Methodology: how the calculator works
The AI Automation ROI calculator is deliberately simple and transparent. This page sets out every formula and assumption so you can check the results, adapt them, or rebuild them in a spreadsheet.
Inputs
| Input | Meaning | Default |
|---|---|---|
| People | Number of people who do the selected tasks | 5 |
| Hourly cost | Fully loaded cost of one hour of their time | 40 |
| Working weeks | Weeks worked per year after leave and holidays | 46 |
| Task hours | Hours per person per week on each selected task | 13 in total |
| Time saved % | Reduction in time on those tasks when AI-assisted | 35% |
| Adoption % | Share of people who actually use the tool | 80% |
| Ramp-up | Months during which benefits are counted at 50% | 2 |
| Setup cost | One-off cost: design, integration, data cleanup, training, pilot | 5,000 |
| Licence per user | Monthly tool cost per active user | 30 |
| Other running costs | Monthly API usage, maintenance, support | 150 |
Formulas
Key assumptions and limits
- Time is valued at cost. The calculator values saved hours at what they cost the business. It does not estimate extra revenue, and saved time only becomes a financial gain if it is redeployed.
- Savings apply to selected tasks only, not to whole jobs. This matches the research: large gains on specific tasks, modest gains across whole jobs.
- Flat over time. Years two and three repeat the year-one run rate. In practice benefits may grow as processes improve, or fade if tools are not maintained. Prices and wages are not inflated.
- No discounting. Values are not discounted to present value. For a three-year horizon at typical discount rates this changes results modestly; use a spreadsheet if you need NPV.
- Quality effects are excluded. Fewer errors or faster customer response may add value; more errors may subtract it. Neither is in the number.
- No tax, financing or depreciation effects.
Why the default time savings are 35%
Controlled studies of AI assistance on specific work tasks have found average gains of about 14% (customer support) to 25% (consulting tasks), with larger gains for less experienced workers, and a slowdown of 19% for experienced developers in one trial. Highly repetitive tasks — formatting, extraction, routine drafting — can see much larger savings. A moderate 35% default on targeted repetitive tasks, with 20% and 50% presets either side, reflects that range. The evidence is summarised in what the research says.
Privacy
All calculations run in your browser. Nothing you enter is sent to us. If you use "Copy link", your inputs are placed in the page address so the scenario can be shared.
Changes and corrections
We review the defaults when significant new research is published. If you think a formula or source is wrong, please tell us.
Sources
- Brynjolfsson, Li & Raymond, "Generative AI at Work", NBER Working Paper 31161 (2023)
- Dell'Acqua, McFowland, Mollick, Lakhani et al., "Navigating the Jagged Technological Frontier", Harvard Business School working paper (2023), as reported by The Harvard Crimson
- Bick, Blandin & Deming, "The Impact of Generative AI on Work Productivity", Federal Reserve Bank of St. Louis (Feb 2025)
- Bick, Blandin & Deming, "The State of Generative AI Adoption in 2025", Federal Reserve Bank of St. Louis (Nov 2025)
- METR, "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity" (July 2025)