“How much will it bring in, and from when?” Your finance director asks this at the first presentation, and rightly so. AI project ROI: three words that every management committee ends up saying, often without a common method for answering them.
ROI, or return on investment, compares what a project brings in with what it costs. This article gives you a five-step calculation method, a fully fictional example to follow the reasoning, and the pitfalls that distort the result.
A simple formula, tricky terms
The formula fits on one line: ROI = (gains − costs) ÷ costs. A positive result means the project returns more than it costs over the chosen period. The payback period indicates, for its part, after how many months the cumulative gains cover the initial outlay.
The difficulty does not lie in the division. It lies in what you put into each term. An AI project generates recurring costs that are easily forgotten, and gains that are just as easily overestimated.
The method in five steps
1. Measure the starting point
Choose a specific process and record four data points: the monthly volume handled, the time spent per unit, the error rate, and the fully loaded hourly cost, meaning salary and social charges, of the people involved. Without this baseline, no gain can be demonstrated.
2. List all the costs
Separate the initial investment from recurring expenses. To price the external items, request a detailed quote that distinguishes the build from the running costs.
- Initial costs: scoping, data preparation, development, integration with your tools, testing, training.
- Recurring costs: model licences or usage fees, hosting, maintenance, supervision.
- Internal time: your teams spend hours on the project, then on checking the results. Count them.
3. Estimate gains by category
Sort gains into four families: time freed up, errors avoided, additional revenue, and avoided expenses. Distinguish measurable gains, which you can find in the accounts, from qualitative gains such as customer satisfaction or team comfort.
4. Apply a realisation rate
No tool handles every case from day one. Adjust the theoretical gain by the share of cases actually automated and by the pace at which users adopt the tool.
5. Calculate over a period, with three scenarios
Choose a duration consistent with the life of the tool, then run a cautious scenario, a central scenario and a favourable one. You then present a range, not a promise.
A worked example, entirely fictional
We use example values, chosen to simplify the calculation. They describe no real company and are not a market benchmark. A trading company wants to automate the entry and checking of its supplier invoices.
- Starting point: 2,000 invoices per month at 6 minutes each, or 200 hours. At 20 currency units per loaded hour, processing costs 4,000 per month.
- Theoretical gain: the tool reduces the time to 3 minutes, human checking included. The team frees up 100 hours, or 2,000 per month.
- Gain retained: you apply a realisation rate of 80%, because the team will still process some invoices by hand. The gain falls to 1,600 per month, or 19,200 per year.
- Costs: 15,000 initial investment, then 400 per month to run, or 4,800 per year.
In the first year, gains reach 19,200 against 19,800 in costs. The ROI comes out at −3%: the project has not yet paid for itself. Over two years, gains rise to 38,400 against 24,600 in costs, and the ROI reaches 56%. The net monthly gain of 1,200 covers the initial investment in 12.5 months.
Now redo the calculation with a 60% realisation rate. The monthly gain drops to 1,200, the net gain to 800, and the payback period approaches 19 months. The conclusion changes with the assumption. That is why you present three scenarios rather than a single figure.
Six pitfalls that distort the calculation
- Counting freed-up hours as savings without saying what the team will do with them. Time saved is only worth something if it serves another useful task.
- Forgetting recurring costs. Model usage, maintenance and supervision last as long as the tool does.
- Ignoring human review. If a person checks every result, include that time in the calculation.
- Assuming full and immediate adoption.
- Counting the same gain twice, for example the time saved and the drop in errors that results from it.
- Comparing the project only with doing nothing. Compare it also with a simple process improvement, without AI.
Reading the result and deciding
A negative first-year ROI is not enough to rule out a project, as the example shows. Look at the payback period, the sensitivity to assumptions, and the share of gains you know how to measure.
If the cautious scenario stays negative over the whole period, reduce the scope or choose another use case. If the calculation holds, set an ROI review date with real figures. You then replace the assumptions with measurements, and decide whether to extend, correct or stop.
What changes by sector
- Transport and logistics: gains come from kilometres, fuel and avoided delays. Your route data provides the starting point.
- Accounting firms: the firm sells time. Measure the data-entry hours freed up and decide whether they feed billable advisory work or new clients.
- Car dealerships: the workshop and after-sales offer regular volumes. Track the rate of kept appointments and the response time to requests.
Costing your project with AISYSNEXT
A credible ROI calculation requires precise costs. Our agency, based in Tunis, works in artificial intelligence and BI, ERP and CRM, and custom development. Our free quote request lets you describe your process and receive a proposal that separates initial costs from recurring costs.
Still unsure which use case to cost first? Contact our team to discuss it before requesting any pricing.
Frequently asked questions
Over what period should you calculate the ROI of an AI project?
Use the period during which you plan to use the tool without a major overhaul. Too short a period penalises the initial investment. Too long a period rests on fragile assumptions.
How do you value a quality or customer satisfaction gain?
Link it to an indicator you already track: complaints, credit notes, response times, renewals. If you find no measurable link, present this gain separately, without adding it to the calculation.
What if the ROI comes out negative?
First check your volume and cost assumptions. Then reduce the scope to the part of the process where most of the time is lost. If the result does not improve, the project does not deserve priority.
