TL;DR
Digital employee experience metrics tend to be whatever the tooling emits, which is why so many of them collapse the moment somebody senior asks what they mean. Five hold up under that question. Three common ones do not, and two of those are in almost every deck.
- Elapsed time beats satisfaction, because it is comparable across teams.
- Report the distribution, not the mean. The tail is the whole story.
- Ticket volume is ambiguous: down can mean fixed or mean given up.
- Repeat contacts is the honest quality measure, and nobody uses it.
- A composite score cannot be argued with, which is why it survives so long.
- Every metric needs a decision attached or it is reporting, not measurement.
Five digital employee experience metrics that hold up
Each is comparable across teams and countries, each points at a specific step, and each has an obvious decision attached.
| Metric | Decision it drives |
|---|---|
| Offer accepted to equipment delivered | Move the ordering trigger |
| Start date to first productive day | Fix whichever step holds the delay |
| Request raised to resolved, by category | Staff or automate that category |
| Repeat contacts on the same issue | Fix the cause rather than the symptom |
| Leaving date to device recovered | Change the leaver trigger |
Why elapsed time rather than sentiment
A satisfaction score moves for reasons nobody can trace, cannot be compared between a team in one country and a team in another, and depends on who happened to respond that week.
Days are days everywhere. When one region takes twelve and another takes three, you know where to look without a further investigation, which is the whole point of a metric.
The one nobody tracks
Repeat contacts. It is the closest thing to a quality measure in this area, because a problem solved once stays solved and a problem patched comes back with the same person attached.
Counting it needs nothing new. Group tickets by person and by subject over a quarter and the repeats are visible immediately. The usual finding is that a small number of underlying causes generate a disproportionate share of contacts, and that none of them was ever logged as a problem in its own right.
The three that flatter the team
All three appear in most reporting, all three are easy to produce, and none of them means what it is presented to mean.
Ticket volume, falling. Presented as improvement. It equally indicates that people have stopped bothering to report things, which is the worse outcome and looks identical on the chart. Pair it with repeat contacts and with a sample of people asked directly, or do not use it.
Average resolution time. The mean is dragged down by the large number of trivial tickets and tells you nothing about the ones that matter. One person waiting nineteen days disappears entirely into an average of four hours.
A composite experience score. Whatever the vendor’s weighting produces. It cannot be decomposed, two products scoring the same fleet will disagree, and its chief virtue is that nobody can argue with it, which is exactly why it survives in decks long after it has stopped being informative.
What to report instead of the average
The ninetieth percentile and the count above a stated threshold. “Forty-one requests took more than five days” is a sentence somebody can act on. “Average resolution 4.2 hours” is a sentence that ends a conversation.
The same applies to every elapsed-time metric in the first section. Report the tail, because the tail is both where the damage is and where the fixable causes cluster. An average tells you the system usually works, which you already knew. The tail tells you who it failed and roughly why, which is the part somebody can act on this week.
What not to measure, and why
Disclosure: RemoAsset, named below, is owned by the same people who publish PeopleOpsHQ. It is described with its limitations stated, on the same terms as every other supplier. This site takes no vendor money.
Two categories to leave alone, for different reasons.
Anything about an individual’s activity. Hours active, applications used, messages sent. It changes behaviour toward the measure, it produces a number managers do not trust, and it carries obligations: the UK Information Commissioner guidance on monitoring workers expects monitoring to be necessary for a stated purpose, and under the consolidated text of the General Data Protection Regulation the data has duties attached regardless of why it was collected.
Anything you will not act on. The larger practical problem. Most dashboards in this category contain charts that have never changed a decision and never will, and each one dilutes the few that would.
The test before adding a metric
Write the sentence: if this number moved by a quarter, we would do X. If X is blank, the metric is decoration. Applied honestly this removes most of a typical dashboard, which is the point.
The measurement most teams are missing entirely
Leaving date to device recovered. It is absent from almost every DEX dashboard because no endpoint agent can report it, and it is the number finance eventually asks about.
Start counting it before setting any target. Most teams discover the number is far worse than anybody assumed, because the cases that drift are invisible by definition: nobody chases a device belonging to a person who left eight months ago, and nobody reports its absence either.
It is also the one that drifts worst without an owner, because nothing forces the issue once somebody has left. Where the finding is a long tail, collection with packaging and a courier arranged is the mechanism that shortens it, and the limitation is the familiar one: no supplier can tell you somebody resigned. The trigger stays with your people system.
Control frameworks such as ISO/IEC 27001, the information security management standard expect the inventory to reflect reality, which makes this a compliance number as well as an operational one.
The decision table
| Situation | Scale | Setup | Primary pain | Recommended starting point |
|---|---|---|---|---|
| Dashboard nobody acts on | Any | Tooling in place | Reporting, not measurement | Delete every chart with no decision |
| Ticket volume falling | Any | Help desk | Ambiguous signal | Pair with repeat contacts |
| Averages look healthy | Any | Any | Tail invisible | Report count above a threshold |
| Board wants one number | 200 plus | Exec reporting | Composite hides everything | Offer three, not one |
| No recovery number at all | Any | Remote fleet | Nobody owns it | Start counting, then set a target |
Review the metrics themselves once a year
Metrics accumulate. Nobody removes one, because removing it looks like hiding something, so dashboards grow until the important numbers are indistinguishable from the decorative ones.
Once a year, take each metric and ask three questions.
Did it change a decision in the last twelve months? Name the decision. If you cannot, that is the answer.
Who looks at it, and when? If the honest answer is that it is generated and nobody opens it, it is costing collection effort for nothing.
Would we notice if it stopped? The most revealing of the three, and frequently the answer is no.
What usually survives
Do the cull in public rather than quietly. A note saying which metrics were removed and why prevents the slow reappearance of the same charts six months later under a different name, which is what happens when the removal is never explained.
Three or four numbers, typically the elapsed times. Everything built from sentiment scores and composite indices tends not to, because nobody can point to a decision that turned on them.
The count matters less than the honesty of the cull. A dashboard of four numbers that each changed something beats one of twenty where nobody can name a single decision, and the second is what most teams have after three years of nobody removing anything.
That is a healthy outcome rather than an embarrassing one. A short set of numbers that genuinely drive decisions beats a comprehensive set that describes everything and changes nothing, and it is substantially cheaper to maintain.
Designing measurement around what people actually experience rather than around what the tooling can emit is the substance of ISO 9241-210, the standard on human-centred design for interactive systems, and an annual cull is the cheapest way to keep that honest. For the related question of what endpoint tooling can see at all, see our piece on DEX tools.
See how the recovery number gets shortened at device retrieval with packaging arranged, or read what DEX tools miss.
Frequently asked questions
Which metrics actually hold up?
Five elapsed times: offer accepted to equipment delivered, start date to first productive day, request raised to resolved by category, repeat contacts on the same issue, and leaving date to device recovered. Each is comparable across teams and countries, points at a specific step, and has an obvious decision attached.
Why elapsed time rather than satisfaction scores?
Because a satisfaction score moves for reasons nobody can trace, cannot be compared between regions, and depends on who responded that week. Days are days everywhere. When one region takes twelve and another takes three, you know where to look without a further investigation, which is what a metric is for.
Why is falling ticket volume ambiguous?
Because it equally indicates that people have stopped bothering to report things, which is the worse outcome and looks identical on the chart. Pair it with repeat contacts and with a sample of people asked directly, or do not use it as evidence of improvement at all.
What is wrong with average resolution time?
The mean is dragged down by a large number of trivial tickets and tells you nothing about the ones that matter. One person waiting nineteen days vanishes entirely into an average of four hours. Report the ninetieth percentile and the count above a stated threshold instead.
Why avoid a composite experience score?
It cannot be decomposed, two products scoring the same fleet will disagree because the weighting is the vendor’s rather than yours, and its chief virtue is that nobody can argue with it. That is exactly why it survives in decks long after it has stopped telling anybody anything.
What is the test before adding a metric?
Write the sentence: if this number moved by a quarter, we would do X. If X is blank, the metric is decoration. Applied honestly this removes most of a typical dashboard, and that is the intended outcome rather than a sign something went wrong.
Which measurement are most teams missing?
Leaving date to device recovered. It is absent from almost every dashboard because no endpoint agent can report it, and it is the number finance eventually asks about. It also drifts worst without an owner, since nothing forces the issue once somebody has already left the company.
PeopleOpsHQ covers the operational side of distributed work. We describe suppliers with their limitations stated, and we take no vendor money.