A business case a CFO can audit.

The problem it solves

The market has corrected from AI euphoria toward rigor. Roughly fifteen percent of AI decision-makers report a positive profitability impact in the past year. Fewer than a third can tie AI outputs to concrete business benefit. Finance functions now discount AI business cases on arrival, and they are right to.

A CFO applies one filter above all others: is this derived from our data or your assumptions? A case built on vendor benchmarks fails that filter immediately. The tool cannot force a consultant to have client data, but it can make the absence visible so nobody is surprised in the room.

What I built

A two-part system. A structured discovery instrument with thirty-two questions across eight sections that captures what a real conversation produces, and an estimator that turns those answers into a risk-adjusted range with the math shown.

Four pillars, not one

Built on Forrester's Total Economic Impact framework, which has been the standard for technology business cases for two decades. Benefits, costs, flexibility, and risk are all first-class. Most ROI calculators model benefits and gesture at the rest. Full total cost of ownership runs across six lines including data preparation, which published post-mortems consistently put at thirty to fifty percent of AI project budgets and which informal estimates almost always miss.

Ranges, not point estimates

Every meaningful input is captured as three points and interpreted as a tenth percentile, a mode, and a ninetieth percentile. Those parameterize a PERT-beta distribution, and four thousand Monte Carlo iterations run on every change. The headline is a P10 to P90 range with a P50 base case, risk-adjusted by an explicit register of named risks with probability and impact.

PERT rather than triangular is a deliberate choice. Triangular treats the extremes as equally dense as the mode, which inflates variance and produces tails no reviewer believes.

The J-curve

Productivity drops for the first several months of an AI adoption before it rises. A flat ramp from day one is dishonest, so the timeline models the dip, the climb, and the asymptote, and payback is computed against the realistic curve rather than the optimistic one.

Provenance and validation

Every input is tagged as client data, industry benchmark, or SME estimate, and the output states the mix on its face. Eight automated validation rules run continuously, catching ranges that are suspiciously tight, payback outside credible bounds, double-counted headcount against labor savings, and risk registers that look decorative.

What it gives an organization

• A value case in the register of a capital allocation memo rather than a vendor pitch.
• An honest answer to where these numbers came from, on the cover slide.
• Projected against actual once the work is live, so the next case is calibrated by the last one.
• A portfolio view that aggregates engagements, which is where a firm learns whether it estimates well.