AI HR Tools 9 min read

Employee Productivity Software and What It Cannot See

Employee productivity software measures activity and calls it output. What the numbers genuinely show, what they miss, and the line not to cross.

Michael Rodriguez Michael Rodriguez • • 9 min read
Employee Productivity Software and What It Cannot See

TL;DR

Employee productivity software counts things that are easy to count: keystrokes, application time, messages sent. None of those is output, and treating them as a proxy changes behaviour in a direction nobody intended. There is a narrow band where this tooling is genuinely useful, and it is not the band it is sold for.

  • Activity is not output. Everything downstream depends on that sentence.
  • Measured people optimise the measure. Reliably, and quickly.
  • Aggregate and anonymous is defensible. Individual scoring is not.
  • Tell people before you install it, or the finding is the smallest problem.
  • The genuine use is finding broken processes, not ranking staff.
  • If you cannot say what you would do with it, do not buy it.

What employee productivity software actually measures

It measures activity, because activity is what a machine can see. The gap between that and output is the whole problem.

What it counts What it implies Why that fails
Active hours Effort Thinking looks identical to idling
Applications used Focus The best hour may be a whiteboard
Messages sent Collaboration Rewards noise over resolution
Documents touched Contribution Counts edits, not judgement

The behaviour change is the real cost

People respond to what is counted, quickly and rationally. Count active hours and the mouse moves during a long read. Count messages and the thread gets longer, not shorter.

Neither of those is dishonesty. Both are people correctly inferring what the organisation now rewards, and the organisation then receives exactly what it asked for rather than what it wanted.

That is the argument against individual scoring, and it holds regardless of the tool’s quality. Better measurement of the wrong quantity produces a more precise wrong answer.

It also damages the measurement for everyone else. Once a team believes the numbers feed a judgement about them, the data stops describing how work happens and starts describing how people think work should look, which makes the dataset useless for the process questions it was genuinely good at.

Free Weekly Briefing Stay ahead of what's changing in HR and people ops.

Join 4,200+ leaders getting practical insights every week — no fluff, just signal.

Join Free →

The line not to cross, and where it sits

There is a real distinction between measuring a process and measuring a person, and it is the one that decides whether this is defensible.

Process measurement. How long an approval takes on average. How many handoffs a request passes through. Which step has the longest queue. Aggregate, anonymous, about the system.

Person measurement. What this individual did at their keyboard, scored and compared. About the person, and a different proposition entirely.

The first improves work. The second mostly produces a number managers do not trust and employees resent, and it rarely survives contact with a competent works council or a determined question at an all-hands.

What the published guidance expects

the UK Information Commissioner guidance on monitoring workers sets out expectations around monitoring at work, including that people should know what is happening and that the monitoring should be necessary for a stated purpose rather than collected because it is available.

Under frameworks such as the consolidated text of the General Data Protection Regulation, the data generated is personal data with the obligations that follow, which is worth knowing before a tool is installed rather than after a subject access request arrives asking for everything it holds.

None of that makes the category unusable. It makes the aggregate, transparent, purpose-limited version the only one worth building on.

The narrow band where it genuinely helps

Disclosure: RemoAsset, named below, is owned by the same people who publish HROpsLab. It is described with its limitations stated, on the same terms as every other supplier. This site takes no vendor money.

Stated plainly, because the useful version is much smaller than the pitch and gets lost in the argument about surveillance.

Finding a broken process. When a request takes eleven days and nobody can say which step consumed them, aggregate flow data answers it. That is a real question with a real answer.

Sizing a support team. Volume and timing data tells you whether a queue is understaffed or badly routed, which is otherwise guesswork.

Checking whether a change worked. You changed an approval threshold. Did the wait actually fall? Without a before number this is an opinion.

Notice what those have in common. Each is a question about the system, each has a specific answer, and none needs anybody’s name attached.

There is a fourth that sometimes qualifies: checking whether a team is under-resourced rather than underperforming. The two look identical from the outside and the distinction matters enormously to the people in that team, so it is worth the effort to tell them apart before anybody draws a conclusion.

The test before buying

Write down the question you would answer and the decision that would follow. If the sentence does not survive being read aloud to the people who would be measured, the problem is the purpose rather than the phrasing.

The thing it will not explain

A fair amount of what looks like low productivity is a slow machine, and no amount of activity data distinguishes a person working slowly from a person waiting for a laptop that should have been replaced two years ago.

Check the hardware before interpreting the dashboard. Refresh timing is the unglamorous version of this problem, and our guide to hardware refresh cycles covers how to decide it. Where a replacement is the answer, configured device procurement and delivery is the mechanism, and the limitation worth stating is that it does nothing about the decision itself.

The decision table

Situation Scale Setup Primary pain Recommended starting point
Requests take weeks, cause unknown Any Multi-step approvals Cannot see the queue Aggregate flow data, no names
Asked to rank individuals Any Pressure from above Measure will be gamed Decline, offer the process version
Support team sizing 50 plus One shared queue Guesswork Volume and timing, anonymised
Works council in place EU Consultation required Consent and purpose Agree purpose before any pilot
Suspected low output Any Older fleet Possibly the hardware Check machine age first

If you do buy it, four conditions

Four conditions that keep this on the defensible side. All should be agreed before installation rather than retrofitted after somebody objects.

Tell people first, in plain words. What is collected, why, who sees it, how long it is kept. Discovering it afterwards does more damage than anything the tool could report, and it is the version that ends up in a tribunal or a press article.

Aggregate by default, with a high bar for anything else. If individual data is ever viewable, write down who may look, on what grounds, and that it is logged.

A retention period, enforced. Data kept indefinitely because deletion was never configured is the commonest failure, and it converts a reasonable tool into a liability as the dataset grows.

A review date with the power to switch it off. Six months, with a named decision maker. Most of these tools are never revisited after the enthusiasm fades, which is how organisations end up running surveillance nobody can justify and nobody remembers commissioning.

And one thing to measure about the tool itself

Whether any decision changed because of it. Not whether the dashboards look impressive. Whether a process was altered, a team resized, a threshold moved.

If nothing changed in six months, the honest conclusion is that the question was never pressing, and the right action is to remove it rather than renew it.

What to tell people who ask why it is there

Somebody will ask, usually in the first fortnight, and the quality of that answer sets how the tool is received for its whole life.

A good answer names the specific question being investigated, says what is not collected, and gives a date when the decision will be revisited. A bad answer talks about insight and visibility, which people correctly hear as having no answer.

If the honest answer is that leadership wanted reassurance rather than information, that is worth knowing internally before installation, because reassurance is not a purpose any published guidance recognises and it will not survive the first challenge.

That is the same test worth applying to any measurement, and it is the one most often skipped. For the related question of measuring systems rather than people, see our piece on workplace analytics.

Check the hardware before the dashboard at configured device procurement and delivery, or read our guide to hardware refresh cycles.

Frequently asked questions

What does employee productivity software actually measure?

Activity, because activity is what a machine can see. Active hours, applications used, messages sent, documents touched. None of those is output. Thinking looks identical to idling, the best hour of the week may be spent at a whiteboard, and counting messages rewards longer threads rather than resolved ones.

Why is individual scoring a problem?

Because people respond to what is counted, quickly and rationally. Count active hours and the mouse moves during a long read. Count messages and threads get longer. That is not dishonesty, it is people correctly inferring what the organisation rewards, and the organisation then gets what it asked for rather than what it wanted.

Where is the line?

Between measuring a process and measuring a person. How long an approval takes, how many handoffs a request passes, which step has the longest queue: aggregate, anonymous, about the system. What this individual did at their keyboard, scored and compared: about the person, and a different proposition entirely.

What do the published expectations say?

Guidance on monitoring at work expects people to know what is happening and the monitoring to be necessary for a stated purpose rather than collected because it is available. The data generated is personal data with the obligations that follow, which is worth knowing before installation rather than when a subject access request arrives.

When is it genuinely useful?

Three cases, all about systems. Finding which step of an eleven-day process consumed the time. Sizing a support team from volume and timing rather than guesswork. And checking whether a change actually worked, which needs a before number. None of the three requires anybody’s name attached.

What is the test before buying?

Write down the question you would answer and the decision that would follow. If that sentence does not survive being read aloud to the people who would be measured, the problem is the purpose rather than the phrasing. If you cannot write the sentence at all, do not buy the tool.

Could low numbers mean something else?

Often, yes. A good share of what looks like low productivity is a slow machine, and no activity data distinguishes somebody working slowly from somebody waiting on a laptop that should have been replaced two years ago. Check the age of the hardware before interpreting the dashboard.

Share on X Share on LinkedIn

What to do next?

Explore More Articles

Dig deeper into HR Ops strategy, tools, and workflows built for real teams.

Browse the blog →
Join the HROpsLab Community

Connect with People Ops practitioners sharing real workflows, tools, and challenges.

Join now →