Whitepaper · roi · 20 min read
The AI ROI Framework for Finance Leaders
A defensible model for measuring the return on enterprise AI - beyond the productivity slide and the vendor case study.
A finance-friendly framework for valuing enterprise AI: counted vs. claimed value, cost of risk, cost of platform, time-to-value, and a worked example for support, sales and operations workloads.
November 22, 2025 · For CFOs - Finance Business Partners - Programme Sponsors
Why most AI business cases fall apart
Most AI ROI decks claim productivity, throughput or deflection without an tracked baseline. Finance correctly asks: where is the cash? This whitepaper offers a model that satisfies both the AI sponsor and the CFO.
Counted vs. claimed value
Counted value is value that shows up in a P&L line: avoided headcount, faster collections, lower contact-rate, higher conversion, reduced spend. Claimed value is everything else. Both can be real; only counted value pays for the platform.
Total cost of platform
Five components: model spend, infrastructure, platform fees, people (engineering + governance), and risk reserve. Underestimating people is the most common mistake.
Time-to-value buckets
Fast (< 90 days): support deflection, marketing copy ops, knowledge search. Medium (3-9 months): sales acceleration, claims triage, onboarding automation. Slow (9-18 months): industry-specific decisioning. Match the workload to the bucket; do not over-promise.
Cost of risk
Add a risk reserve: probability × impact for the failure modes documented in the security whitepaper. A platform with strong guardrails, audit and rollback shrinks this reserve materially.
Worked example: support workload
Baseline 250k tickets/year at $9 fully loaded handle time. AI deflects 28% in year one with 15% reduction in handle time on the rest. Counted savings ≈ $1.4M; platform cost ≈ $480k; net ≈ $920k. Reserve $150k. Net counted value ≈ $770k.
Worked example: sales workload
Baseline 30 reps with 14% conversion at $48k ASP. Agent assist lifts conversion 1.6 pp and reduces ramp time 30 days. Counted lift ≈ $2.1M; platform cost ≈ $360k; net ≈ $1.7M. Reserve $200k. Net counted value ≈ $1.5M.
Worked example: operations workload
Baseline 18 FTE on document-heavy ops. Workflow agents reduce manual hours 40% with HITL on 12%. Counted savings ≈ $1.1M; platform cost ≈ $300k; net ≈ $800k. Reserve $100k. Net counted value ≈ $700k.
Designing a pilot finance can underwrite scale-up on
Instrument the baseline before you ship. Choose a workload with clean metrics. Set the bar at the floor of counted value. Make the rollback button obvious. Report weekly. The next conversation is about doubling the budget, not defending the first one.
More papers on the whitepapers index.