Employees have no incentive to disclose how productive AI makes them

The harder a company pushes to measure individual AI productivity, the stronger the reason it gives each employee using it to keep that productivity out of view. So the gains leaders most need to see become the ones employees have the least reason to reveal. Ricardo Viana Vargas , Joao Henrique Jacinto and Antonio Nieto-Rodriguez call this the “AI disclosure paradox”.

An AI productivity trial can report 26 minutes saved per person per day and still leave the hardest management question unanswered: which of those minutes became value, and who felt safe enough to say so?^[1]

That second question is the one most leaders rarely ask. The tools perform, employees save time and early satisfaction rises. Yet when leadership looks for that gain in enterprise performance, it is often nowhere to be found. The saved time exists. It has simply gone quiet.

In late 2024 the British government ran its largest Microsoft 365 Copilot experiment to date, involving 20,000 employees across 12 organisations. Participants reported saving an average of 26 minutes a day. The finding was encouraging. But the limitation was more revealing. The savings were self-reported and no one could verify how the recovered time was reallocated. A later evaluation at the Department for Work and Pensions added a firmer signal. Comparing Copilot users with non-users through regression analysis, it estimated 19 minutes saved per day across eight routine tasks, with statistically significant gains in efficiency, job satisfaction and perceived quality of work.^[2][3]

The pattern is not unique to one vendor. In a Fortune 500 study of customer-support agents , access to a generative AI assistant raised issues resolved per hour by 14 per cent. In a field experiment with Boston Consulting Group consultants , those using AI finished faster and at higher quality.^[4][5]

Read together, these studies frame the problem precisely. The gain is real enough to survive a control group and invisible enough that even the organisation producing it cannot trace where it went. Someone writes, summarises, prepares or decides faster and the gain is real before it reaches a budget, a service level or a cycle-time metric. This is the invisible AI dividend: the productivity a company may never capture because it cannot see it.

The paradox of disclosure

The invisibility of AI productivity is behavioural as much as technical. The first person to know that AI saved time is usually the person who saved it, and that person has to judge what disclosure will mean. Revealing the gain might bring more work, tighter targets or questions about whether the role needed that time at all. Silence becomes understandable. The problem is not bad faith; it is incentives.

A second, quieter reason needs no threat from management. In many workplaces, visible effort still serves as a proxy for value. Admitting that AI helped produce the work can make the output look less earned, even when the judgment behind it remains valuable. The gain stays private because revealing it carries social risk even when it carries no managerial one.

This is not hypothetical. In Microsoft and LinkedIn’s 2024 Work Trend Index , 78 per cent of AI users were already bringing their own tools to work, and 52 per cent were reluctant to admit using AI for their most important tasks, many fearing it would make them look replaceable. The concealment is already happening, at scale and outside any official programme – a problem to manage, not a risk still to prevent.^[6]

The instinctive response to invisible productivity is to measure more, and in AI programmes that instinct can make the problem worse. When productivity data attaches directly to individuals, the number acquires a potential defendant: it can reward, but it can also compare, pressure or justify reduction. That creates a rational incentive to understate the gains AI makes possible. Managers have long known that a measure stops being a good measure once it becomes a target. AI adds a sharper variant, in which the distortion can happen before the number is ever recorded. The number may never be born.^[7]

This is the AI disclosure paradox. The harder a company pushes to measure individual AI productivity, the stronger the reason it gives each person to keep that productivity out of view. The gains leaders most need to see become the ones employees have the least reason to reveal.

Disclosure-safe measurement

The way out is disclosure-safe measurement: measure the workflow where AI changes the work, separate that data from individual performance review and make reinvestment credible enough to be believed.

Start with the unit of analysis. Consider a commercial proposal process. The personal question — how many hours each analyst saved by using AI — quietly invites concealment. The workflow question has no defendant. How long does it now take from a client request to a delivered proposal? What share of proposals return for rework? None of these points at a person; all of them reveal whether AI changed the work. When a gain belongs to a workflow rather than a person, disclosure stops feeling like exposure and becomes operational intelligence. Watching how work moves across functions without turning data into a verdict on people is precisely the modern mandate of the project management office, transformation office or AI centre of excellence. It is also the group best placed to trace the gain from task to workflow to strategic outcome — the discipline of benefits realisation — which is what raises the stakes from saved minutes to whether the strategy actually lands.

Measurement, though, is only half the design. People reveal gains when they trust what happens next. And that trust has to be earned before the disclosure, not after. A reinvestment contract makes the trade explicit: leaders state, in advance and in the open, how recovered capacity will be used, and they name the use specifically enough that the team can later tell whether the promise was kept.

What makes such a contract credible is not the language of the commitment but the cost of breaking it. A vague pledge that saved time will support higher-value work reassures no one, because it can be quietly abandoned. A specific, visible commitment cannot. Return to the proposal cycle. Leadership declares that the roughly two hours a week AI frees for each analyst will be reinvested in named work the team can verify — deeper client research and earlier legal review, protected on the calendar rather than quietly absorbed by a higher proposal quota. A quarter later, the team can see whether the promise held and the commitment becomes self-enforcing. A team that watches its disclosed gain fund deeper work learns that disclosure is the rational choice. A team that watches nothing change learns the opposite lesson, and it learns it once.

The harder part is human

The next phase of AI adoption will be decided less by the sophistication of the tools than by the social architecture of disclosure. The companies that pull ahead will not be the ones that extract the most individual productivity claims. They will be the ones that make productivity safe to reveal before they try to make it easy to measure.

The hard part was supposed to be the technology; it turns out the harder part is human. It means persuading the people who hold the gains that telling the truth about them will be rewarded rather than punished. Get that wrong, and the dashboard keeps improving while the dividend stays hidden, held quietly and rationally by the people who earned it. Get it right, and the invisible dividend finally has a reason to come into view.^[8]

This article gives the views of the author, not the position of LSE Business Review or the London School of Economics. You are agreeing with our comment policy when you leave a comment.