Beyond ROI: Reframing IRR for AI Investments
The Deliberate AI Leader — A Series for Executives Who Want to Get This Right – Part 14
IRR — internal rate of return — is a familiar tool in any finance team’s kit. Point it at a set of cash flows, and it tells you the annualized return a project is expected to generate. The problem with applying it to AI initiatives is right there in the definition: it assumes you already know the cash flows.
Traditional IRR assumes predictable cash flows. AI projects rarely begin that way. That mismatch is exactly why so many leadership teams give up on quantifying AI return altogether and fall back on enthusiasm instead, and it’s the wrong response to a real problem.
What to Estimate Instead of Waiting for Certainty
Rather than waiting for perfect data, we ask executives to build an initial estimate across a working set of variables:
- Initial investment
- Annual labor savings
- Revenue uplift
- Error reduction
- Working capital improvements
- Speed-to-decision improvements
- Infrastructure costs
- Ongoing maintenance
None of these numbers need to be precise on day one. They need to be defensible — a reasonable, documented estimate that the team is willing to stand behind and revisit.
A Worked Example
Take a hypothetical AI-assisted intake process in an operations team. The initial investment might include software licensing and implementation hours. Annual labor savings could be estimated from the number of hours currently spent on manual intake, multiplied by a conservative automation rate — say, 40% rather than the vendor-promised 80%, to leave room for reality. Error reduction can be estimated from the cost of rework tied to current mistakes, discounted heavily since the new system’s error rate is still unknown. Infrastructure and maintenance costs get estimated from comparable existing systems.
None of these figures are exact. Together, they produce a rough IRR range — say, 15% to 35% — that’s honest about its own uncertainty but still usable for comparison against other initiatives in the portfolio, which is the entire point.
Precision Isn’t the Goal. Discipline Is.
The point of putting a number on an AI initiative isn’t to predict the future with accuracy. It’s to force the same discipline onto every AI project that capital budgeting already forces onto every other kind of investment: write down your assumptions, compare them against alternatives, and be willing to revisit them as reality provides better information.
A project with a rough but honest IRR estimate is more useful to a leadership team than a project with no estimate at all — even if the rough number turns out to be wrong. At least it can be compared, tracked, and corrected
A Caution Worth Naming
There’s a failure mode on the other end of this spectrum too: AI-washing the math. Inflating projected labor savings, ignoring maintenance costs, or assuming 100% adoption on day one produces an IRR estimate that looks impressive and means nothing. The estimate is only useful if the team is willing to be conservative and revisit it honestly as real data comes in. A leadership team that only ever sees rosy projections should ask harder questions, not fewer.
Refining the Number as the Project Matures
Initial estimates should tighten over time. A pilot phase generates real usage data — actual labor hours saved, actual error rates, actual adoption levels — that should replace assumptions as they become available. Treat the first IRR estimate as a hypothesis to test, not a number to defend at all costs. Revisiting and correcting an estimate isn’t a failure of the original analysis; it’s what disciplined capital allocation looks like in practice.
This closes our three-part series on AI capital discipline. Together, these three articles make the case that AI decisions deserve the same rigor as any other capital investment — no more, and no less.
Want help building an initial IRR estimate for your organization’s AI initiatives? Book a Strategy Call and we’ll work through it together.
About WHIM Innovation
WHIM Innovation helps organizations harness the practical power of AI, automation, and custom software to work smarter and scale faster. We combine deep technical expertise with real-world business insight to build tools that simplify operations, enhance decision-making, and unlock new capacity across teams. From AI strategy and workflow design to custom monday.com apps and fully integrated solutions, we partner closely with clients to create systems that are efficient, intuitive, and built for long-term success.