How to Build an AI Business Case
A practical framework for building a compelling business case that gets AI investment approved. Covers costs, ROI, risks, and how to present to leadership.
You know AI could help your business. But the people holding the budget need more than enthusiasm. They need numbers, timelines, risk assessments, and a clear explanation of why this investment makes sense compared to all the other things competing for the same budget.
This guide gives you a framework for building an AI business case that survives contact with a finance director. It is the structure I would use myself, and it works at any scale, because the questions a board asks are the same whether the number is four figures or seven.
Step 1: Define the Problem, Not the Solution
The most common mistake in AI business cases is leading with the technology. Executives do not care about large language models or neural networks. They care about business problems.
Start by clearly defining the problem you are trying to solve. Be specific and quantify the impact:
Weak:“We should use AI to improve customer service.”
Strong:“Our customer service team spends 35% of their time answering the same 20 questions. This costs us £180,000 per year in staff time and results in average response times of 4 hours, contributing to a customer satisfaction score of 72%.”
The strong version gives leadership everything they need: a clear problem, a cost, and a measurable impact on the business. Now they are motivated to hear about the solution.
Step 2: Estimate the Full Costs
Be thorough and honest about costs. Nothing kills credibility faster than unexpected expenses after approval. Include everything:
The figures below are round numbers to show the shape of the calculation. They are not quotes, and they are not what I charge. My own prices are published on the services page.
Technology costs
Software subscriptions, API usage fees, infrastructure, development tools
ChatGPT Teams: £20/user/month x 15 users = £3,600/year
Implementation costs
Consultant fees, internal staff time, system integration, data preparation
Consultant: £5,000 for setup and integration; Internal team: 40 hours at £45/hr = £1,800
Training costs
Workshop delivery, materials, staff time away from normal duties
Half-day workshop: £1,500; Staff time: 15 people x 4 hours x £30/hr = £1,800
Ongoing costs
Subscriptions, maintenance, updates, additional training, support
Annual: software £3,600 + quarterly reviews £2,000 = £5,600
Step 3: Project the Return on Investment
ROI projections need to be realistic, not optimistic. Use conservative estimates and show your working. Decision-makers respect honesty more than impressive-sounding numbers they cannot verify.
Calculate ROI across three categories:
Direct cost savings
Time saved multiplied by hourly cost. If AI saves your team 20 hours per week at an average loaded cost of £35 per hour, that is £36,400 per year in reclaimed capacity.
Revenue impact
Faster response times, higher customer satisfaction, and improved lead conversion can drive revenue growth. Be conservative: if you project a 5% improvement in conversion, show the data that supports it.
Avoided costs
Hiring that you will not need to make, errors that will not need correcting, penalties that will not need paying. These are legitimate but should be clearly labelled as projections.
Example ROI calculation
Illustrative figures, invented to show the shape of the arithmetic. Replace every one of them with your own measurements.
Step 4: Assess and Address the Risks
Every business case should include an honest risk assessment. Acknowledging risks does not weaken your case; it strengthens it by showing you have thought things through.
| Risk | Likelihood | Impact | Mitigation |
|---|---|---|---|
| Low team adoption | Medium | High | Phased rollout, champion programme, training |
| Data privacy breach | Low | Very High | Approved tools only, AI policy, data classification |
| AI output errors | Medium | Medium | Mandatory human review, quality checks |
| Cost overrun | Low | Medium | Fixed-price consultant, capped API spend |
| Technology changes | Medium | Low | Vendor-agnostic approach, quarterly reviews |
For a deeper look at AI risks and how to manage them, see the AI risks guide.
Step 5: Define the Timeline
Leaders want to know when they will see results. Create a realistic timeline with clear milestones:
Discovery and planning
Map current processes, define success metrics, select tools
Setup and configuration
Deploy tools, configure integrations, create AI policy
Training and pilot
Train pilot team, run parallel processes, gather feedback
Optimise and expand
Refine based on feedback, begin wider rollout
First review
Measure against baseline, report results, plan next phase
Step 6: Present to Leadership
The best business case in the world fails if it is presented badly. Here is how to structure your presentation:
- 1
Start with the problem
Two minutes maximum. Focus on the business impact, not the technology.
- 2
Show the cost of doing nothing
What happens if you do not act? Lost time, missed opportunities, competitive risk.
- 3
Present the solution briefly
What you plan to do, in plain language. Keep technical details to a minimum.
- 4
Share the numbers
Costs, projected savings, ROI, and payback period. Use the conservative estimates.
- 5
Address risks proactively
Show you have thought about what could go wrong and how you will prevent it.
- 6
Propose a pilot
If full commitment is too big an ask, propose a smaller proof of concept with a clear success criteria.
- 7
End with a clear ask
Exactly what you need: budget amount, team resources, timeline, and decision deadline.
Frequently Asked Questions
What ROI should I expect from an AI project?
There is no typical figure worth quoting, and any you find online are averages across businesses nothing like yours. A simple automation and a custom predictive model have completely different return profiles. Build the number from your own measured costs, discount it heavily, and you will rarely be caught out.
How long does it take to see ROI from AI?
It depends entirely on what you built, and a board will respect that answer more than an invented timeline. A quick win such as email drafting is fast to switch on and fast to read. Anything requiring integration and adoption takes far longer to show a signal, and looks like a failure while it does. Name the review date in the business case so nobody has to guess later.
What if my business case gets rejected?
Ask for specific feedback on what would change their minds. Often the answer is 'show us proof on a smaller scale first.' Propose a cheap pilot that addresses exactly that concern. A small proof of concept with real numbers from your own business persuades a board far more reliably than a large proposal built on assumptions.
Should I hire or outsource AI implementation?
For most SMEs, outsourcing the first project is the cheaper answer, because a permanent senior hire is a recurring annual cost with employer NI, pension and notice attached, and the first project is a one-off. Hire once you know what you are hiring for. Until then, make knowledge transfer to your existing team part of the brief.
How do I measure intangible benefits of AI?
Employee satisfaction, decision quality, and competitive positioning are harder to quantify but still important. Use proxy metrics: employee NPS scores before and after, time-to-decision for key business choices, customer satisfaction scores, and qualitative feedback from team leads.
Related Guides
How to Use AI in Your Business
Complete guide to getting started with AI
AI Risks for Business
Understand and mitigate AI adoption risks
AI Consulting Cost UK
Budget numbers for your business case
AI ROI Guide
Real ROI data to strengthen your proposal
How to Automate With AI
Identify and automate your best opportunities
AI Consulting Services
Get help building and presenting your business case
Need Help Building Your Business Case?
Tell me what you are proposing and what it currently costs you, and I will help you structure the case. The first conversation costs nothing, there is no booking system, and if the numbers do not work I will tell you before you take it to a board.
Get in touch