A company introduces AI and reports that employees have saved 500 hours in a year. The project is described as a success. Yet staff costs are unchanged, revenue has not increased and the software now adds another expense.

Was the investment worthwhile? Possibly. But the answer cannot be found in the hours saved alone. I would want to know what the business actually gained from those hours, what it cost to achieve them and whether the result can be repeated.

The gap between productivity and financial performance

An OECD survey of more than 5,000 SMEs across seven countries, including the UK, found that among generative AI users 65% reported improved employee performance, whilst 26% reported increased revenue. These are self-reported outcomes collected in 2024, not measured UK productivity gains or evidence that AI caused a particular financial result. [1] Separately, qualitative interviews in UK government research found that many businesses had no formal metrics for attributing changes in revenue or productivity to AI. [2]

There is no contradiction in a business becoming more productive without reporting higher sales. Faster administration might improve service quality, reduce pressure on employees or allow growth without extra recruitment. Those are legitimate outcomes. However, they are different from an actual reduction in cost or an increase in profit.

The distinction matters because a spreadsheet can make a theoretical saving look rather more substantial than it is.

The 20-person business: six minutes at a time

Article 3 considered a hypothetical 20-person equipment servicing company, testing AI-assisted preparation of engineer service reports. The business produces 40 reports each week. Preparation and checking take 18 minutes per report before the pilot and 12 minutes afterwards, including human review and corrections.

That difference releases four hours each week, or approximately 184 hours over 46 working weeks. At an illustrative fully loaded staff cost of £30 an hour, the notional value of that time is £5,520 a year. These figures are assumptions for the example, not observed client results.

It would be tempting to enter £5,520 as a saving. But unless employment costs, overtime or another actual expense fall, the amount paid by the business has not changed. What has changed is the capacity available to perform other work.

The Distinction That Matters

184 hours released × £30 per hour = £5,520 of notional staff time. It is not £5,520 of cash savings. The financial case depends on how that capacity is used, any avoidable costs, and the additional expense of the AI solution.

Where might the value appear?

There are several possible routes. The company might avoid planned overtime or temporary assistance, with a real reduction in expenditure. It might use the available hours to complete additional chargeable work, improving contribution to profit if there is sufficient customer demand. Faster and more reliable reporting might also improve customer retention, although that benefit would need its own evidence. Alternatively, employees could spend less time on repetitive administration and more on service quality or training. That may be worthwhile, but should not be recorded as cash profit.

For businesses charging by the hour, there is a further complication. Completing the same work faster can reduce billable time unless pricing and delivery arrangements change. A fixed-fee service may benefit differently because better efficiency can improve margin. AI does not remove the need to understand the commercial model.

A small business case, with the assumptions visible

Suppose the pilot requires £1,800 of initial configuration and training, plus £1,200 in licences, monitoring and support over the first year: £3,000 in total. The £1,200 allowance is additional to the report-checking time already included in the six-minute reduction.

If the 184 hours simply become unused capacity, there is no demonstrated direct financial return against that £3,000 cost. There may still be service or staff benefits, but these require separate assessment.

If the company can genuinely redeploy half the released time — 92 hours — into additional customer work generating £55 of contribution per hour after incremental delivery costs, that could produce £5,060 in additional contribution. Less the £3,000 project cost, the illustrative first-year net benefit would be £2,060. This assumes that the extra work exists, the staff have the necessary skills and the additional contribution would not have arisen without the change.

A different outcome might be £4,000 of evidenced temporary labour or overtime genuinely avoided. Against the same £3,000 project cost, that would leave £1,000. Those two routes must not be added together if they represent alternative uses of the same released hours.

Measurement needs to survive the pilot

I would keep the benefits record fairly simple. Establish the baseline using completed work, rather than a handful of impressive demonstrations. Measure the total effort after implementation, including checking, correction and exception handling. Record licence, integration, training and ongoing support costs. Then agree who is responsible for turning released capacity into a business outcome.

The comparison must also account for what would have happened anyway. Seasonal demand, staffing changes, different job complexity or an improved template can all affect the figures. A small pilot may not support a precise causal claim, but a representative sample, consistent definitions and a comparison period will produce a more useful management decision than an isolated estimate of minutes saved.

Quality matters too. A report prepared in twelve minutes but requiring a later customer correction is hardly an improvement. Turnaround time, factual accuracy, rework, customer response and employee workload belong alongside financial measures.

The most useful question comes after the time saving

The decision should not be whether AI has produced an impressive number of hours. It should be whether the business has improved a result that matters — and whether the improvement justifies the cost, risk and effort involved.

My view is that this is where many AI projects need more discipline. The technology may be capable of saving time; realising the value is a management responsibility. That is a less exciting headline than announcing hundreds of hours saved, but considerably more useful to a founder or board.

The next article will consider a related workforce question: if AI increasingly carries out junior-level tasks, how will the next generation develop the experience and judgement required for more senior roles?

Richard James Bowles is Founder & Principal of Richard James Advisory Ltd, providing independent senior technology, transformation and board-level advice to founder-led and growing businesses.

Sources and methodological notes

[1] OECD, Generative AI and the SME Workforce: New Survey Evidence (5 November 2025). Late-2024 survey of more than 5,000 SMEs in seven countries, including the UK; self-reported outcomes. Source

[2] DSIT, AI Adoption Research (28 January 2026, updated 13 February). 2025 UK fieldwork; qualitative interviewees often lacked formal measures for attribution. Source

Worked example: all prices, volumes, time savings, cost assumptions and contribution scenarios are hypothetical. Financial comparisons are illustrative, not forecasts, client outcomes or evidence of typical AI returns.