Introduction

How much does AI actually bring in for your company? It is the question every SME leader asks before investing. The risk is well known: projects launched without a quantified objective, which cost a lot and which no one can say were worthwhile. The answer comes down to one discipline: calculating the ROI of AI before, during and after the project. It is one of the pillars of our guide AI strategy for SMEs: key steps and priorities.

This guide provides a concrete method for measuring the return on investment of your AI projects, with a formula, a worked example and the mistakes to avoid. The systems we build for each business function, presented on the page diagonall AI systems, all start from this calculation.

Why measuring AI ROI is essential

The problem with projects that have no quantified objective

As early as 2024, Gartner predicted that at least 30% of generative AI projects would be abandoned after the proof of concept, notably because of rising costs and poorly defined business value. Many companies jump in because "everyone is doing it", without saying what they expect.

The classic mistakes:

  • launching a tool without measuring beforehand the time spent on the task;
  • counting only the provider's invoice and forgetting the rest;
  • confusing usage ("the teams use it") with results ("we save time or money");
  • waiting until the end of the project to ask the question.

How to calculate the ROI of an AI project

The basic formula

ROI is simple to calculate:

ROI (%) = (gains over the period − costs over the period) / costs over the period × 100

It is often complemented by the payback period: the number of months needed for cumulative gains to cover the initial investment. For AI, each component needs to be broken down.

The costs to take into account

Type of cost Examples
Initial development Scoping, POC, integration with your tools
Subscriptions and usage Tool licenses, AI model consumption, hosting
Training User teams, managers
Maintenance Monitoring, fixes, enhancements, model updates
Internal time Coordination, testing, adoption (often underestimated)

Ask each provider for a separate quote for each of these lines, and add your own estimate of internal time.

The gains to measure

The gains from AI fall into three categories:

1. Productivity gains (the easiest to measure)

  • hours saved per week on a task, multiplied by the fully loaded hourly cost;
  • additional volume handled without hiring.

2. Cost reductions

  • errors avoided (disputes, credit notes, re-entry of data);
  • fewer external services or overtime hours.

3. Revenue gains (harder to attribute)

  • a better conversion rate thanks to faster follow-ups;
  • better-served customers, and therefore more loyal ones.

For a first calculation, stick to productivity and cost gains: they can be measured without debate.

A worked example: the ROI of a customer service assistant

The example below is fictional. It is meant to show how the calculation works; replace each assumption with your own figures.

Context

An online retail SME receives 100 customer requests per working day. A good share are simple questions: order tracking, returns, frequently asked questions. Each simple request takes an average of 6 minutes to handle.

Assumptions

  • The assistant prepares a reply for 50 simple requests per day, validated in 1 minute by an agent instead of 6.
  • Gain: 50 × 5 minutes = 250 minutes per day, or about 4 hours.
  • Over 220 working days: about 900 hours per year.
  • With a fully loaded hourly cost of 35 euros: about 31,500 euros of time freed up per year.

Costs (assumptions)

  • Setup in the first year: 15,000 euros.
  • Subscriptions, usage and maintenance: 6,000 euros per year.
  • Internal time and training: 3,000 euros in the first year.

Calculation

  • Year 1: costs of 24,000 euros, gains of 31,500 euros. ROI = (31,500 − 24,000) / 24,000, or about 31%.
  • Year 2: costs of 6,000 euros, gains of 31,500 euros. ROI of about 425%.

The important point is not the result of this example, but the method: with different assumptions, year 1 ROI can be negative. It is often the second year that makes the difference, once the initial investment has been absorbed. And the time freed up only has value if it is reinvested in useful tasks.

The 5 essential indicators

1. Time saved per process

Measure the time before and after for each automated task. It is the most reliable indicator.

2. Adoption rate

An unused tool has zero ROI. Track the share of teams that really use it.

3. Error rate

Does AI reduce errors? Compare before and after.

4. User satisfaction

Ask the teams: "On a scale of 10, would you recommend this tool to a colleague?"

5. Cost per case

Divide total costs by the number of cases handled. AI should bring this ratio down.

The mistakes that drag down ROI

Mistake 1: underestimating hidden costs

Internal time spent (meetings, coordination, testing) is rarely counted. Yet it can weigh heavily in the total cost.

Mistake 2: not training the teams

A poorly used tool brings in nothing. Plan a training budget from the start. This is also what the AI Act expects: since July 27, 2026, its Article 4 requires taking measures to support the AI literacy of teams.

Mistake 3: aiming too big from the start

Start with a POC on a specific use case, validate the ROI, then scale up. See AI POC: method, budget and KPIs in business.

Mistake 4: forgetting maintenance

An assistant needs to be updated, models change, and so do prices. Plan an annual monitoring budget.

How diagonall approaches ROI

We never launch a project without having agreed with the client on the success indicators and their baseline value. In practice:

  • measuring the time spent on the task before the project;
  • a detailed cost breakdown, line by line;
  • tracking the indicators during the pilot, then after deployment.

To build your own analysis, see also Prioritizing your AI use cases: the matrix.

Frequently asked questions

What ROI can an SME expect from an AI project?

There is no universal figure: it all depends on the volume handled, the time saved per case and the cost of the project. On a frequent, well-scoped task, ROI generally becomes clearly positive once the initial investment has been absorbed.

How long does it take to see the ROI of an AI project?

The first gains from a pilot can be measured within a few weeks. The full return on investment often takes 12 to 24 months for a broader project. Simple automations can pay off faster.

How do I convince my management to invest?

Present a quantified case: the cost of the current situation, the expected gains based on your assumptions, and a payback timeline. Also show what happens if the project fails: a POC limits the loss.

Should freed-up time be counted as a gain?

Yes, provided you say what it will be used for. A time saving that changes nothing in the organization remains theoretical.

Conclusion

Calculating the ROI of AI is not optional: it is the condition for avoiding projects that stall. Set your indicators before you begin, measure regularly, and do not hesitate to change course if the results are not there.

Book a demo