How to Ask a Skeptical Board for an AI Budget Without Promising Magic
A practical way to ask a board for AI budget: tie it to a known cost, release money in stages, agree measures up front and name a clear exit.
A skeptical board is doing its job. Most directors have sat through at least one technology pitch that promised transformation and delivered little more than a new line item. When you ask for an AI budget, they will listen for the same pitch again. You get the money by making a request that sounds nothing like it.
Ask for a decision, not a belief
Weak AI pitches ask the board to believe something: that AI will reshape the industry, that competitors are ahead, that waiting is dangerous. A board can't vote on a belief, and it shouldn't try. Ask it to approve a bounded decision instead. "We want €120,000 over six months to test whether automated first-pass review cuts contract turnaround from nine days to four. If it doesn't, we stop."
That sentence has a cost, a timeframe, a result you can measure and a way out. A board can approve it without anyone in the room having to become an enthusiast.
Start from a cost the board already knows
Pick a problem that already appears in the board pack. It might be slow quote turnaround, a support backlog, a month-end close that takes twelve days, or a compliance review that ties up senior people. The board already knows what that problem costs because it has been discussing it for months. That gives you a baseline nobody disputes and a result they will recognise when it arrives.
Don't lead with the technology. "We want to deploy a language model" invites questions about the model. "We want to halve the time our engineers spend writing incident reports" invites questions about incident reports, which is where you want the discussion to be.
Put numbers on the downside too
Directors trust a proposal more when it names its own risks, so say what could go wrong and how you will handle it. The output may be right only 80% of the time, so a person reviews every item before it leaves the building. Staff may not use the tool, so adoption is one of the measures. The data may be messier than you expect, so the first weeks go on checking it before anything gets built.
Show the full cost. The software licence is usually the smallest part of it. Most of the money goes on connecting the tool to existing systems, preparing the data, staff time for review and training, and ongoing monitoring. If your budget shows only the software price, someone will notice, and then the board will doubt everything else you told them.
Structure the money in stages
Boards don't like large, open-ended commitments to uncertain outcomes, so don't ask for one. Split the request into stages, and release each stage's money only when the previous stage hits a stated threshold.
- Discovery (4–6 weeks): confirm the data exists, measure how the process performs today, and define success in numbers.
- Pilot (8–12 weeks): run the system on real work with a small team, alongside the existing process, and compare the two.
- Rollout: extend it to the wider operation only if the pilot met its targets, with a budget based on what the pilot actually cost.
This breaks one big bet into small steps you can reverse. It also leaves the board in control of the spending at every stage, which is often what a skeptical director actually wants.
Define what "working" means before you start
Agree the success measures in the meeting, not after the pilot. Candidates include hours saved per week, error rate compared with the current process, turnaround time and customer complaints. Choose two or three, write them down and set the threshold for continuing. That way nobody can move the goalposts once the results come in, and that includes you.
Include a stop condition, such as: "If accuracy is below 90% after eight weeks, we stop and report what we learned." A clear way to fail makes approval easier because it shows you are testing an idea, not defending one.
Drop the words that set off alarms
Some words reliably lose the room: "revolutionise", "transform", "game-changer", and any claim that AI will replace a whole department. Skip the headline figures from vendor reports too, because directors know who paid for them. Use your own numbers, however modest. A realistic 15% gain on a process the board cares about persuades more than a borrowed promise of 40%.
Name an owner. Put one executive in charge of the outcome and have the pilot team report to them. That answers the question every board asks, even if nobody says it out loud: who do we call when this goes wrong?
Report back whether it worked or not
Your first AI budget is really a bid for the second one. At each stage, report against the measures you promised, including the ones that disappointed. If a board sees a pilot stopped cleanly with a clear account of why, it will fund your next proposal faster than if it watched a project drift.
If you want help with the groundwork (measuring the baseline, costing the work, designing the pilot), that is the heavy lifting Hercules A takes on for clients. Either way, the approach is the same: ask for a small, measurable decision, and let the results make the case for the bigger one.