How to Build an AI Business Case That Survives Finance Scrutiny

ROI & Cost Optimisation

5 January 2026 | By Ashley Marshall

You know AI can help your business. So does your CFO - in principle. The gap is the number on the slide: what will it cost, what will it return, and how confident are you in those figures? Building an AI business case that survives finance scrutiny and board review requires a different approach to most technology investments - because AI's value often comes from decisions made better or problems avoided, not just tasks completed faster.

Start With the Business Problem, Not the Technology

The most common error in AI business cases is leading with technology capability: "large language models can automate document processing, which we should take advantage of." This framing puts technology in search of a problem and invites the obvious question: what specific problem are we solving?

A stronger structure starts with the problem: "Our contract review process takes an average of 4.2 days, delaying deal closure and costing us approximately X deals per quarter in lost revenue. AI-assisted review could reduce this to under 24 hours." The technology is the solution, not the premise.

This matters beyond framing. When you start with a specific business problem, you can calculate the value of solving it with much more precision - and you can define success metrics that actually connect to business outcomes.

The Four Value Drivers to Quantify

AI investment typically creates value through four mechanisms. A complete business case should address each, even if some are harder to quantify precisely.

Cost Reduction and Efficiency

This is the most straightforward to calculate. Identify the volume of the process you are automating or accelerating, the current cost per unit (staff time at fully-loaded cost plus any third-party costs), and the expected cost after AI deployment. The difference is the efficiency gain.

Be honest about partial automation. AI rarely eliminates a process entirely - it typically reduces the human effort required by a significant proportion. A document review task that takes 4 hours might be reduced to 45 minutes with AI assistance. Calculate the cost saving based on realistic time reduction, not theoretical full automation.

Revenue Enablement

Some AI investments enable revenue that would otherwise not occur or enable it faster. Faster contract review accelerates deal closure. Better customer health scoring reduces churn. Personalisation at scale improves conversion rates. These revenue effects are real but require careful attribution - not all revenue improvement in the same period as an AI deployment was caused by that deployment.

Use conservative assumptions and base them on comparable evidence where available. If similar organisations have deployed comparable tools and measured revenue effects, that data is much more credible than a theoretical model.

Risk Reduction

AI investments often reduce risk in ways that do not show up directly in revenue or cost lines: better fraud detection, more consistent compliance, improved data quality, reduced key-person dependency. Risk reduction is the hardest value driver to quantify but often the most strategically significant.

A useful approach is expected value modelling: estimate the probability and cost of the risk event occurring without the AI investment, versus with it. Even a rough estimate of risk reduction can be meaningful in a business case if the underlying risk is significant.

Quality and Capability Improvement

AI sometimes enables things that were not possible before, not just the same things done more efficiently. Better analysis, faster insight, more personalised customer experience, coverage of processes that were too expensive to scale manually. These improvements may not have a direct revenue or cost line but contribute to competitive position and customer satisfaction.

For this value driver, qualitative evidence and customer/user testimonials are more useful than speculative financial modelling.

How to Structure the Financial Model

Finance teams evaluating AI investment will expect to see costs, benefits, and a timeline modelled over a multi-year horizon - typically three to five years for a significant investment.

Costs to include: Software licensing or API costs, implementation and integration resource, change management and training, ongoing governance and maintenance, any required data infrastructure investment. Many AI business cases underestimate implementation costs, which damages credibility when actuals emerge.

Benefits to model: Use three scenarios - conservative, central, and optimistic - with explicit assumptions behind each. Identify which assumptions have the most material effect on the outcome (sensitivity analysis) and explain why your central case is defensible.

Timing: Be realistic about the ramp period. AI deployments rarely deliver full benefit on day one. A typical deployment delivers 20 to 30 per cent of anticipated benefit in the first six months, reaching full run-rate benefit after 12 to 18 months. Modelling instant full-value delivery destroys credibility.

The Pilot as Business Case Validation

The most credible business case is one where the assumptions have been validated by a real pilot. Rather than asking for full investment upfront, proposing a funded pilot with defined success criteria and a clear decision gate is often more effective - and more intellectually honest.

A well-designed pilot should be scoped to test the key assumptions in your business case. If your case rests on a 65 per cent time saving in a specific process, run the pilot on that process with real work and measure actual time saving. Three months of pilot data is worth more than three years of modelled projections.

What Finance and Boards Look For

Finance teams evaluating AI proposals are increasingly experienced with AI hype and appropriately sceptical of optimistic projections. The qualities that build confidence:

Specific baselines. "Our current cost per unit is X, based on X staff at X salary processing X volume per year" is much more credible than "we estimate significant cost in this area."

Conservative assumptions with upside, not optimistic assumptions with caveats. Build your central case on assumptions you are confident you can achieve. Then model the upside separately if conditions are favourable.

Named comparable deployments. "Organisation X deployed a comparable tool and reported Y saving" is stronger than theoretical analysis. Industry benchmarks and vendor case studies provide useful reference points even when they cannot be taken at face value.

Clear success metrics. Define before you start how you will measure whether the investment delivered. Vague claims about transformation are much weaker than "we will measure this through metric X, with a target of Y within Z months."

Frequently Asked Questions

What is a typical payback period for AI investment?

For well-scoped efficiency-focused AI deployments, payback periods of 12 to 24 months are common. Strategic AI investments with significant revenue or risk reduction components may have longer payback periods but higher overall returns. Business cases that claim sub-12-month payback should be scrutinised carefully for completeness of cost modelling.

How do you handle uncertainty in AI benefit projections?

Acknowledge uncertainty explicitly rather than presenting a single point estimate. Three-scenario modelling (conservative, central, optimistic) with clear assumptions for each is the standard approach. Sensitivity analysis identifying which assumptions most affect the outcome helps focus scrutiny appropriately and demonstrates rigour. A funded pilot to validate key assumptions before full commitment is the strongest risk management approach.

Should an AI business case include productivity gains that don't reduce headcount?

Yes, but they need careful framing. If AI frees 30 per cent of a team's time and that time is reallocated to higher-value work, the benefit is real - but it is a quality improvement and capability expansion, not a cost saving. Frame these gains as what they actually are: improved throughput, ability to take on more work without additional resource, or higher-quality outputs. Claiming headcount savings that will not materialise undermines credibility.