Last reviewed:
What is AI ROI? Definition and pitfall to avoid
AI ROI, modelled on classic financial ROI, is a deceptively appealing metric for evaluating an AI deployment. Three structural confusions invalidate it: underestimation of hidden costs, lack of robust pre-AI baseline, and concentration on operational efficiency to the detriment of the other two dimensions of value.
The AI ROI metric is widely repeated in business press in 2026, but it masks three major methodological confusions. First confusion: visible costs (subscriptions, licences, API consumption) are often accounted for, hidden costs are not. Application integration, operational supervision, team time to adapt, production incidents: all items missing from the initial budget, and RAND (2024) notes that organisations often underestimate the time and cost of preparing their data. Second confusion: the pre-AI baseline is rarely rigorously measured. Without precise knowledge of processing time, error rate, or satisfaction before deployment, the improvement measured afterwards is only an impression. Third confusion: only operational efficiency (time gained, volume processed) is reported. The other two dimensions of value (decision quality, strategic option) are passed over in silence, although they are often more durable and harder for competitors to reproduce.
Concrete example
Illustrative case. A customer service company announces internally an AI ROI of 233% on its first deployment, calculated by relating the net gain (1.3 million euros of productivity gains over 12 months, minus the investment) to the declared investment (390,000 euros). Three elements are absent from the calculation. The 280,000 euros of internal team mobilised for 9 months. The 90,000 euros of supervision and governance in the first year. The baseline of processing time before deployment, which had never been formally measured. ROI after correction: about 71% (1.3 million minus 760,000 euros of costs, divided by 760,000 euros), still positive, but more than three times lower than the initial communication.
FAQ
Why be wary of a reported AI ROI figure?
Because the number often hides three confusions. Hidden costs (integration, supervision, team adaptation time, production incidents) are left out. The pre-AI baseline is rarely measured rigorously, so any improvement is just an impression. And only operational efficiency is reported, not the other dimensions of value.
How do you evaluate the return of an AI project seriously?
Require a full TCO (visible and hidden costs) and an independent audit of the baseline before any return calculation. Without a baseline measured before deployment, no percentage gain is reliable. RAND (2024) notes that organisations routinely underestimate the time and cost of preparing data.
See also
Further reading
The Root Causes of Failure for Artificial Intelligence Projects, RAND Corporation, 2024
Sources
- The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, Ryseff, De Bruhl & Newberry, RAND Corporation, August 2024 (65 interviews with data scientists and engineers). https://www.rand.org/pubs/research_reports/RRA2680-1.html
- Generative AI at Work, Brynjolfsson, Li & Raymond, Quarterly Journal of Economics vol. 140 n° 2, 2025 (heterogeneity of gains). https://academic.oup.com/qje/article/140/2/889/7990658