Why the standard ROI model doesn't fit
A traditional technology investment case compares a known cost against a reasonably predictable productivity gain, over a time horizon where both sides of the equation are stable enough to model. AI investment breaks that model on both sides: the cost is variable and usage-dependent, and the return often compounds, a workflow that gets modestly faster today can become categorically different in eighteen months as the underlying capability improves.
Evaluating that kind of investment with a static ROI spreadsheet built for a software license misprices it in both directions, sometimes overstating the near-term case and almost always understating the longer one.
Treat it as a portfolio, not a single bet
The more useful frame is a portfolio with a defined risk-return profile: a small number of high-conviction, well-scoped bets where the decision being improved and the P&L impact are both named up front; a larger number of smaller, faster experiments with strict time and budget boxes; and a deliberate limit on how much total capital and attention goes to the second category before it has to earn its way into the first.
This does two things a single-bet framework can't. It gives the board a coherent way to evaluate the whole program rather than approving initiatives one demo at a time, and it gives the organization permission to run cheap experiments without every one of them needing to justify itself as a standalone business case.
The AI investments that actually move the P&L are rarely the ones that make for a good demo. Weight the boring case more than instinct suggests.
Tahmid Islam, Chief Financial Officer
Ask for the boring case
The AI investments that actually move the P&L are rarely the ones that make for a good all-hands demo. A model that shaves two days off a reconciliation cycle is a boring case with a clean, defensible number behind it. A generative system that produces impressive creative output is a compelling demo with a much harder number to attach to it.
We tell CFOs to weight the boring case more heavily than instinct suggests. The demo that gets applause in the room is not a reliable predictor of which investment will still be paying back in two years.
Build the exit before you build the case
Every AI investment we help structure gets a kill criterion before it gets approved: a specific, measurable threshold that, if unmet by a specific date, triggers a wind-down rather than a renewal conversation. This is standard discipline for any other capital allocation decision, and AI investment gets a pass on it far too often because the technology feels too important to be seen killing.
That discipline is exactly what makes it possible to keep funding the program at all. A portfolio with a real kill criterion earns the credibility to keep making new bets. One without it eventually accumulates enough underperforming initiatives that the whole program becomes a target.
