The AI pilot is back on the executive committee’s agenda. The results look promising, the team has learned a lot, and they’re asking for another quarter to be sure before anything bigger is committed.
I’ve been in an executive committee meeting where an AI pilot in finance made exactly that request. Nobody around the table, me included, wanted to be the one to say stop, so it was noted and we moved on.
I understand why that happens. Another quarter feels like the careful choice. Nobody has to back the pilot at scale or write off what’s been spent, and the meeting gets to the next item on time. But it’s still a spending decision, and nobody actually takes it. The team stays on the pilot, the budget stays committed, the benefits in the business case wait, and whatever those people would otherwise be working on does too.
An AI pilot extension commits the team, the budget and the business case benefits for three more months, whether or not anyone decides to.
Here’s how I’d pitch an extension that could change the decision from "no," and what I’d ask for before funding either.
The short version
Treat another quarter as a new funding request: the cash, the work the same people aren’t doing, and the benefits that arrive later.
Ask what the next quarter will tell you that this one didn’t, and whether that answer could change your decision.
Before the review, ask for a one-page case in numbers, including what the improvement is worth versus what it would cost to run.
The second time a pilot comes back, choose scale, change or stop, with a named decision-maker and a date.
Another quarter is a new investment decision
You don’t need to be told that what’s already been spent shouldn’t decide what happens next. The extension still gets through, because it doesn’t look like a decision. It looks like continuity: same team, same budget, one more quarter.
So I’d price it like any other request for money. Say the pilot costs $200,000 a quarter for the delivery team, licences, and running the model. That excludes the finance people testing it alongside their day jobs.
Judge another quarter of an AI pilot against everything else the same money and the same people could do next quarter.
Approve the first extension and two more after it, and you’ve committed another $600,000. That’s only the cash. The pilot team isn’t working on the next priority, and if go-live moves with each extension, the business case benefits arrive nine months later. Those are three different costs, and only the first appears on the pilot’s budget line.
If the executive committee faces the same question at the third review as it did at the first, that $600,000 bought nine more months of the same question. There may have been useful learning along the way. What matters is whether any of it changed what the committee can decide.
This is about pilots funded to prove a business case. Exploratory work with a capped budget, set up to learn, is a different kind of spend, and we judge it by what it learns.
The question I’d put to any extension is this: “What will the next quarter tell us that this one didn’t, and could that change what we decide?” If nobody can answer it, you already have what you need to decide today.
Why executive committees keep saying yes
Everyone around the table knows this. So why does the extension keep getting approved?
Research published in Harvard Business Review in August helped me name it. Morgan Blangeois and Thomas Roulet spent three years with leadership teams in 11 European IT services firms, through 23 interviews and three workshops in which executive committees debated their own AI strategy. In private, the executives had clear views about what AI required of their firms. In the workshops, they quietly dropped those views. As the authors put it, “Nobody fought AI; nobody decided anything about it either.”
An AI pilot extension is the only option that asks no one on the executive committee to say what they think.
One scenario they gave every team was a multimillion-dollar AI plan with no return yet, and whether to keep paying for it. Individual views shifted, but no team reached a position it could state out loud together, and even with their sector’s productivity figures in front of them, teams asked for more time. The authors call it AI leadership drift.
Their study was of IT services firms, so applying it to other executive committees is my judgement, not their finding. It matches the meeting I described. An extension is the one option that asks nobody to say what they think: scaling means someone backs the pilot in front of their peers, and stopping means someone says the money didn’t work.
Better numbers alone didn’t break the drift. The authors’ answer has two parts: tie every pilot to the bigger decision it’s meant to feed, with a date, and make it one person’s explicit job to keep raising the question everyone else would rather drop.
Before the pilot comes back, I’d ask who in the meeting has that job. If the answer is nobody, the extension is already approved.
The one page I’d want before funding another quarter
If the request arrives as a progress update, with what the team did, what it learned and what it’ll do next, that’s useful to the team. It doesn’t tell the person deciding whether another quarter is worth paying for.
Before the next review, I’d ask the programme director for one-page that answers five questions.
What has the pilot proved, against a baseline finance has signed off? Promising results aren’t enough here. I want the measure before the pilot, the measure now, and how you took both.
Where did the saved time go? If the pilot saves hours, I want to know what happened to them: a close finished earlier, cost avoided, more work absorbed without hiring, lower risk, or nothing, because the time disappeared into the next queue.
What’s the one question still open, and what evidence would settle it by the decision date? If the page lists several, I’d ask which one would change the decision.
What does it cost, and what is it worth? Another quarter all in, including the business people testing it; what it would cost to run in production, because a pilot that’s cheap to test can be expensive to run; and what the improvement is worth against that.
What’s the recommendation, who decides, and what would change it? Scale, change, or stop, with the decision date.
Each question removes one way to avoid the decision. The baseline stops “promising” standing in for a number, the time question stops counting hours as money, and putting the cost next to the worth makes the extension compete with every other use of the same budget. The fifth makes someone commit to a view, with their name and a date, before the meeting starts.
Before funding another quarter of an AI pilot, separate three questions: does the technology work, does the process improve, and is the improvement worth what it costs to run?
If the papers that reach you usually arrive as options rather than recommendations, I wrote about what to do when every decision is sent up to you for review.
Here’s the page filled in for a made-up pilot. Say a finance team has spent a quarter testing an AI agent that compares the ledgers of two group companies each night and flags the entries that don’t match, so mismatches surface before the month-end close. The figures are illustrative.
What it proved: over three closes, the system flagged mismatches on day two of the close instead of day six, against a baseline the group financial controller signed off before the pilot started.
Where the time went: the analysts spend less time finding differences, but the close hasn’t got any shorter. Clearing a mismatch still needs procurement, billing and the other company’s finance team to agree.
The open question: if the agent also sends each mismatch straight to the people who have to agree it, does the close get shorter? The evidence: two more closes, with the close date measured against the same baseline.
Cost and worth: $150,000 for the quarter, plus two analysts spending about a day a week on it. Running it across 40 companies in the group would cost roughly $400,000 a year. Nobody has yet put a value on a shorter close.
The recommendation: change it. Fund one more quarter on the routing question only, and ask the group financial controller to value a shorter close before the January review, where the CFO decides. If the close doesn’t get shorter, or a shorter close isn’t worth $400,000 a year, stop it.
The example separates three things a pilot report can blur: whether the technology works (mismatches found on day two), whether the process improves (the close finishing earlier) and whether the improvement is worth what it costs to run. Only the third tells you whether to fund it.
The second question is the one I’d read first. In the reconciliation work I’ve seen, finding a difference is quick; getting several teams to agree on it, often across business units and time zones, takes time. A pilot can speed up the quick part, look like a success, and leave the close where it was.
That page changes the conversation. Instead of “Do we give it more time?”, the executive committee is deciding whether a shorter close is worth $150,000 and one more quarter, and it already knows what it’ll do if the answer is no.
Three real decisions on the review date
When the pilot comes back with its page, the executive committee has three real decisions. Noting it and moving on isn’t one of them.
I treat a returning pilot the way I treat a risk at a steering committee. Noting a risk the first time is reasonable. The second time it comes back, I want one of three outcomes: mitigate it, explicitly accept it, or change the plan, scope, budget, or date. A pilot coming back a second time gets the same treatment.
If the meeting minutes record a returning AI pilot as noted, the executive committee has paid for another quarter without deciding to.
If it has a business owner who’ll answer for the results, a baseline finance has signed off, results that hold up outside the pilot, a cost to run that the value covers, and a fallback that’s been tested once without the agent, scale it. Tie each release of money to the measures in the business case moving, and keep measuring after it scales. A steering group in the owner column doesn’t count; I wrote in August about why a committee cannot own a change.
If one specific question is still open, change it. Change means the next round differs from the last: one question, the evidence that would settle it, a capped budget, a decision date, and the name of the person who’ll decide. If the next quarter's plan is the last one with new dates, I’d send it back.
Some evidence does take time. You can only measure an I pilot on a quarterly process at quarter end, and sometimes a dependency won’t be ready, so pausing makes sense. Both are fine as decisions, as long as you agree on the question and date before the review. If they first appear in the meeting, you're putting off a decision.
If the evidence isn’t there any more, stop it. Write down what the pilot taught you, get the sponsor to sign off, and move the people and money in the same meeting so the budget doesn’t drift into the next pilot by default. A pilot that answers its question with a no has still done its job. The point of all three is to get the strong pilots to scale sooner and the weak ones off the budget, so the next idea gets its turn.
Doing nothing has a cost too, and it’s the one the minutes don’t show. If the minutes say the pilot was noted, the executive committee has paid for another quarter without deciding to.
Keep the person who asks apart from the person who decides
None of this needs a new governance process. One page, a date and clear names will do.
Around every returning AI pilot, name who recommends, who answers for the value, who decides on the money, and who keeps asking.
For each AI pilot coming back this quarter, write down the decision it’s meant to inform and the date it has to inform it by. Then I’d keep four jobs apart: the programme director brings the recommendation, the business owner answers for the value, the sponsor or the investment committee makes the funding decision, and one person with nothing riding on a yes keeps the question on the agenda until it’s answered.
That person doesn’t need the authority to decide. In the research I mentioned, the role usually fell to people one level below the leadership team, near enough to the work to make the questions specific. On an AI pilot, that might be the AI governance lead or a finance business partner.
A board can ask its own version of this: “Which AI pilots have come back more than once without a decision, and who is answering for each one?”
Before the next review, I’d ask the programme director for that one page, in time to read before the meeting.
If the next quarter can’t change the decision, take the decision now.
Roman
New here? Start with what to do when your business case runs years longer than your programme plan.
The observation that an extension "doesn't look like a decision" but "looks like continuity" describes exactly how pilots survive without anyone owning them. Treating another quarter as a new funding request, with scale, change or stop and a named decision-maker, is the same discipline that a project review applies to a change order. When the team asks for more time, which single question has most often revealed that the next quarter would not actually change the decision?