Insights / Reporting and AI

How to measure the ROI of AI in a £5m to £50m business

20 October 2026 · 4 min read · EYRA Advisory

UK businesses are using AI more than ever. The British Chambers of Commerce found 54% of SMEs actively adopting AI in 2026, up from 35% a year earlier. Yet very few can say what it is worth. In BGF's September 2026 research with 285 leaders of businesses turning over £5m to £50m, only 6% said they measure AI return on investment confidently, and 34% named ROI uncertainty as a barrier. Only 10% had integrated AI across the organisation.

The gap is rarely about technology. It is about measurement: nobody recorded how long the task took before, the costs of running and checking the tool were left out, and hours saved were counted as money saved. This article sets out a method any finance director can use.

The method in six steps

  1. Define the process and its unit. Pick one process and the unit it produces: an invoice coded, a report issued, a case progressed, a file note written. ROI is far easier to measure per unit than per department.
  2. Record a baseline. For at least a month, measure time per unit, volume per month, error or rework rate and cycle time. Write the definitions down so the after measurement uses the same ones.
  3. Count all the costs. Build or set-up cost, licences, usage charges, the time people spend checking AI output, maintenance as systems change, and the training and change effort. Oversight time is the cost most often left out, and for regulated work it can be significant.
  4. Sort the benefits into types. Hours released, hires avoided, faster cash, fewer errors, and new revenue. Be honest about hours: they only become money if they avoid a hire, reduce overtime or contractors, or are redeployed to work that earns.
  5. Measure again at 30, 60 and 90 days. Same definitions, same units. Early results often dip while people learn the new process, so judge on the trend.
  6. Decide to scale, adjust or stop. Set the decision rule before you start, for example: scale if time per unit falls by at least 40% with no rise in errors.

A worked example

The figures below are illustrative, to show the arithmetic.

A finance manager spends four days a month building the management reporting pack. At a loaded cost of around £300 a day, that is £14,400 a year. After the data feeds and pack are automated, the same report takes half a day a month to check and finalise, or £1,800 a year. The time released is worth £12,600 a year.

If the build and the first year of running and maintenance cost £15,000, the time saving alone pays it back in a little over 14 months. That understates the value, because the pack now arrives in the first week of the month instead of the third, and decisions on pricing, hiring or cash are taken two weeks earlier. Where you can, put a number on that too, for example the cost of a month's delay in spotting a loss-making contract.

The simple ROI formula is: (annual benefit minus annual cost) divided by annual cost. Show year one, including set-up, and a steady-state year separately, because the second is usually much better.

What to include in the business case

Section What it should say
Process and owner The process, its unit, and who is accountable for the outcome
Baseline Current time, volume, error rate and cycle time, with definitions
Costs Set-up, licences, usage, oversight time, maintenance, training
Benefits By type, with the assumption behind each and whether it is cash or capacity
Risks and controls Data handled, where it is stored, who checks outputs
Decision rule The result at 90 days that means scale, adjust or stop

A one-page version of this is enough for most decisions. The discipline matters more than the format.

Why external help is common at this stage

The BGF research found 83% of leaders thought an external AI adviser would be useful, while fewer than 1% had one. Skills gaps (40%), change management (38%) and security constraints (37%) were the most common internal barriers. A short, structured outside view is often the quickest way to get a baseline measured and a business case written without pulling the finance team off month end.

Where EYRA fits

EYRA's review is built to answer the ROI question before money is spent. We map the processes where time goes, measure the baseline, cost the options, and give you a ranked plan with a decision rule for each use case. If you go ahead, the build includes the measurement from day one, so the before-and-after numbers are there for your board, your investors or your next buyer.

Frequently asked questions

How do you calculate the ROI of AI?

Measure the baseline for one process, count every cost including human oversight, value the benefits by type, then divide the net annual benefit by the annual cost.

What is a good payback period for an AI project?

Many operational automation projects aim to pay back within 12 to 18 months. Set your own threshold before you start and judge against it.

Are hours saved the same as money saved?

No. Hours become money only when they avoid a hire, reduce overtime or contractors, or are redeployed to revenue-earning work.

Why do AI pilots fail to show ROI?

Usually because there was no baseline, the costs of checking output were ignored, or the pilot never reached daily use.

What should an AI business case include?

The process and owner, baseline, full costs, benefits by type, risks and controls, and a decision rule for scaling or stopping.

Related reading

Sources

Book a review

Find out where your numbers are getting stuck.

Tell us a little about your firm and what's taking up your team's time. We'll reply within two working days to set up a 20-minute call. If we're not the right fit, we'll say so.