Measuring AI ROI: a model that boards actually believe

"Productivity uplift" is not a number a CFO will fund a second year of work on. "$2.4M reduction in cost-per-resolution, audited" is. The difference between AI projects that get a second budget and those that don't is whether they connected to a P&L line.

The four-quadrant model

1. Cost reduction — fewer humans needed for the same work. The most defensible quadrant; cleanly measurable.
2. Revenue lift — better conversion, larger basket, lower churn. Harder to attribute, larger upside.
3. Risk reduction — fewer compliance breaches, fewer fraud losses. Hardest to measure, often the highest-value.
4. Capability creation — things you couldn't do before. Don't try to ROI these in year one. Frame them as options.

The metric your CFO actually wants

Cost-per-X. Cost-per-resolution. Cost-per-claim. Cost-per-lead-qualified. Pick the unit your business already tracks and report your AI investment as a delta on that unit. Boards understand unit economics. They do not understand "productivity uplift."

How we instrument

From day one of any engagement, we put a line of code that logs the AI-handled outcome alongside the human-handled outcome it replaced. Six months in, we have a clean dataset that lets us compute cost-per-X with and without AI. No assumptions, no surveys. Just outcomes.

The trap of "hours saved"

Hours saved is the easiest metric and the worst one. It assumes the saved hours convert to value, which only happens if the people whose time you saved are doing higher-value work with the recovered time. Most aren't. Stop reporting hours saved. Report decisions made or units processed.

What to put on the board slide

• Baseline. Cost-per-X before AI, with sample size and methodology.
• Today. Cost-per-X now, with sample size and methodology.
• Quality. Has the quality of the X held? (CSAT, accept rate, dispute rate.)
• Coverage. What share of X is now AI-handled?
• Trajectory. What does the next 12 months look like?

Framing the un-quantifiable

"Capability creation" gains — things your business literally could not do before — are hard to ROI in the first year. Don't pretend you can. Frame them as options: "this capability lets us enter Market X. We are not modeling that revenue yet. We are quantifying the cost of not having the capability." Boards respond well to honesty.