Free tool
AI ROI calculator.
Run the numbers on a proposed AI project before you commit. Payback period, annual benefit, and the assumptions you'll need to defend in a board meeting. No download, no email.
Inputs · adjust the sliders
A realistic estimate after the system is live and adopted.
How many people in the team the system supports.
Salary plus on-costs ÷ working hours. Roughly 1.4× the headline salary divided by 1,800 hours.
One-off cost to design, build and deploy.
Hosting, API costs, monitoring, light maintenance.
Estimate
Annual time freed
2,080 hours
Annual gross saving
£93,600
Net annual benefit (year one)
£57,600
Year-one saving minus build cost and one year of running cost.
Payback period
4.1 months
Months until cumulative savings cover the build cost.
If the payback period is under 12 months and you have the team time to run a pilot, the project is probably worth the diagnostic to firm up the assumptions.
How to use it
Be conservative. Then halve it.
The biggest mistake in AI business cases is generous savings estimates. The system saves five hours a week on paper, but only the first three hours show up in practice because the team has to learn it, edge cases come back to the human, and review still happens. Halve your initial number and you'll usually be closer.
The second biggest mistake is forgetting running costs. AI systems aren't free to run — API calls, hosting, monitoring, retraining. Budget at least 15-25% of build cost annually unless the system is unusually simple.
The third is treating saved hours as cashable savings. They're not, unless you genuinely reduce headcount or redeploy the time to revenue work. For most teams, the win is capacity rather than payroll. Be explicit about which it is when you take this to a board.
Benchmarks
Real ranges from our projects.
| Project type | Typical cost | Typical annual saving | Payback |
|---|---|---|---|
| Document processing automation | £15–25k | £40–60k | 4–6 months |
| Predictive maintenance | £20–35k | £50–80k | 5–8 months |
| Quality control (computer vision) | £25–40k | £60–100k | 5–8 months |
| Demand forecasting | £15–30k | £30–50k | 6–10 months |
| Internal retrieval / RAG | £25–45k | Often non-cash (time, errors avoided) | 6–12 months |
Ranges, not guarantees. Specifics depend on data quality, integration complexity, and how mature the underlying process is before you add AI.
When the maths doesn't work
Four reasons to walk away.
-
The problem is too small. Saving two hours a week for one person doesn't justify a £20k build, no matter how clever the system.
-
Your data isn't ready. Data cleanup is real engineering work. If the cleanup costs more than the savings, the savings aren't real.
-
The process changes too often. If the rules behind the work shift every quarter, the system spends its life chasing the spec.
-
Your team won't use it. Adoption failure is the silent killer. If the people doing the work aren't in the design, the system will go live and quietly not be used.
Numbers stack up?
Good — the diagnostic is the cheapest way to firm them up. Two weeks, fixed price, board-ready report. If we can't make the case ourselves at the end of week one, we tell you and refund the second week.