Research · AI Strategy & Adoption

Business Case & ROI

An AI business case must survive the question every CFO asks: what will this cost, what will it return and how will we know? Credible cases count the full cost of ownership and measure outcomes, not activity.

The AI Value Ledger

An AI business case is usually written once to win funding and then forgotten. The AI Value Ledger is a CorpExcellence framework that treats the case as a living account instead. It records what the process costs before AI, what the AI really costs to run, how much of the change the AI actually caused, and whether the benefit reached a budget line. Leaders update it each quarter, so the case that approved a project is the same one used to scale it or stop it.

1. Set the baseline before you build

Measure the current process first: cycle time, cost per transaction, error or rework rate, and the customer outcome you care about. Use figures you already report so finance trusts them. Without a baseline, every later benefit is an estimate presented as a result.

2. Count the full cost

Model and token charges are only the visible part. A complete cost line includes integration and data preparation, evaluation and red-teaming, human review of outputs, security and governance work, training, and the support that keeps prompts and connectors working as vendors update their models. Model the cost under at least two usage scenarios, because consumption, not license count, now drives most AI bills.

3. Attribute the change

Compare a group using the AI with a similar group that is not, over the same period. Where a control group is impossible, use a staggered rollout and compare early and late adopters. Agree on the measurement method with finance before the pilot starts, not after. Report adoption separately from outcomes. High usage shows interest, not value.

4. Confirm the value was realized

Saved time counts as a benefit only when it turns into something measurable: fewer contractor hours, volume growth absorbed without new hires, faster revenue or lower loss rates. Before launch, name the budget owner who will show the change in their own numbers. If no owner will commit, treat the benefit as unproven.

5. Re-forecast every quarter

Prices and usage patterns move quickly. Each quarter, update cost and benefit against the plan and apply the stop rules agreed at approval. A case that is never revised is a sales document.

Know the true cost of one successful transaction

AI Inference Cost Control gives you the cost-per-transaction method behind this ledger, with a worksheet, a cost-driver decomposition and an executive AI cost scorecard.

Get the full report →

Where the industry is now

AI cost behavior has changed. The FinOps Foundation notes that per-token prices keep falling while reasoning and agentic workloads use many more tokens per task, so total spend often rises. Pricing models are shifting as well: Salesforce introduced consumption credits charged per agent action in 2025, and Intercom prices its AI agent per outcome. Cost per business outcome is now the most useful unit in a business case, because it can be compared directly with the baseline.

Common pitfalls

• Counting hours saved with no plan to redeploy them.
• Pricing the pilot and assuming production costs grow in a straight line.
• Leaving evaluation and human review out of the cost line.
• Approving a case with no stop rule.