AI gave employees time back. What did the business do with it?

There is growing evidence that AI can make individuals more productive. A field experiment involving 7,137 knowledge workers across 66 firms examined the effect of giving employees access to a generative AI tool embedded in applications they already used for email, meetings and writing.

During the second half of the six-month experiment, workers who used the tool spent around two fewer hours per week on email and reduced the amount of work they did outside normal working hours. But the researchers did not detect broader changes in the quantity or composition of work resulting from individual access to the tool.

That distinction matters. People can become more efficient without the organisation fundamentally changing how work gets done. And that is not necessarily a problem. Less administrative effort, fewer hours worked outside normal schedules and a better employee experience can all be worthwhile outcomes.

The problem begins when those improvements are treated as though they automatically become financial value. If an employee saves five hours a week using AI, the organisation has created capacity. It has not necessarily created a cash saving. The employee is still employed.

Their salary is still being paid. The operating model may be unchanged. And the five hours that have been released may or may not be used differently.

Multiplying those five hours by an hourly employment cost can indicate the scale of the capacity created. It does not, by itself, show what value the business realised. Those are different things.

When AI reduces the effort needed to perform work, management has a decision to make. What should happen to the capacity that has been released? The business might use it to process more transactions, reduce a backlog, improve service, strengthen controls, pursue additional revenue or redirect people towards work that previously received too little attention.

It might reduce overtime or external contractor expenditure. If demand is growing, it might avoid some of the additional hiring that would otherwise have been required. It may also deliberately allow some of the benefit to flow to employees through reduced \ administrative burden or a more sustainable workload.

All of these can be legitimate outcomes. But they are not economically equivalent. A reduction in current expenditure is different from avoiding a future cost. Avoiding a future cost is different from creating additional capacity.

And creating additional capacity is different again from using that capacity to generate incremental revenue or improve customer outcomes. The discipline is not to force every benefit into a financial category. It is to make the value claim match what actually happened.

Consider an insurance company introducing AI into its claims operation. If employees can process claims faster, the technology has created capacity within the team. What happens next depends on the circumstances of the operation. If there is a growing backlog, the company might use that capacity to process more claims without increasing staffing proportionately.

If claims volumes are stable, it might redirect some of the capacity towards complex cases, customer communication or fraud investigation. If demand is expected to grow, the business may be able to absorb some of that growth without making the hires previously anticipated.

And if the operation already has spare capacity, making existing activities faster may produce relatively little economic value unless something else changes. The same productivity improvement can therefore have very different value in two apparently similar businesses.

It depends on the constraint the capacity is being used to address. That is why an AI value case should start with the business constraint, not the percentage of time a technology claims it can save.

Before approving a major AI investment, the sponsoring business leader should be asked a simple question: If the technology performs exactly as promised, what will actually change in the operation? Whatwill the team do differently? Which constraint will be removed?

What capacity will be created? How will that capacity be used? And what outcome should improve as a result? Those answers should sit inside the investment case alongside the technology costs and expected productivity gains.

That changes the logic of the business case. Instead of starting with hours saved and trying to convert them into money, the organisation starts with an operating outcome it wants to change and works backwards. Once the solution is deployed, management can then track whether the expected chain of events actually occurred.

Was capacity created? Was it put to the intended use? Did the operational outcome change? And was the outcome worth the full additional cost of achieving and sustaining it, including AI operating, oversight and change costs?

These distinctions matter because value can disappear between each step. AI may save employees time, but that time may never be redeployed. Capacity may be redeployed, but towards work that produces little incremental benefit. An operational metric may improve, but other changes in demand, staffing or process design may explain part of that improvement.

A credible assessment needs to account for those factors before attributing the result to AI. Sometimes the technology will work exactly as expected and the organisation will still discover that the original financial benefit cannot be realised without a broader operating-model change.

That is not a failed AI initiative. It is useful management information. It tells the executive team that the next decision is no longer about the technology.

If the technology performs exactly as promised, what will actually change in the operation?

The question that should sit inside every investment case