TL;DR
Most IT operations metrics measure activity, because activity is what the ticketing system records. Tickets closed, average response time, volume by category. They are easy to produce and they do not answer the question a budget review actually asks, which is whether people can work and what it costs to keep them working.
- Activity metrics measure the IT team. Outcome metrics measure the company.
- Time to a working person beats time to close a ticket.
- Report the median and the tail, never the mean alone.
- Cost per equipped person is the number finance understands.
- Any metric a team can improve without helping anybody will be.
- Four numbers beat a dashboard nobody opens.
Why most IT operations metrics do not survive scrutiny
Ticket metrics exist because ticketing systems produce them. That is the whole origin story.
They describe how busy the team is and how fast it closes things, which is genuinely useful for managing the team. It is not what a budget conversation is about.
A budget review asks whether the money is buying something the business needs. Tickets closed does not answer that, and it has an awkward property: it goes up when things are going badly.
The gaming problem
Any metric that can be improved without helping anybody eventually will be, not through bad faith but through ordinary optimisation.
Average response time improves by auto-replying. Tickets closed improves by splitting one issue into three. Neither change helps a single person work better, and both look like progress.
The test worth applying to any candidate metric: could this improve while the experience of the people we support gets worse? If yes, it is a management metric rather than a reporting one.
Four numbers worth reporting
These resist gaming because improving them requires actually improving something.
| Metric | What it answers | Why it is hard to game |
|---|---|---|
| Time to a working person | How long from request to productive | Spans every team involved |
| Devices not reporting in 30 days | How much of the fleet is out of reach | Comes from the devices, not a queue |
| Recovery rate on leavers | Whether owned hardware comes back | Binary per device, hard to argue with |
| Cost per equipped person | Whether the money is efficient | Finance can verify it independently |
The first is the one to lead with, because it is what everybody outside IT actually experiences. It spans procurement, shipping, configuration and enrolment, which is also why no single team owns it and why it rarely gets measured.
The second is the honest version of fleet compliance. A device that has not reported in a month is not a compliance percentage, it is a machine you cannot patch, cannot wipe and may not possess.
the NCSC device security guidance frames the ability to reach and update a device as the foundation the other controls sit on, which is why this number deserves to be reported on its own rather than folded into a percentage.
Why the third one belongs in a budget conversation
Recovery rate converts directly into money. Every device not recovered is a replacement purchase for hardware already paid for, which makes it the easiest of the four to translate into a funding argument.
Our note on IT offboarding covers the process that number depends on.
Report the median and the tail
Averages mislead badly in operations because the distributions are skewed. Most requests are fine and a small number are terrible, and the mean sits somewhere that describes neither.
Two numbers tell the real story. The median, which is the typical experience. And the count above some multiple of it, which is the set of people who had a bad time.
That second figure is also the useful agenda item. A list of eleven requests that took more than three weeks is a conversation with specific causes. An average of nine days is not a conversation at all.
Watch the tail by geography
Long tails in this work are almost never random. They cluster in particular countries, where lead times are long, customs is slow or no local supplier exists.
Breaking the tail down by market usually shows that a general performance problem is actually a specific supply problem in two or three places, which is a much more tractable thing to fix.
The decision table
What to measure first depends on what is currently invisible.
| Situation | Scale | Setup | Primary pain | Recommended starting point |
|---|---|---|---|---|
| No metrics at all | Under 50 staff | Informal support | Nothing is visible | Time to a working person, nothing else |
| Ticket metrics only | 50 to 200 | Service desk in place | Numbers look fine, people complain | Add outcome metrics alongside |
| Under budget pressure | Any | Needs a case | Activity data does not persuade | Cost per equipped person plus recovery rate |
| Distributed across markets | Series B+ | Mixed suppliers | Averages hide the problem | Median and tail split by country |
| Large unmanaged fleet | Any | Enrolment incomplete | Compliance number is fiction | Devices not reporting in 30 days |
Row five is the uncomfortable one. A compliance percentage calculated only over devices that check in is a measure of the devices you can see, which is exactly the wrong denominator.
Presenting four numbers, not a dashboard
Dashboards fail in budget reviews for a simple reason: nobody reads them, so the argument has to be made verbally anyway and the dashboard becomes decoration.
Four numbers, each with a direction and a short sentence of context, does more. The context sentence is the part that matters and the part usually omitted.
Saying median time to a working person is nine days is a fact. Saying it is nine days, down from fourteen, with the remaining delay concentrated in two markets where we hold no stock, is an argument with an obvious next step attached.
Show one thing that did not work
Every reporting period contains something that went badly. Including one of those, with what was learned, does more for credibility than an all-green summary.
Reviewers discount reports where everything improved, correctly, because operations does not work like that.
It also changes what the meeting is for. A report that only contains good news invites scrutiny of the numbers. One that names a real problem invites a conversation about fixing it, which is a considerably better use of the time and usually the reason the meeting exists.
Our piece on device health monitoring covers where the second metric in this list comes from.
Where the numbers come from
Three of the four do not come from the ticketing system, which is why they are rarely reported.
Time to a working person needs a start event when a request is raised and an end event when the device enrols successfully. Both exist in most environments and are simply in different systems.
the NIST Cybersecurity Framework is a reasonable reference for why visibility of assets underpins almost everything else, and the practical consequence is that enrolment data is the most useful operational signal most teams already have and do not use.
Devices not reporting comes from endpoint management directly.
Recovery rate needs offboarding to record an outcome per device, which is usually the piece that has to be built.
Cost per equipped person needs the hardware costs that are scattered across budgets gathered into one figure, including shipping, recovery and disposal.
Start with two
Instrumenting all four at once usually produces none. Time to a working person and recovery rate are the pair with the best ratio of effort to argument, and both improve with the same operational changes.
For distributed teams the second is bounded by whether collection is possible at all. RemoAsset handles recovery across markets, which is what turns recovery rate from a measure of your courier luck into a measure of your process.
The first metric has a similar constraint at the other end of the lifecycle. Time to a working person is bounded by how fast a device can reach somebody, so in markets without local supply the number reflects geography rather than effort. Local procurement and delivery is what makes it a measure of the process again.
RemoAsset recovers devices across markets, which is what makes recovery rate a measure of process rather than luck. More at PeopleOpsHQ.
Frequently asked questions
Which IT operations metrics actually matter?
Four that describe outcomes rather than activity: time from request to a person actually working, devices not reporting in the last thirty days, recovery rate on leavers, and cost per equipped person. Ticket counts and average response times measure how busy the team is, which is useful for managing the team and does not answer the question a budget review asks, namely whether the money keeps people working.
Why are ticket metrics a poor basis for a budget case?
Because they describe activity and have an awkward property: they rise when things go badly. They are also easy to improve without helping anybody, not through bad faith but ordinary optimisation. Average response time improves with an auto-reply, tickets closed improves by splitting one issue into three. The test for any metric is whether it could improve while the experience of the people you support gets worse.
Should we report averages or medians?
Median, plus a count of how many cases exceeded some multiple of it. Operational distributions are skewed, with most requests fine and a small number terrible, so the mean describes nobody’s experience. The tail count is also the better agenda item, since eleven requests that took over three weeks is a conversation with specific causes while an average of nine days is not a conversation at all.
What is wrong with a fleet compliance percentage?
The denominator, usually. A compliance figure calculated across devices that check in measures only the devices you can already see, which excludes precisely the machines worth worrying about. Devices not reporting in thirty days is the honest version, because a device that has not reported is one you cannot patch, cannot wipe and may not actually possess any more.
How do we translate IT metrics into money?
Recovery rate converts most directly. Every device not recovered from a leaver is a replacement purchase for hardware the company already paid for, which makes it straightforward to express as a funding argument. Cost per equipped person is the other one finance engages with, provided it includes the costs usually scattered elsewhere such as shipping, customs, recovery and disposal.
How should these be presented in a review?
Four numbers with a direction and one sentence of context each, rather than a dashboard. The context sentence is the part that does the work and the part usually left out. Nine days is a fact; nine days, down from fourteen, with the remaining delay concentrated in two markets where we hold no stock, is an argument with an obvious next step. Include one thing that did not work, because all-green reports get discounted.
Which metric should we instrument first?
Time to a working person and recovery rate, as a pair. They have the best ratio of effort to argument and both improve through the same operational changes. Instrumenting all four at once reliably produces none. The first needs a start event when a request is raised and an end event at successful enrolment, which usually already exist in different systems and have simply never been joined up.
RemoAsset recovers devices across markets so recovery rate reflects your process, not your luck.