- Accuracy is an input; the only thing that gets a budget approved is a number on the P&L.
- Value = (volume × value per event × improvement) − total cost to build and run.
- Most pitches inflate the improvement and ignore the run cost.
- Lead with financial impact and conservative assumptions; let accuracy be the supporting footnote.
Every AI pitch leads with accuracy. "95% precision!" The CFO does not care — and they are right not to. Accuracy is an input, not an outcome. The only thing that gets a budget approved is a credible number on the profit-and-loss statement. After fifteen years building the business cases behind AI programmes, I have found that one simple formula separates the projects that get funded from the ones that get politely declined.
The formula
It is deliberately not complicated, which is exactly why it works. For any AI initiative:
Value = (volume × cost-or-revenue per event × improvement) − cost to build and run.
Every term is something a finance team can interrogate. The volume of events the AI touches. The cost or revenue tied to each event. The realistic improvement the AI delivers. And the true, fully-loaded cost to build, integrate, govern and run it. The discipline is in being honest about each one.
Where the numbers usually go wrong
Two terms are almost always wrong in an unfunded pitch, and they are wrong in the same direction:
- The improvement is inflated. A model that scores 95% in a test rarely delivers 95% improvement in production, once you account for the cases it cannot handle and the ones that still need a human.
- The run cost is ignored. Integration, monitoring, retraining, governance and support are real recurring costs. A business case that counts only the build cost is not a business case.
Be conservative on improvement and complete on cost, and a genuinely good project still clears the bar comfortably — while a weak one is exposed before you have spent the money.
Why accuracy is the wrong lead
A five-point accuracy gain on a low-volume, low-value process is worth almost nothing. A modest improvement on high-volume, high-value work can be transformational. The same model can be a terrible investment or a brilliant one depending entirely on what it is pointed at — which is why the volume and value-per-event terms matter far more than the accuracy headline. Lead with the financial impact; let accuracy be the footnote that supports it.
The same model can be a terrible investment or a brilliant one depending entirely on what you point it at.
What CFOs actually approve
- A clear line from the AI output to a specific cost or revenue line in the accounts
- Conservative, defensible assumptions that survive scrutiny
- A payback period they can believe, with the run cost included
Start from the number
The teams that get AI funded do not start with the model and look for a use; they start with a costly, high-volume problem and ask what a realistic improvement is worth. That single reversal — number first, model second — is the discipline behind every engagement I run, from a short readiness assessment to a full build. Get the formula honest, and the budget conversation gets a great deal easier.