It measures hands at a keyboard. Work is what gets achieved

The question behind the request

The request arrives as a product request: we should get something that shows what people are doing. Underneath it is a concern, and the concern is rarely about software.

The distinction in “It measures hands at a keyboard. Work is what gets achieved” is important when operational records are interpreted. Organisations researching employee monitoring software can use employee monitoring software to add time and project context, while outcomes, direct feedback and human review remain necessary to explain what the numbers do not show.

What is usually said is that output has dropped, or that nobody can tell what people are working on. What is usually meant is that nothing visible marks progress, that the manager has no way to raise a concern without a confrontation, or that somebody above is anxious and action must be seen to be taken.

For an independent reference relevant to “It measures hands at a keyboard. Work is what gets achieved”, consult the European Commission data-protection resources; it provides a useful external check on scope, terminology, governance and the claims made during procurement or review.

Two questions separate these quickly. "If you had this data tomorrow, what would you do differently?" — which distinguishes a management question from a reporting one from a feeling. And "what would change your mind about this person?" — because if the answer is nothing, monitoring is being sought as evidence rather than as information.

What the software actually captures

Which window is in focus and for how long. Keyboard and mouse activity, usually as counts. Periods with no input, reported as idle. Sites visited. Screenshots, at intervals. Login times.

What is inferred from that: productive and unproductive time, from a categorisation somebody chose; a score, from weights the vendor chose; attention, from input patterns. Each is an interpretation presented as a measurement.

What is not captured at all: whether anything was achieved, thinking, reading on paper, a conversation, a call taken on a mobile, work on another device, quality, difficulty — and whether the person was doing the right thing, which is the only question the business has.

The inversion

The relationship between activity and output is not merely weak. It gets weaker as the work gets more skilled, and in places it reverses.

Data entry has a genuine correlation between keystrokes and output. Design, analysis, writing, diagnosis and advice have close to none — the person who solved it in ten minutes of thought scores worse than the one who spent two hours flailing.

There is also an averaging trap. Across a large group over a long period, activity and output correlate somewhat, because people doing nothing do show up. That aggregate relationship is then applied to individuals, where it does not hold at all. The harm happens at that step.

Gaming is not a discipline problem

Within weeks of deployment, measured behaviour changes. Mouse movers. Documents kept open and scrolled during calls. Work split into more, smaller visible actions. Email sent at times that look good rather than when written.

This is rational. The organisation announced that an indicator matters, and people optimise toward indicators that matter — in every field where this has been studied. Somebody who ignores the measure is disadvantaged relative to somebody who manages it, which means the honest response is the one that gets punished.

Within months the figures describe awareness of the measure rather than work, and the gap between high and low scorers becomes a gap between those who adapted and those who did not. Widespread gaming is a finding about the measure: people do not believe it reflects their work, and they are usually right.

The costs that are not on the invoice

"It damages trust" is true and too general to act on. What actually changes is a set of behaviours, each with a practical cost.

People stop reporting problems early, because a problem is a dip in the numbers. Issues surface later and larger, and this is invisible in any accounting.

People stop asking for help, because asking looks like not knowing. An hour is spent on what a colleague would have answered in two minutes — and the measured activity goes up while the organisation is worse off.

People stop taking on difficult work, because hard work has long thinking periods and uncertain outcomes while easy visible work scores better.

And the people who leave first are the ones with options. The number who go is usually small; the composition is the problem, and one departure can exceed the annual licence for a whole team.

Reading what the vendor tells you

Every product in this category carries a percentage improvement. The figures are produced in a consistent way worth understanding.

The most common claim is that productive hours rose — and productive is defined by which applications the vendor's categorisation approves of. So the claim is that after deployment, people spent more time in applications the vendor counts as good. That may be true and it is not a statement about output.

Published benchmarks work the same way: an average of the vendor's metric across the vendor's own customers, who are organisations that bought monitoring software. The figures are low, and that is the point — a low benchmark makes an ordinary working day look like a crisis. Nobody has ever worked eight concentrated hours.

Three questions change the conversation. How is productive time defined, and may we see the category list? Was there a control group? And may we speak to a customer who deployed more than two years ago? The last is the most useful and the least expected.

The category list nobody looks at

Underneath every productive-time figure is a list labelling software as useful or not. It ships from the vendor, built from general assumptions about office work, and in most deployments it is barely adjusted.

Which means your organisation's definition of productive work was written by somebody who has never seen it. Research on industry forums lands in unproductive. Video platforms are unproductive even when the video is training. Documentation reading, which for engineers is a large share of competent work, frequently counts against them.

And the asymmetry nobody notices: a person idly scrolling inside an approved application scores as productive, while a person reading exactly the right technical article does not. The list measures which window is open, and nothing can make it measure what is happening in the window.

A category list employees cannot see is a secret standard they are graded against. If it requires this much care to be fair, and will not get it, that is an argument against relying on productive-time figures at all.

Who the measure penalises

The burden is not evenly distributed, and the differences track things unrelated to performance.

Anybody whose work happens on the telephone, in meetings, on site or on paper appears less active because their work is less typed. Senior people in analytical roles routinely score below juniors doing mechanical work. New starters appear as low performers for their first months, which is exactly when a manager's impression forms.

People with caring responsibilities have fragmented days — a school run, an appointment, an elderly parent — which produces an irregular pattern and disproportionately affects women. Disabled colleagues using assistive technology, voice input or managing fatigue produce patterns the measure reads as poor, and then must explain their figures repeatedly, which is a burden placed on them rather than on the system.

Which means an individual ranking is driven partly by role, caring responsibilities, disability and tenure. And it will be used, because it exists.

What it does to management

Monitoring is justified as giving managers information. What it frequently does is replace the activity that produced better information.

Before: the manager asks how it is going, hears about a blocker, helps. After: the manager checks the dashboard, sees a number, and either does nothing or raises the number. The conversation stops happening because something has apparently taken its place — and the something contains far less.

It is also most attractive to the managers least comfortable with the conversation, which means the tool is most used by the people it should least be given to, and it entrenches the weakness rather than supplementing it.

The honest test, six months in: are your managers having more conversations with their people, or fewer? Ask the people rather than the managers.

Where it is the right answer

This collection does not argue that monitoring is always wrong. Five cases justify it: a regulatory recording obligation, billing clients for time, specific privileged security roles, safety in lone or hazardous work, and a defined investigation with a stated basis.

They share four properties — a named obligation, a defined population, a defined signal, and a limit — and none of them is "we want to see how hard people are working".

Dressing a general productivity concern in the language of compliance is the most common way these deployments are justified, and it does not survive the consultation or the first challenge.

If you deploy anyway

Report at group level and hold that line. It is the single decision that determines whether the deployment is tolerable, and it answers every legitimate organisational question: workload distribution, meeting load, whether a pattern changed after a process change.

Configure to the narrowest thing that answers your question — most defaults collect everything the product can. Turn off screenshots, keystroke content, individual scores and covert mode. Narrow the population before narrowing anything else.

And set a date, in advance, at which you ask whether it did anything: which business measure should have moved, by how much, and what happens if it has not. Without that written down before purchase, the licence renews because switching off would be an admission.

What the core notes deliberately avoid

No product rankings appear inside the core notes. Named products are kept in separate comparison guides so the method and governance analysis does not depend on one supplier.

No productivity improvement figures, because they are produced by measuring the vendor's own metric on the vendor's own customers.

And no claim that this is a moral question. It is a decision with costs that do not appear on the invoice, taken by people who are usually trying to solve a real problem with the wrong instrument.