Every board I speak to right now asks the same question: what is the ROI?
Last year MIT found that 95 per cent of enterprise AI projects show no measurable return.[1] Boards have been quoting it ever since as a reason to wait.
They have misread it entirely.
The MIT study measured ROI six months post-pilot. It used P&L impact as the metric.[1] Berkeley’s response put it well: we are applying “industrial-era metrics to a cognitive-era transformation”.[2] The problem is not that AI does not work. It is that the ruler is wrong.
Lessons from electricity and the iPhone
Nobody in 1890 could model the ROI of electrification. The value chain had not been reimagined yet. The ROI of electricity was not in replacing candles.
It was in enabling factories, cities, and eventually everything that came after. You could not see that from 1890. The calculation would have been wrong.
The iPhone launched in 2007. The most valuable things it enabled, Uber, contactless payments, the entire remote work economy, were not on any product roadmap. An ROI model at launch would have looked at calls and emails. It would have missed everything that mattered.
Three years in
ChatGPT launched November 2022.[3] We are little more than three years into a transformation that will take decades to play out. Demanding ROI proof at year three is premature optimisation.
The real question is not what AI has returned so far. It is what it will cost you to catch up in 2027 when you look back at this moment and have to explain why you waited.
If your board is weighing up how to judge AI investment, I would be happy to help frame the question. There is more about how we work on our services page, or you are welcome simply to drop me a line.
This article was first shared as a post on LinkedIn.
Sources
- MIT NANDA, The GenAI Divide: State of AI in Business 2025 (methodology: “ROI impact measured 6 months post-pilot”).
- UC Berkeley Executive Education, Beyond ROI: Are We Using the Wrong Metric in Measuring AI Success?, 17 September 2025.
- OpenAI, Introducing ChatGPT, 30 November 2022.