Traditional ROI models, suited for IT investments like ERP systems or any form of work automation, will not capture the value that AI brings. The trap that I see organisations fall into, because we have trained leadership to behave this way, is to measure the efficiency gains of an IT capital investment through metrics of the corporate strategy, as if AI were just another technology platform.
Say, for example, €1 million invested in a dunning system should produce a 10-day improvement in Days Sales Outstanding, freeing up €3 million in cash, providing a quarter-million annual return. Payback in 4 years. That's a reasonable approach and we have trained business and IT leadership to think that way. Traditional IT investments often aim for efficiency gains, with clear metrics and predictable returns.
This is the distinction between tokenised and non-tokenised intelligence. Tokenised intelligence — the kind we buy in software, in consulting days, in pre-packaged solutions — can be measured, accounted for, and amortised. Non-tokenised intelligence — the emergent capability of augmented humans, the unplannable insight, the pattern that can't be commoditised — doesn't fit the spreadsheet. And most organisations are optimised for the wrong kind.
The most common HR approach for measuring investment in people is the Kirkpatrick model: measure reaction, learning, behaviour, results — and only then, ROI. We accept that training ROI is assessed ex-post. We don't pre-approve a leadership programme by demanding a Net Present Value calculation. Why should AI be different?
Think of AI investment akin to investing in human capital. Just as we assess the value of training or employee development through a staged evaluation process, only considering ROI ex-post, that's how we should be thinking about AI. You are not going to know the ROI until after the fact.
Drop the Templates. Look for Cognitive Stress.
My core argument for companies now is to throw out ROI templates and look for cognitive stress points at points of process criticality and complexity, and just start there without strategy or ROI planning. Especially because before even worrying about ROI, one needs to assess culture through introspection — assessing if innovation and talent engagement is fertile soil for AI capability to take seed. It's not about what AI can do for you. It's about what AI demands from you.
Another anecdote helps round out the point. Years ago, our leadership took a bold step: increasing post-grad interns to 20% of the workforce. Practically overnight, the average age of the organisation was halved, and Gen Z became a sizeable majority. It wasn't easy. We had to learn how to train non-experts and bridge generational divides. Eight years on, that investment paid off. We didn't just build a deep talent pool — we changed our culture and accelerated our pace.
Investing in AI is investing in your workforce's cognitive capabilities, necessitating a broader, more nuanced approach that is closer to HR than it is to Finance. That's how you eventually improve customer value and shareholder returns.
This article is adapted from "You Don't Need an AI Strategy! Part 2: ROI" and subsequent writing by Paul Devalier.