Most AI projects fail the CFO test — not because the technology didn’t work, but because nobody measured them in finance terms. The internal champion built a demo, the COO signed off, the bills started arriving, and twelve months later a question landed: "what did we get?" Here’s how to design the scorecard before you build, so the answer in month twelve doesn’t depend on storytelling.
Four buckets the CFO actually cares about
Revenue impact, cost impact, risk reduction, and capability building. Most AI projects only measure one bucket (usually cost) and underdeliver because the other three weren’t designed for. The scorecard treats all four as first-class. Each gets a baseline, a target, and a measurement method agreed at project kickoff.
Revenue impact
Three KPIs that survive scrutiny. Lead response time — does AI-drafted outreach get answered faster? Sales-qualified lead conversion — do AI-prepared briefings raise win rate? Account expansion — does AI-generated cross-sell intel surface real opportunities? Track each against a baseline period before AI was deployed; three months pre / three months post is the minimum honest window.
Cost impact
Cycle time on the workflows you targeted. Manual labour reallocated (not eliminated — the CFO knows people don’t disappear after AI ships, they shift to higher-leverage work). Direct cost per task before and after. Be specific: "ticket resolution went from 12 minutes to 7 minutes average, across 4,200 tickets per month, freeing 7 FTE-hours per day" beats "AI reduced support cost."
Risk reduction
The bucket most projects skip and CFOs care about most after their first audit. Hallucination rate on evaluated queries. Audit-log completeness — every AI decision traceable to inputs. Compliance posture — where AI replaced a control, what replaced it. Risk-reduction wins matter even when revenue and cost numbers are mixed; they convert AI from "experiment" to "infrastructure" in the CFO’s mind.
Capability building
The longest-horizon bucket. How many teams are now AI-literate. How many workflows are automation-ready (mapped, not yet automated). How many production-ready AI components exist that future projects can reuse. This is where the 18-month return shows up, and it’s the bucket that justifies the AI-platform investment when the first-12-month numbers are middling.
The 30 / 60 / 90 cadence
Don’t wait until month twelve to discover the numbers. Day-30 check: are the baselines actually captured and is the measurement pipeline shipping data? Day-60: directional movement visible? Day-90: full scorecard rendered, course-correct decision made. This is the cadence that catches "the demo worked, the production system didn’t" before twelve months of budget evaporate.
IDS AI Solutions builds the scorecard into the AI Audit Sprint deliverable — every engagement starts by selecting which KPIs the AI must move and what the baseline already is. Talk to our team.