, for any reason, fails.
• One visit, excluding parts-related returns. The most common contractor-preferred definition. If a part had to be ordered, the job is excluded from the denominator, on the argument that parts availability is not within the technician's control.
• One visit, fault made safe or asset made operational. A softer reading that counts a temporary restoration as a fix. Popular where the operational priority is availability rather than a permanent repair.
• One technician, any number of visits inside the SLA. A definition that is really measuring ownership and SLA compliance, not first-time fix at all, but frequently labelled as such.

None of these is wrong in the abstract. What is wrong is not writing down which one you mean, because the gap between the strictest and the loosest reading on the same estate can be large enough to swing a performance judgement entirely. The exact size of that gap varies enormously by estate, asset mix and how much of the work is specialist, so I would not quote a typical figure; what matters is that you measure both the strict and the excluded-parts version on your own data and see the spread for yourself.

My recommendation: define first-time fix strictly, and report the exclusions separately rather than removing them from the denominator. So the headline is the strict number, and beneath it sit the reasons for failure: part not available, wrong trade attended, access denied, further diagnosis required, permit not in place, specialist required. That breakdown is far more actionable than a single flattering percentage, because each reason points at a different fix. Part not available is a stores and van-stock problem. Wrong trade attended is a triage and coding problem. Access denied is a coordination problem. Permit not in place is a process sequencing problem, which is where the permit to work integration discipline earns its place.

5. Respond versus resolve, and why the pair matters more than either

Response time is the metric that gets written into contracts because it is easy to define and easy to police. Resolution time is the metric that tenants and operations actually experience. Manage only response and you build a service that is excellent at arriving and mediocre at finishing.

The behaviour this produces is recognisable. Technicians are pulled off productive work to make a compliance attendance on a new high-priority job, stop the clock, assess, and leave. The response SLA is met on both jobs and neither is progressing. Multiply that across a busy estate and you get a service provider hitting every response target while the backlog quietly grows, which is precisely the green-dashboard-unhappy-client meeting I described at the top.

The corrective measures I would build in:

• Never report response without resolution alongside it, at the same priority granularity. The pair is the unit of measurement, not either one alone.
• Measure time from response to first substantive activity. If the average gap between "attended" and "work started" is material, the response clock is being stopped by arrival rather than by engagement.
• Report resolution as a distribution, not a mean. A mean hides the tail, and the tail is where the reputational damage lives. The ninetieth percentile and the count of jobs over a threshold are more honest than an average that a handful of fast jobs can flatter.
• Track the percentage of jobs that met response but missed resolution. This single derived figure exposes the respond-and-run pattern better than anything else I have used.

For how resolution performance ties back into the age and composition of what is outstanding, the maintenance backlog and downtime tracking discussion is the companion piece, and the underlying reliability arithmetic sits in MTBF, MTTR and availability.

6. SLA design: the decisions that determine whether the numbers mean anything

An SLA is a measurement specification disguised as a commercial document. Most of the disputes I have seen in service contracts are not disagreements about performance, they are disagreements about definition that surfaced eleven months into the term. The decisions below need to be made explicitly and written down before the contract starts.

Response versus rectification. Decide whether each priority level carries a response target only, or a response target and a rectification target. My strong preference is both, with the rectification target set realistically rather than aspirationally, because an unachievable rectification target simply teaches everyone to ignore it.

Business hours versus clock hours. A four-hour target means very different things if the clock runs continuously or only during the working day. Define the calendar precisely: working hours, weekends, public holidays, and which priorities run on a twenty-four hour clock regardless. In the Gulf this needs extra care around Friday arrangements, Ramadan hours and a public holiday calendar that is partly announced at short notice. Write the rule, do not assume it.

The priority matrix. Priority should be derived, not chosen. A matrix of impact against urgency, with asset criticality feeding impact, gives a repeatable answer and removes the argument at the point of logging. Free-text priority selection by whoever raises the job is the single largest source of SLA noise I encounter, because everything urgent to the person reporting it becomes priority one.

Clock stops. This is the hard one and it gets its own section below.

Credits and penalties. Decide what failure actually costs, and make sure the mechanism is proportionate to the harm rather than to the ease of measurement. The commercial structure of the agreement itself, how the service charge is built and how penalties interact with the annual maintenance contract, belongs to the contract discussion rather than here.

For the detailed mechanics of building the matrix itself, including priority bands and target derivation, see the SLA matrix design walkthrough, and for how priority interacts with job classification, work order types in a CMMS.

7. The clock-stop problem

Clock stops exist for a genuine reason. If a technician cannot enter a tenant's server room until Tuesday, or a compressor part has a three-week lead time from Europe, it is not reasonable to hold the service provider's resolution clock running through a delay they do not control. Every mature SLA therefore has hold states: awaiting access, awaiting parts, awaiting client decision, awaiting permit, awaiting third party.

And every clock stop is also the most effective SLA compliance tool available to a contractor who wants one. I have reviewed contract performance data where the reported resolution figures were excellent and the total elapsed time from report to genuine fix was multiples of the target, with the difference sitting entirely in hold states that nobody audited.

What I would insist on:

• Report gross and net elapsed time side by side. Net is the contractual number. Gross is the tenant's experience. Both belong on the report, always.
• Every hold requires a reason code and evidence. Awaiting access needs a named contact and a timestamped attempt. Awaiting parts needs a purchase order or supplier reference. A hold with no evidence is not a hold.
• Holds require client acknowledgement to count. A hold the contractor applies unilaterally and the client never sees should not stop a contractual clock. Making acknowledgement a condition changes the behaviour immediately.
• Cap the number of hold cycles per job. A job that has been on and off hold five times is not a hold problem, it is a management problem, and it should escalate automatically.
• Report hold time by reason as a standing metric. If awaiting parts dominates, the answer is a stores and van-stock review, not an SLA renegotiation. The hold breakdown is genuinely one of the most useful diagnostic reports in field service, because it points at the constraint rather than the symptom.

Treated this way, clock stops stop being a loophole and become a diagnostic. That is the reframe I would push for in any contract review: the hold register is not an excuse log, it is the list of things stopping your estate from being fixed, and most of those things are within the client's control rather than the contractor's.

8. Utilisation, wrench time and the measurement that gets confused

Utilisation and wrench time are routinely used interchangeably and they measure almost opposite things.

Utilisation is the proportion of available paid hours that are allocated to work. It is a planning and commercial metric. It tells you whether you are carrying more labour than you have work for, and it is the number a contractor watches because it drives their margin.

Wrench time is the proportion of a technician's day actually spent working on an asset, as distinct from travelling, waiting for access, collecting parts, finding drawings, obtaining permits, and completing paperwork. It is a productivity and process metric, and it is the number the client should care about, because everything it excludes is waste the client is paying for.

High utilisation with low wrench time is the classic picture of an over-allocated, under-supported field team. The technicians are busy all day and very little of the busyness is repair. The fix is almost never more technicians; it is removing the friction. Better job information at dispatch. Parts staged before attendance. Permits raised in advance. Access coordinated by someone other than the technician at the door. Drawings and asset history available on the mobile device rather than in a filing cabinet, which is one of the more concrete arguments for field-ready mobile maintenance apps.

A caution on measuring wrench time: doing it properly requires either time-stamped mobile activity capture or a sampled observational study, and both have costs. Self-reported wrench time is worthless, because the person reporting it knows what the number is for. If you cannot measure it credibly, measure its proxies instead: travel time per job, jobs per attendance, and the reason breakdown on first-time fix failures. Those three together will tell you where the day is going without pretending to a precision you have not earned.

9. Contractor management: pre-qualification through to scorecard

Field service performance is increasingly delivered by people who do not work for you. That changes the management problem from supervision to governance, and governance has a lifecycle.

Pre-qualification. Before a contractor touches an asset, you need trade licence and registration, insurance with adequate limits and correct interests noted, relevant certifications for the work scope, financial standing sufficient for the contract size, references you actually call, and safety performance history. This is administrative work and it is where most estates are weakest, because pre-qualification tends to happen once at onboarding and then decays.

Competence and currency records. Individual competence matters more than corporate accreditation for field work. For each person who will attend, you want the specific competencies the work requires, and critically, their expiry dates. A contractor with a valid company certification and three technicians whose high-voltage authorisations lapsed last quarter is non-compliant in the way that actually matters. Expiry tracking with automated warning is the single highest-value piece of contractor data management, and it is also the most commonly absent.

Induction and permit compliance. Site induction, site-specific rules, emergency procedures, and the permit regime. The governance question is not whether inductions happened but whether the system can refuse a job assignment to an uninducted person. If it cannot, the induction record is documentation rather than control.

Ongoing performance scorecards. Covered below.

Review and offboarding. A periodic review with consequences, and a clean exit process: access revoked, keys and passes returned, asset documentation handed over, outstanding work transferred with history intact. The handover of work history is the one that gets skipped, and it is the reason many estates lose all record of what was done to an asset every time a contractor changes.

10. Managing one contractor versus managing a panel

These are different disciplines and conflating them causes real problems.

With a single total facilities management provider, your leverage is contractual and your main risk is opacity. The provider controls the data, the reporting, the classification and often the system the work is recorded in. You are measuring performance using numbers produced by the party being measured. The governance emphasis therefore falls on definition control, audit rights, access to raw records rather than summaries, and independent verification by sampling. I would rather have the right to pull a hundred raw job records at random than a beautifully formatted monthly pack.

With a panel of specialist contractors, your leverage is comparative and your main risk is inconsistency. You can benchmark lift contractors against each other, and the discipline of a shared scorecard does most of the work for you. But you now own the integration problem: multiple data formats, multiple definitions, multiple systems, and the boundary disputes that appear whenever a fault could belong to two trades. The governance emphasis falls on standardisation: one job taxonomy, one priority matrix, one definition of first-time fix, one asset register that everybody references, and mandated submission into your system rather than theirs.

The panel model also creates the interface problem that nobody budgets for. A chilled water fault could be the chiller contractor, the pump contractor, the controls contractor or the water treatment contractor, and while they establish which, the tenant is warm. Whoever owns the estate has to own triage, and triage cannot be outsourced to the parties whose scope is in question. In practice this means keeping a small in-house or managing-agent capability whose job is diagnosis and allocation, and resourcing it properly, because it is the thing that makes a panel work.

11. A contractor scorecard with weightings

A scorecard has to be small enough to be maintained and weighted in a way that reflects what you actually care about. The weightings below are a starting point I would adjust per contract: a statutory-heavy fire and life safety contract should push compliance weight up, a tenant-facing reactive contract should push responsiveness and experience up.

Three rules that make the difference between a scorecard that changes behaviour and one that generates paperwork:

• Some dimensions are gates, not scores. Competence currency and permit compliance should not be tradeable against a good first-time fix rate. Below a defined floor on safety and compliance, the overall score is capped regardless of performance elsewhere.
• Score from your data, not their submission. Every metric above should be computable from records in your system. Anything that can only be supplied by the contractor as a summary is an assertion, and assertions do not belong in a score.
• Attach a consequence. A score with no effect on work allocation, contract extension or commercial position will be treated exactly as seriously as it deserves. On a panel, allocating a share of discretionary work by score is the cleanest consequence available, because it is proportionate and it needs no dispute process.

If you are building this alongside a broader performance framework, the FM KPI framework covers the estate-level measurement theory this scorecard sits inside, and building a KPI tree is the method for linking a contractor-level metric to something the board recognises.

12. Why contractor data is worse, and what to do about it

Contractor-generated maintenance data is reliably poorer than in-house data, and the reasons are structural rather than a matter of attitude.

• The data is used to judge them. Any field whose value affects a score or a payment will be completed in the way that is most favourable. This is not dishonesty, it is a predictable response to an incentive you designed.
• They work in multiple systems. A contractor serving eight clients uses eight portals with eight taxonomies. Yours is not their primary system, and data entered into a secondary system is always thinner.
• Administrative effort is unpaid. Detailed fault coding, asset referencing and evidence capture take time that the rate does not usually cover, so they get compressed.
• Their staff turn over. Training on your conventions decays continuously, and nobody tells you when the technician who understood your asset numbering left.
• They do not own your asset register. Correct asset referencing depends on knowing the register, and a visiting technician often does not. So jobs land against a location or a parent, and asset-level history, the foundation of repeat visit measurement and most reliability analysis, quietly fails to accumulate.

What actually helps, in rough order of return:

• Reduce what you ask for, then enforce it absolutely. Five mandatory fields that are always correct beat twenty-five that are half-filled. Decide the minimum viable record and make closure impossible without it.
• Make asset identification physical. A scannable tag on the asset that resolves to the right record removes the largest single source of contractor data error. This is unglamorous capital spend with a very high return on data quality.
• Constrain rather than instruct. Pick lists, conditional fields and validation rules outperform training and guidance notes, because they work regardless of who turns up.
• Pay for the administration, explicitly. If evidence capture and coding are contractual deliverables, price them. Expecting unpriced administrative rigour is the reason it does not happen.
• Publish a data quality score back to them. Contractors respond to being measured on data completeness surprisingly well, particularly on a panel where the comparison is visible.
• Sample and verify. Pull a random set of closed jobs each month and check the record against reality: was the asset right, does the evidence support the closure, does the time sequence make sense. A small, visible, consistent sample does more for data integrity than any amount of written procedure.

The wider discipline here is asset master data, and if the register itself is unreliable then no amount of contractor governance will produce usable numbers. Master data management for assets is the prerequisite, not an optional refinement.

13. The reporting cadence that works

Most estates report monthly because that is when the meeting is, and monthly is the wrong frequency for almost everything. A cadence that works separates the operational rhythm from the governance rhythm.

Two details that matter more than the table. First, the daily view should exist whether or not anyone formally reviews it, because a breach you can see coming is a breach you can prevent, and prevention does not require a meeting. Second, the definition review at annual cadence is the item most often omitted and the one that keeps the whole system honest: once a year, go back to the written definitions, check whether practice has drifted from them, and check whether any metric has been optimised to the point of meaninglessness. Metrics wear out. Plan for it.

The idea to walk away with

Field service measurement fails in a predictable way. Somebody picks a set of reasonable-sounding KPIs, writes them into a contract, and then spends years watching the numbers improve while the service does not. The cause is always the same: each metric had a cheap shortcut, nobody built the counterweight, and the data used to judge the performance was produced by the party being judged.

The discipline that fixes it is not more metrics. It is fewer metrics, each with an explicit written definition, each paired with a structural check that makes the shortcut visible, computed from raw records in your own system, reported at a cadence that allows intervention rather than post-mortem, and backed by a contractor governance regime that treats competence currency and asset-level data integrity as non-negotiable gates rather than scoreable dimensions. None of that is sophisticated. All of it is work.

Final thoughts

If I could change one thing about how most estates measure field service, it would be to stop reporting means. Averages are the enemy of field service management, because service failure lives entirely in the tail and a mean is designed to hide tails. Report distributions, percentiles and counts over threshold, and the conversation changes from "we are at ninety-four percent" to "these eleven jobs took more than a week and here is why", which is a conversation that can actually produce a fix.

And if you are standing up a measurement regime from scratch, resist the temptation to start with the dashboard. Start with the definitions, then the data quality, then the counterweights, then the cadence, and build the dashboard last. A dashboard built on undefined metrics and contractor-supplied summaries is not a management tool; it is a reassurance machine, and reassurance is the last thing an underperforming estate needs.

For the standards background behind competence, asset management policy and maintenance specification, ISO , BSI and SFG20 are the canonical starting points for maintenance task definition and asset management frameworks.

Related reading: SLA matrix design for FM operations, FM KPI framework, Maintenance backlog and downtime tracking, MTBF, MTTR and availability, Permit to work integration with a CMMS, Field-ready mobile maintenance apps, Master data management for assets, Building a KPI tree.