How to build an AI business case your board will trust
By Keith Hamilton · · 5 min read
Boards and owners are right to be wary of AI proposals that promise large savings from a single, confident number. An AI business case your board will trust starts from what is happening today, shows a realistic range of outcomes and is honest about costs and risks. This guide walks through how to build one, with a simple worked example you can adapt.
Why do so many AI business cases fail to convince?
Most weak business cases share the same problems. They start with the tool rather than the problem. They quote one headline saving with no explanation of where it came from. They leave out running costs, staff time and the effort of changing how people work. And they say nothing about what happens to customer data.
A board that has seen a few of these will rightly push back. The answer is not a longer document. It is a clearer one, built on evidence.
Start with a baseline
Before you estimate any saving, measure how the work is done now. For the task you want to improve, record:
- how many hours a week it takes, and who does it
- how often it happens, such as orders processed or queries answered
- current quality measures, such as error rates, response times or complaints
- what it costs, based on the real cost of the people doing it
A baseline turns opinion into fact. It also gives you something to measure against later, which is what makes the board's second conversation easier than the first.
To work out the cost of staff time, take the salary, add employer on-costs of about 20%, and divide by about 1,650 paid working hours a year. That gives a realistic hourly cost.
Show a range, not a single number
Automation rarely removes all of a task. Typically it removes somewhere between a third and two thirds of a repetitive piece of work, with people still handling exceptions, checks and judgement calls.
So show three cases:
- Low: about a third of the task's time saved
- Central: about half
- High: about two thirds
A worked example
Here is a purely hypothetical example with round numbers. Suppose one administrator on a salary of £30,000 spends 10 hours a week keying in order details from emails.
- Salary plus 20% on-costs is £36,000.
- Divided by 1,650 hours, that is about £21.82 an hour.
- Ten hours a week over 52 weeks is 520 hours, costing about £11,346 a year.
The potential saving is then about £3,782 in the low case, £5,673 in the central case and £7,564 in the high case. These are not promises. They are a range to test.
Be clear about what the saving means. Unless the time is redeployed to valuable work, or reduces the need for overtime or new hires, it is capacity rather than cash.
Include implementation and running costs
A trustworthy business case counts everything, not just the licence. Include:
- Implementation costs: set-up, configuration, connecting systems, testing and any outside help
- Running costs: subscriptions, usage charges, support and ongoing maintenance
- Internal time: staff hours spent on training, testing and checking outputs, especially in the early weeks
- Governance: policy, data protection checks and supplier due diligence
Continuing the hypothetical example, suppose implementation costs £4,000 and running costs are £1,000 a year. The net annual benefit is then about £2,782 in the low case, £4,673 in the central case and £6,564 in the high case.
Work out the payback period
Payback is simply how long it takes for the net benefit to cover the upfront cost. Boards like it because it is easy to understand.
In the example, payback is roughly 17 months in the low case, 10 months in the central case and 7 months in the high case. Showing all three makes the decision honest. If the low case is still acceptable, the board can approve with confidence. If only the high case works, that tells you something important too.
Be honest about risks and data protection
Every AI business case should include a short risk section. Cover:
- what personal data the AI will touch, your lawful basis under UK GDPR, and whether a data protection impact assessment is needed
- where the supplier processes data, and whether it leaves the UK
- what happens when the AI gets something wrong, and who checks its outputs
- whether any decisions significantly affecting people will need human review
- the impact on staff, and how you will bring them with you
- what you would do if the supplier changed its terms or prices
Naming the risks, with a plan for each, usually increases a board's confidence rather than reducing it.
Start small, measure and report back
The most persuasive business case asks for a small first step. Propose a limited pilot on one task, with a clear time frame and success measures agreed in advance.
When the pilot ends, measure the same things you recorded in the baseline. Report the actual result against the low, central and high cases, including anything that went less well than expected. That report becomes the evidence for the next decision, and it builds a track record your board can rely on.
Frequently asked questions
What should an AI business case include?
At a minimum, a baseline of the current work, a low, central and high estimate of benefit, all implementation and running costs, a payback period and a section on risks and data protection. It should also propose how results will be measured and reported. Keep it short enough to discuss in one board meeting.
How accurate do the savings estimates need to be?
They need to be reasonable and transparent rather than precise. Showing a range, explaining the assumptions and using real baseline data is more credible than a single exact figure. The pilot is where you replace estimates with evidence.
What if the pilot does not deliver the expected savings?
That is useful information, gained cheaply. Report it honestly against the baseline, look at why, and decide whether to adjust, try a different task or stop. A board will trust future proposals more if it sees results reported openly.
Where Forwardcycle fits
Building business cases and reporting results to boards is a core part of our commercial leadership service. To see which tasks in your business might be worth looking at first, try the free AI Readiness Assessment or arrange a free call.