We see a lot of AI investment proposals go to SMB boards, and a depressing number of them fail to get approved - or get approved at half the scale needed to actually prove anything. The reasons are remarkably consistent: the proposal is too vague about outcomes, too vague about risk, and too optimistic about timeline. Boards are not anti-AI. They're sensibly cautious about anything that smells like a hype investment with no clear test for whether it worked.
Here's how to write a proposal that gets a confident yes.
Lead with the business problem, not the technology
The single most common opening of a failed AI proposal is 'AI is transforming how businesses operate' followed by a paragraph of generalities. Boards have read that paragraph 30 times this year. It tells them nothing about your business.
Lead instead with a specific operational problem you already know they're worried about. 'Our sales team currently spends roughly 12 hours per person per week on proposal drafting and meeting follow-ups - time that could be spent on active selling.' Or: 'Month-end close takes us nine working days and is the constraint on getting timely commercial information to leadership.' That gets the board's attention because it's a problem they already recognise.
AI then enters the conversation as a proposed solution to that problem, not as a topic in its own right.
Be specific about what you'll buy and what it'll do
Vague proposals get vague responses. Be concrete. '20 Microsoft 365 Copilot licences at £24 per user per month, deployed to the sales and marketing teams, with a 30-day structured pilot targeting proposal turnaround time and meeting follow-up volume.' Total cost: roughly £5,800 over six months including a one-off training engagement. Expected outcome: time savings of 3-5 hours per user per week, equivalent to roughly 1.5 FTE of reclaimed capacity, plus a measurable reduction in proposal turnaround.
Numbers like those let the board test the proposal. Numbers like 'unlock significant productivity gains' don't.
Address risk before they ask
Boards know AI carries risk. If you don't mention it, they conclude you haven't thought about it. If you mention it briefly and pragmatically, you build trust and shorten the meeting.
Cover three risks explicitly. Data security: 'We've checked our SharePoint permissions and the sensitivity labels on our highest-risk folders before going live. Copilot inherits our existing Microsoft 365 security model.' Quality and accuracy: 'Our policy requires human verification of any AI-generated content that goes to a customer or regulator. We've trained the team on this.' Adoption risk: 'We've structured this as a 30-day pilot with explicit go/extend/stop criteria, so if it doesn't deliver we'll know quickly and can withdraw without sunk-cost commitment.'
Each risk paragraph is two sentences. The board doesn't need a risk register. They need to see you've thought about it.
Define success on day zero
Write down, in the proposal, exactly what you'll bring back to the board at the end of the pilot and at six months. 'In 30 days we'll report usage rates by team, self-reported time savings, three named use cases that have stuck, and a recommendation to roll out wider, extend, or stop. In six months we'll report measured changes in proposal turnaround and meeting follow-up volume against today's baseline.'
This does two things. It tells the board what evidence they should expect to see, which makes the approval feel safer. And it forces you to actually run the measurement, rather than letting it drift.
Show the alternative
Boards approve investments by comparison, not in isolation. Make the comparison easy. The relevant alternative is almost never 'do nothing'. It's either 'spend this on another priority' or 'do nothing about this productivity problem and watch competitors do it'.
If the proposal is £5,800 over six months and the comparable alternative is a part-time contractor who would deliver less reclaimed capacity at greater cost, say that. If the alternative is doing nothing and watching one of your three biggest competitors invest in Copilot publicly (it's worth checking - they probably are), say that too. The case becomes much harder to defer.
Get the right people in the room beforehand
Don't surprise the board. Walk the proposal past the FD privately first - their objections are usually about ROI math and risk language, both of which are easier to fix in a one-to-one than in a meeting. Walk it past the chair if you have one. Walk it past the most sceptical board member specifically. By the time the proposal hits the actual board meeting, the only question should be 'are we agreed?' rather than 'what is this and why'.
This is not politicking. It's the basic discipline of getting good decisions made quickly.
Keep it short
The proposal should fit on two pages. Problem statement, proposed investment, expected outcome with numbers, three risks and how they're managed, success criteria for the pilot, recommendation. Anything longer signals that you haven't done the thinking to compress it. Boards prefer crisp papers from people who've done the work to papers that demonstrate effort.
What to do if they say no
Sometimes the answer is no. That's information. Ask specifically what would have changed the answer - smaller pilot, different team, more measurement, longer lead time, different vendor. Often the no is to the current shape of the proposal, not to the underlying idea. A revised proposal three months later, with the original objections explicitly addressed, very often gets approved.
The honest summary
AI proposals to SMB boards succeed when they lead with a recognisable business problem, propose a specific and bounded investment, address risk pragmatically, define what success looks like in advance, and compare honestly against the realistic alternative. They fail when they lead with technology trends, use vague numbers, ignore risk, and ask for open-ended approval. Boards aren't suspicious of AI. They're suspicious of fuzzy thinking. Bring them clarity and they'll usually back you.