Measure the return
AI ROI: measure the return from a bigger business change
AI ROI compares the money gained of an AI project with its full cost. For a service firm, the benefit may come from a new offer. For a large business, it may come from serving more teams or removing delays. Track how the work changes, then show which changes produce a return in money.
Measure the first project honestly. Track what it makes possible next, without counting future results as money already earned.
Start with the full cost
AI ROI (%) = (financial benefit − full project cost) ÷ full project cost × 100
Use the same period for cost and benefit. State whether costs include staff time as well as added spending. If the cost is zero, check for missing costs before trying to calculate a percentage.
The formula cannot tell you whether people use the service or whether it helps them. Keep those measures beside the money.
Review your AI ROI
Use existing service and finance records where possible. Add measures that help you make a decision, rather than building a dashboard nobody uses.
Follow the change through to the business
| Change | What to measure | Financial treatment |
|---|---|---|
| More clients or paid services | Extra work sold and delivered, less all direct costs. | Count the extra contribution, after any existing work it replaced. |
| Faster internal service | Waiting time, completed work and effects on the receiving team. | Use a money value only when you can support it. |
| Less staff time on a task | Time after checking, correction and upkeep. | Wages are not saved if the wage bill stays the same. |
| Lower bills | Spending that actually fell for comparable work. | Count the supported saving, less new costs. |
| A foundation for another project | Which records, rules or connections were reused. | Show reuse separately; do not count hoped-for future sales. |
Contribution means the money left from a sale after the costs of doing that work. If freed hours let you earn that contribution, do not count them again as wages saved.
Compare work of similar size and difficulty. Record other changes, such as more staff or simpler requests, so AI does not receive all the credit.
Worked example: a new expert review service
Imagine a specialist firm using AI to prepare client plans for expert approval. It uses some of the time released to offer extra paid reviews. This is an invented example, not a client result or a LeverageAI quote.
By month six, 60 comparable plans take 160 staff hours instead of 240. Both totals include checking and correction. Keeping source records current takes another eight hours, leaving 72 hours for other work.
Two extra reviews use 24 of those hours. The example assumes the old workload prevented those sales and that they replace no existing paid work.
Count the costs
| Cost | Calculation | Amount |
|---|---|---|
| Build and external setup | One-off | £48,000 |
| Staff setup, testing and training | 240 hours at £50 | £12,000 |
| Total setup | £48,000 + £12,000 | £60,000 |
| Monthly software, support and upkeep | Includes 8 upkeep hours at £50 | £6,000 |
| Full cost through six live months | £60,000 + 6 × £6,000 | £96,000 |
The running cost excludes the extra reviews’ delivery costs, which are deducted below. Staff still spend time checking plans. We claim no wage saving for that time.
Count the benefit
Each extra review sells for £12,000. Twelve staff hours cost £600 and other direct delivery costs are £2,400. That leaves £9,000 per review.
The service needs time to find buyers. It delivers no extra reviews in month one, one in month two and two a month from month three.
| Month | Extra reviews | Contribution | Running cost | Net total after setup |
|---|---|---|---|---|
| 1 | 0 | £0 | £6,000 | −£66,000 |
| 2 | 1 | £9,000 | £6,000 | −£63,000 |
| 3 | 2 | £18,000 | £6,000 | −£51,000 |
| 4 | 2 | £18,000 | £6,000 | −£39,000 |
| 5 | 2 | £18,000 | £6,000 | −£27,000 |
| 6 | 2 | £18,000 | £6,000 | −£15,000 |
Nine reviews × £9,000 = £81,000 financial benefit Net benefit = £81,000 − £96,000 = −£15,000 Six-month ROI = −£15,000 ÷ £96,000 × 100 = −15.6% (rounded)
The latest month contributes £12,000 after running costs. The project has still not paid back its setup cost. If demand and costs stay steady, it would break even during month eight.
These figures are before tax and financing. The six live months start after setup. Include the build period when planning cash and judging how long the project takes to pay back.
Decide whether to expand
For a large business, judge the first rollout before adding more regions. Check whether the new teams face the same work and rules. Include the cost of adapting, training and supporting them.
For a service firm, check repeat demand, client results and expert workload before selling to more clients. A busy launch month does not establish a lasting service.
Worked example
An ROI review after six live months
All figures and decisions are invented.
Period: setup plus six full live months
Financial benefit: £81,000 from nine extra reviews
Full cost: £96,000, including staff setup time
Net benefit: −£15,000
ROI: −15.6%, rounded
Latest month: £18,000 contribution less £6,000 running costs = £12,000
Payback: during month eight only if those monthly figures continue
Decision
Continue within the agreed budget and check demand and quality again.
Do not approve wider expansion from one strong month.
Next possible use
Progress reviews using the approved plans and existing client records.
Not yet built or sold. No future sales counted in this result.
Price the extra data work, testing and support before deciding.
Owner: service lead, with finance checking the figures
Next review: end of month seven
Change the plan if quality drops or the expected demand does not hold.
Our workshop helps you decide what gains are worth aiming for. During the build project, we map the current work and agree how to measure the change. Delivering a workflow within 90 days is not a promise of payback within 90 days.
Further reading
- National AI Centre: measuring return on investment
- Journalist-request case study
- Lean AI: the method behind our tools
Written by Said Lopez, founder of LeverageAI and author of Lean AI. Our guides draw on the method in the book and our work with clients.