How to build a business case for AI investment in a mid-market company
Most AI business cases fail before they reach the board. Not because the technology is unproven, but because the person building the case framed it wrong. They led with capability - what the AI can do - rather than with commercial impact - what the business will gain, or stop losing.
At the £500m to £1.5bn revenue mark, this mistake is particularly costly. You are large enough that inefficiency compounds at scale. You are not large enough to absorb a misdirected investment or survive a board that loses confidence in your digital agenda. The margin for error is real.
The other trap is waiting. Leaders at this stage frequently defer AI investment until the picture is clearer - until the technology matures, until a competitor moves first, until a consultant produces a roadmap that never gets acted on. The picture will not get clearer by waiting. The cost of inaction accrues quietly.
This article sets out how to build a business case for AI investment that will survive scrutiny from a CFO, land with a board and point directly at value.
Start with a problem worth solving, not a technology worth deploying
The weakest AI business cases begin with the technology. "We want to implement a large language model in our finance function." That framing puts the burden of proof in the wrong place. The board is not being asked to approve a technology - they are being asked to approve a use of capital.
Start with an operational or commercial problem that already has a cost attached to it. That cost might be explicit - labour hours, error rates, customer churn - or it might be implicit, such as decisions being made on stale data, or a sales team that cannot access the right account intelligence at the right moment.
A useful test: can you describe the problem without mentioning AI at all? If you can, you have a real problem. If you cannot, you have a solution looking for a problem.
Consider a distribution business turning over £800m annually. Their procurement team spends roughly 40% of their analyst time extracting and formatting supplier data before they can do any actual analysis. That is a describable, measurable problem. AI is not the starting point - reducing wasted analyst capacity and improving procurement decisions is. AI becomes the mechanism once the problem is established.
Quantify the value in terms the CFO will accept
Once you have identified the problem, you need to translate it into financial terms. This is where most AI advocates - even experienced ones - underdeliver. They produce ranges so wide they become meaningless: "between £500k and £5m in annual benefit." A CFO sees that and hears: "we do not actually know."
Build your value case around three categories:
- Cost reduction: labour reallocation, error remediation, process automation. These are the easiest to model and the easiest to defend.
- Revenue impact: faster decision-making, better customer targeting, reduced churn. Harder to model precisely, but credible if you anchor estimates to specific workflows rather than overall revenue uplifts.
- Risk reduction: regulatory compliance, data quality, scenario planning capability. Hardest to quantify, but relevant to boards managing liability or operating in regulated sectors.
Do not try to claim all three simultaneously in a first business case. Pick the one category where you can make the strongest quantitative argument and build there. A tight, credible £400k cost reduction case will move faster than a sprawling £3m aspiration with weak assumptions.
One practical approach: model the "cost of doing nothing" as its own line item. If your customer service operation is handling 60,000 contacts per year and 35% of those are queries an AI agent could resolve, calculate the fully loaded cost of that volume. Then ask: what is this costing us each year we delay? That reframes inaction as an active choice with a price.
Define what success looks like before you approve anything
A business case is not just an approval document - it is a measurement contract. If you do not define what success looks like before the investment is made, you will not be able to defend the investment afterwards.
This means specifying, in advance: the metric you are moving, the baseline you are measuring from, the timeframe in which you expect to see movement and the person accountable for delivery.
In practice, most organisations skip this step because it feels premature. The instinct is to get the budget approved and then figure out measurement. That is backwards. Investors and boards have become more sceptical of AI investments that cannot be tied to specific, reportable outcomes. If you cannot define your success metrics upfront, you are not ready to invest.
A useful structure is to separate leading indicators from lagging indicators. A leading indicator for an AI-assisted pricing tool might be the percentage of quotes generated using model recommendations. The lagging indicator is margin improvement. You need both: the lagging indicator tells you whether you succeeded; the leading indicator tells you early whether you are on track.
Sequence investments to demonstrate value quickly
Not every AI investment should be a multi-year programme. In fact, at the mid-market level, multi-year programmes are often how AI investments die. They create dependency on sustained organisational will, on technology that may shift significantly over the programme period and on executive sponsors who may not still be in post at go-live.
A better sequencing approach: identify one use case where the data already exists, the problem is well-defined and the impact can be demonstrated within 90 days. Use that to prove the model - commercially and culturally. Then build.
This is not about thinking small. It is about generating the board confidence and organisational muscle memory that makes the larger investments possible. A financial services firm at £1.2bn revenue that starts with an AI-assisted credit memo tool and demonstrates a 30% reduction in analyst preparation time has a fundamentally different conversation at their next board cycle than one that pitched a three-year data transformation programme and has nothing to show for it yet.
Rodan's diagnostic engagements are structured exactly around this logic - identifying the highest-value, fastest-to-prove opportunity within an organisation before committing to broader transformation.
Address the risks the board will raise before they raise them
Any serious board will ask four questions about an AI investment. Anticipate them.
What are the data risks? Boards are increasingly aware of data governance, privacy and model hallucination. Address this directly. Show that you have assessed your data quality, that you understand where the AI will and will not be relied upon for decisions and that there is a human review layer where it matters.
What does this cost to run? Build-and-forget is not a model. AI systems require ongoing monitoring, retraining and governance. Include operational costs in your case, not just implementation costs.
Who owns this? Name the internal owner. AI investments that lack a clear internal champion and accountable operator have a poor delivery record. The board knows this.
What happens if it does not work? A reversibility or off-ramp plan is not a sign of weak conviction - it is a sign of commercial discipline. Define the conditions under which you would pause, pivot or stop.
A business case that surfaces and addresses these questions in advance reads as mature and credible. One that ignores them invites the board to generate their own objections, often in the meeting itself.
Build the case now, or pay for the delay later
The mid-market window for AI differentiation is narrowing. Organisations that have already moved are building data assets, refining models and accumulating the organisational capability that compounds over time. Those advantages are not insurmountable yet. In twelve months, some of them will be.
The business case you build today does not need to be perfect. It needs to be honest, specific and financially grounded. A tight case for a well-defined problem will get funded. A grand vision with soft numbers will not.
If you are not certain which problem to prioritise, or you want external rigour applied to your assumptions before they go to the board, that is exactly the starting point for a Rodan diagnostic engagement. We will identify the highest-value AI opportunity in your business, stress-test the financial assumptions and give you something a CFO will take seriously.
Book a diagnostic at rodan.io.
