Too many AI business cases get shelved not because the technology is wrong, but because the case doesn't speak the language of the people who control the budget. The CFO asks what it does to the operating margin, and there's no good answer. That's a solvable problem.
Why most AI business cases fail
The typical AI business case makes three mistakes.
Mistake 1: leading with technology
"We'll implement a large language model to process customer enquiries." This tells the board what you're building but not why it matters. Get clear on the outcomes you will be delivering on: revenue, cost, risk, speed. The technology is a means, not an end.
Mistake 2: vague benefits
"This will improve customer experience." That's a hope, not a business case. How much improvement? Measured how? Over what timeframe? If you can't quantify the benefit, you can't justify the investment.
Mistake 3: ignoring the change cost
The AI model itself might cost £200K. But what about the change management, the process redesign, the training, the integration with existing systems? Most business cases dramatically understate the total cost of ownership because they only price the technology and solutions. Price the whole thing. Every time.
A structure that works
Here's a four-section structure that helps to get board approval on the right terms.
Section 1: The business problem
Start here, not with the technology. What's the problem costing the organisation today? Be specific. "Our claims processing team handles 50,000 claims per month with an average cycle time of 14 days. This drives £3.2M in annual operating cost and is the primary driver of customer complaints, which run at about 18% of processed claims."
The board now understands why this matters. You haven't mentioned AI once.
Section 2: The proposed solution
Now introduce the technology, but frame it around the outcome. "An AI-assisted triage system will categorise and pre-process claims, reducing average cycle time to 3 days and redirecting about 60% of simple claims to automated processing."
It's not "we'll deploy an LLM." It's "we'll reduce cycle time and automate simple claims."
Section 3: The financial model
Build a three-scenario model:
Conservative: 40% adoption in year one, 60% by year two. Net benefit: £1.1M over three years.
Expected: 60% adoption in year one, 85% by year two. Net benefit: £2.8M over three years.
Optimistic: 80% adoption in year one, 95% by year two. Net benefit: £4.2M over three years.
Include all costs: technology, integration, change management, training, ongoing operations. Show the payback period for each scenario. The CFO will focus on the conservative case, so make sure it still works.
Section 4: The risk register
Boards don't expect zero risk. They expect you to have thought about it. Cover three areas:
Technical risk. What happens if the model doesn't perform as expected? What's the fallback?
Adoption risk. What if staff resistance slows rollout? How will you manage the change?
Regulatory risk. Are there compliance implications? Have you engaged legal and compliance early?
For each risk, show the mitigation and the residual risk.
The golden rule
Every number in your business case should trace back to a P&L line item. If it doesn't appear on the income statement or balance sheet, it's not a commercial business benefit. It's a nice-to-have.
"Improved employee satisfaction" alone won't get you funded. "12% reduction in staff turnover in the claims team, saving £480K in annual recruitment and training costs" will.
What this means for you
If you're building an AI business case right now, test it against these four questions:
Does it start with the business problem? If your first slide mentions a technology, rewrite it.
Are the benefits quantified and time-bound? "Improved efficiency" isn't good enough. Put a number and a date on it.
Does the cost model include everything? Technology, people, process change, ongoing operations.
Have you built three scenarios? Conservative, expected, and optimistic. The board will stress-test you on the conservative case.
Get these right and you won't just get approval. You'll get a board that actively sponsors your AI programme because they understand exactly what it's worth.