AI Saved the Time. Leadership Must Decide Where It Goes.

“Time saved” has become the most popular way to talk about AI progress. It’s simple to measure, simple to scale across a workforce, and reassuringly simple to present to a board. But an hour saved on drafting, searching, or summarising doesn’t automatically turn into value. It only matters once the organisation decides what that hour should be used for — and most never do. The hours are being released regardless, and they’re quietly absorbed into the day without any deliberate choice. 

That’s why the savings show up in surveys and board packs but rarely in the P&L. To change that, leaders need to treat AI‑created capacity as a resource that must be intentionally allocated, workflow by workflow, just as rigorously as budget or headcount. 

In this article, I’ll show you how to use the Capacity Budget to decide what each workflow’s released time should buy and who is accountable for the result. Applied well, it turns an hours‑saved metric into something a board can act on: a clear account of what that time actually delivered. 

BCG’s fourth annual AI at Work survey, published in June 2026, puts numbers on the gap. Among frontline employees who use AI regularly, 42% say it saves them a full working day or more each week. 66% say they get limited or no guidance on what to do with that time, and more than half are not moving it into strategic work. These are fluent users, so the finding cannot be explained as slow adoption. The capacity is being produced but the instructions for using it are missing.  

The pattern is not confined to one survey. Gartner’s July 2025 survey of HR leaders found that only a small minority of organisations give employees any guidelines on how to use the time that AI saves them. BCG’s own conclusion is that without real transformation, saved time simply leaks out of the organisation.  

In both surveys, the technology performed as intended but was operating inside a management system that hadn’t been updated to make use of it. 

Time that nobody allocates does not vanish. It gets spent by default, and the defaults are predictable, the most common being more of the existing job. When HMRC ran its randomised Copilot trial, most users said the time went back into routine work, into the quality of what they produced and into their own wellbeing. None of these are a bad outcome. But neither was chosen, and neither was measured as a benefit the investment was initially meant to deliver. 

Some of the saving is used up by the AI itself. Workday’s research with Hanover Research found that a large share of the gained time goes straight back into correcting, verifying and rewriting output that wasn’t good enough, and BCG found that many employees now spend more time directing AI than doing the underlying work. Checking and steering AI may be exactly where human judgement belongs. It still means the headline figure for time saved is larger than what the organisation can realistically redeploy 

More time disappears between the individual and the team. Gartner found that supply chain staff who reported individual time savings showed only a fraction of that gain once it was measured across the team, with no clear improvement in output or quality. Each person got a little faster, but the handoffs, meetings and approvals between them stayed as they were, so the gain had nowhere to accumulate. 

Taken together, these patterns explain how an AI programme can report impressive savings that are hard to find anywhere else in the business. The time is real at the level of the individual task. It thins out at every step between that task and a result anyone would notice: into more of the same work, into managing the AI, into handoffs nobody redesigned and into queues where faster work simply waits longer for the next approval. The released capacity needs somewhere to go, and until someone decides where, the default will keep taking it. 

“Free people up for higher-value work” is the line that closes most conversations about AI productivity. It sounds like a decision, which is exactly why it’s so good at ending the discussion. Yet it never says which work, or who would notice if it happened. And it assumes the best use of every released hour is something more strategic, which for many teams is simply wrong. A stretched service team may need the hour for faster responses to customers, while a regulated finance team may need it as more thorough control checks. 

The phrase also papers over a disagreement about whose priorities the time should serve. HR may want freed hours spent on projects and development outside the core role, while a line manager answerable for this quarter’s numbers wants them back on the day job. Left unresolved, that argument settles itself: people follow whatever their targets reward, and in most teams that means more volume. That is still an allocation, just not one anyone made on purpose. 

The alternative is to treat released capacity as a budget line: something with a source, an owner and an intended use, reviewed like any other allocation of a scarce resource. Call it the Capacity Budget. Its purpose is modest and practical. It forces a decision that otherwise gets made by accident, one workflow at a time. 

For each significant workflow, the budget records four things. The first is the capacity the organisation can count on once review, rework, fragmentation and downstream bottlenecks are netted off, which is usually well below the headline saving. The second is the destination, meaning what that capacity is being used to buy. The third is the measure that will show whether it arrived there, and the fourth is the person accountable for the result and the date the allocation will be reviewed. 

Every capacity budget also contains a line nobody writes down: the default, where time goes when no one has decided. In most organisations today that unwritten line holds nearly everything, absorbed as extra volume, rising expectations and gradual workload creep. The discipline lies in shrinking it. AI hands leaders a capacity budget; if nobody allocates it, the organisation spends it by default. 

Run consistently, the budget becomes a learning cycle rather than an annual exercise. Leaders sense where AI is releasing time, interpret how much of it is usable, decide where it should go, then act and learn whether it landed. An allocation that produces nothing measurable after a quarter should move, and so should one that has pushed a team into overload. 

A budget is only as reliable as its opening balance, and HMRC offers a useful model of restraint. It gave out Copilot licences at random, measured the results cautiously and found a saving of roughly an hour a week for a full-time employee. It also cut the self-reported savings by around a fifth, to allow for staff who didn’t use the tool and for keener users being more likely to reply. The wider cross-government trial, which HMRC’s early results informed, had reported average savings of about 26 minutes a day. The two studies covered different populations, but the gap shows how far method alone can move the number a business case rests on. – no idea what is being said here 

HMRC also described the benefit with care. Its analysts valued the rollout as capacity-generating net productivity benefit, which is a different claim from a cash saving. Capacity becomes money only when something in the cost base changes: hiring avoided, overtime reduced, contractors released or roles removed. Finance teams that book theoretical hours as savings make the investment case look stronger today and the P&L look weaker later.  

That argues for measuring net capacity at workflow level rather than adding up individual estimates. Five minutes saved fifteen times across a week rarely becomes an uninterrupted hour for anything useful unless roles and workflows are redesigned to consolidate it. The opening balance should reflect the time the organisation can use, whatever employees feel they have saved. 

In professional services, the allocation decision has an external party with opinions. Thomson Reuters’ 2026 Future of Professionals research found that most corporate buyers regard AI-enabled quality improvements from their advisers as very important or essential, yet only 6% say most or all of their providers deliver them, and close to a third are reconsidering relationships with firms falling behind. Thomson Reuters sells AI tools to these professions, so the figures deserve some caution, but the direction is hard to dismiss. Clients increasingly expect to see part of the productivity dividend.  

For firms whose economics rest on billable time, capacity allocation becomes a pricing and relationship question. The time AI saves can be kept as margin, returned as better quality, faster turnaround or lower fees, or reinvested in new propositions and in developing junior professionals, whose training has traditionally come from the very work AI now speeds up. Each of those is defensible. A firm that has not chosen will find its clients drawing their own conclusions about which one it picked. 

Some organisations are writing the destination down before the savings arrive. The Ministry of Justice reports that pilots across HR, policy, probation and the courts service saved staff around half an hour a day, and its AI action plan commits, in its second year, to reinvesting time saved in better experiences for the public and for staff. That remains a broad intention, but it is a stated allocation with a direction attached, which is more than most business cases carry.  

Survey evidence suggests the organisations getting financial results think about capacity in a similar way. McKinsey’s 2026 State of AI survey found that its small group of high performers is far more likely than other organisations to have fundamentally redesigned workflows around AI, and more likely to pursue growth and innovation alongside efficiency. This is an association rather than proof of cause. It is consistent with the view that redesigning the workflow is where released capacity gets consolidated and pointed somewhere.  

Workforce reduction is one legitimate line in the budget, and the evidence suggests it is rarely the main one. In EY’s US AI Pulse survey of 500 senior leaders, organisations seeing AI productivity gains most often reinvested them in expanding and developing AI capability, cybersecurity, R&D and reskilling, while only 17% said the gains were being used to reduce headcount. Where the workforce line is used, leaders should be precise about which version they mean, because vacancy avoidance, redeployment, natural attrition and structural reduction carry very different consequences for the people involved and for the organisation’s capacity to grow later.  

Every AI investment case should answer the following question: what will the capacity created be used to buy, and who decides? Leaving the answer to individual employees and line managers is an allocation too, just one nobody would defend in a board meeting. The business leader who owns the workflow should own that decision, with finance separating capacity from cash, HR translating it into roles and skills, and operations redesigning the work so the time can be used. 

Some released capacity should be protected for learning. AI will keep changing the work, and an organisation that converts every saved minute into more output removes the space its people need to adapt to the next change. That trade-off will not appear in this year’s savings figures. It will appear later, in how quickly the organisation can absorb what comes next. 

AI is changing how long work takes, and it will keep doing so. It cannot decide what the organisation should do with the difference. The technology creates the capacity; management decides whether it becomes value.