AI needs a recovery budget: why every minute saved shouldn’t become more work

By Dr Gleb Tsipursky CEO of AI consultancy Disaster Avoidance Experts and author of eight books, including The Psychology of AI Adoption at Work: From Resistance to Results.

UK employers have a legitimate productivity problem. Office for National Statistics figures show output per hour in the first quarter of 2026 was only 0.4 per cent higher than a year earlier, while growth remains weak compared with trends before the 2008 financial crisis.

It is understandable, then, that leaders are attracted to AI productivity measures: minutes saved, cases closed and drafts generated.

But these measures capture speed while overlooking the human work required to make AI output accurate, appropriate and sustainable.

AI might produce a draft in 10 minutes, for example, only for an employee to spend another 20 checking sources, correcting tone, reconciling conflicting instructions and deciding whether something needs to be escalated.

This matters because when every minute saved by AI is converted into a higher target, technology can increase the intensity of work even while the productivity dashboard suggests improvement.

Research published in Technovation found workplace automation tends to introduce more or stronger demands than resources, with AI workloads creating particular challenges for mental and relational wellbeing.

The UK can ill afford to ignore those risks. The Health and Safety Executive estimates 964,000 workers experienced work-related stress, depression or anxiety in 2024/25, while the Office for National Statistics recorded 148.8 million working days lost to sickness or injury in 2025.

But this isn’t an argument against AI.

Research involving 5,179 customer-support agents found generative AI increased productivity by an average of 14 per cent, with the greatest gains among newer and lower-skilled workers. A later randomised experiment across 66 firms found active AI users spent around two fewer hours a week on email and reduced work outside normal hours.

AI can create genuine breathing space. The question for employers is what they do with it.

Don’t automatically fill the space AI creates. This is where I believe organisations need a recovery budget.

By this, I mean planned capacity for the attention, verification, learning, human contact and decompression required by AI-intensive work.

It isn’t another employee wellbeing benefit. It is part of work design.

If AI increases the number or speed of decisions an employee makes, organisations should ensure there is enough time, staffing, support and discretion for people to continue making those decisions well.

That could mean protected review periods following high-volume AI work, fewer meetings after predictable peak periods or lower throughput expectations for work requiring significant verification.

It could also mean giving employees clear permission to stop, seek a second opinion or return to a manual process when the consequences of getting something wrong are significant.

This aligns closely with the Health and Safety Executive’s approach to work-related stress, which looks at demands, control, support, relationships, role and change. AI has the potential to affect every one of them.

Look beyond time saved

Employers also need to reconsider how they measure AI’s impact.

If output rises after AI is introduced but errors, complaints, rework or audit exceptions rise too, has productivity genuinely improved?

Likewise, if individual tasks become faster but employees are making more decisions, switching between more contexts or working longer hours, the organisation may simply have created denser work.

Measures such as time saved and output still matter. But they need to sit alongside indicators including quality, rework, after-hours activity, absence, workload, employee autonomy and confidence in challenging AI-generated decisions.

Employers should also pay attention to where the hidden work lands.

A small group of confident AI “superusers”, for example, can easily become the people who absorb everyone else’s troubleshooting, checking and informal coaching.

There is a similar risk for junior employees. If AI removes the routine tasks through which people traditionally develop knowledge and judgement, employers need to deliberately replace those learning opportunities.

Protect human judgement

Research involving 319 knowledge workers found generative AI shifted critical-thinking effort away from gathering information and towards verification, integration and task stewardship.

In other words, AI doesn’t necessarily remove cognitive work. It can change where that work happens.

That distinction becomes particularly important when AI is involved in decisions affecting pay, performance, safety, health, legal rights or vulnerable customers.

Employers should therefore distinguish between low-risk work, where a quick human review may be sufficient, and higher-risk decisions requiring stronger checks and clear escalation routes.

Human contact matters too. Teams need time for peer review, discussing exceptions and coaching. Experienced employees need capacity to transfer knowledge rather than becoming permanent AI correction layers, while less experienced employees need opportunities to develop the judgement required to challenge machine output confidently.

Recovery belongs inside the working day

Perhaps most importantly, employers shouldn’t make employees responsible for recovering from poorly designed AI-enabled work in their own time.

Recovery needs to happen within paid working hours, rather than being pushed into evenings, weekends or personal resilience programmes.

Nor should employers use this approach to create another form of employee surveillance. Measures should primarily help organisations understand whether a workflow is sustainable, not monitor whether individual employees are working hard enough.

The principle is relatively simple. Before introducing AI into a workflow, understand how the work currently operates. Once AI is introduced, measure not only what becomes faster but what new work is created around checking, correcting and escalating its output.

Then ask whether targets, staffing and working practices still make sense.

The biggest mistake would be to identify a time saving in a small AI pilot and immediately convert it into higher quotas.

AI offers employers a genuine opportunity to improve productivity and reduce unnecessary work. But if every efficiency is instantly absorbed into greater output expectations, organisations risk losing the very benefit the technology promised.

Over the next year, employers should therefore ask a more demanding question than simply how much time did AI save?

Did it create additional value while preserving people’s health, judgement, capability and trust?

A recovery budget makes that hidden part of AI-enabled work visible. And as AI becomes embedded across UK workplaces, protecting human capacity should be considered part of responsible implementation, not something employers address after people begin to struggle.

Adapted from The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).

Disclaimer: The views expressed in this opinion piece are those of the author and do not necessarily reflect the views of The Well Crowd. This content is for information and discussion purposes only and should not be taken as medical, health, or professional advice.