Making the case for an AI maintenance project to your board

An AI maintenance project does not usually die in the boardroom for lack of interest. It dies because the person championing it walks in with a conviction and walks out without a budget, having failed to put numbers on the table that the board could check. Enthusiasm does not get funded. A business case does.

Building that case is not a sales exercise. It is the opposite: it is how you earn the right to say honestly what the project will deliver, and to be believed precisely because you did not overpromise.

Start with the current cost, the only solid number

The first half of the calculation is the only part that is genuinely reliable, and it is the half most often skipped: what the task you want to lighten costs you today.

That cost can be measured without any tool. You take a repetitive task, estimate the time it takes per occurrence, its frequency over a year, and the number of people involved. Pulling together an equipment's history before a decision, re-keying values from a report into the CMMS, preparing an audit file, hunting for a procedure: each of these tasks has an annual cost you can express in hours, then convert using the fully loaded hourly rate.

This number has a virtue that projections do not: it is verifiable by the person listening to you. A production manager can confirm that reconstructing a history really does take half a day. That is what makes the case credible before you have even mentioned a saving. The "what AI could save you" tool, in the blog's free tools, exists precisely to build this first half, moment by moment.

State the saving as a hypothesis, and own it as one

The second half of the calculation is a hypothesis, and saying so is a mark of seriousness, not weakness.

No tool brings these times down to zero. Human verification is still required, difficult documents always need reading, and part of the work resists. So the right way to frame the saving is not "we will save forty per cent", but "this task costs us two hundred and forty hours a year, of which I reckon I can claw back half, and here is why". The top number is measured, the bottom one is owned, and the gap between the two is what the discussion is about.

This framing has a decisive tactical advantage: it turns the board into a co-author of the number. Instead of defending a percentage the board can reject outright, you are discussing a share that everyone can adjust. A case that invites you to correct the assumption is stronger than a case that imposes it.

What the case must count and everyone forgets

An honest case also counts what it costs, not just what it returns. That is what sets it apart from a pitch.

The entry cost first. An AI project means migrating the existing history, putting the tool in place, and training the teams. That cost is real, often underestimated, and ignoring it is paid for later in lost credibility. The heaviest item is migrating the history, covered in turning PDF reports into usable data.

The verification time next. AI shifts the work from lengthy re-keying to targeted, shorter verification, but that verification exists and takes qualified time. A saving calculation that forgets it announces economies that will not materialise, and the disappointment will do more harm to the next project than to the current one. The reason for that verification is developed in what AI decides and does not decide.

The cost of non-adoption finally, harder to quantify but real. A tool the teams do not use fills up with data that is wrong by omission. The budget for training and change management is not a side expense: it is what decides whether the rest of the investment produces anything at all.

What an honest case does not put a number on

You have to be able to say what cannot be calculated, because that is where dishonest cases give themselves away.

Avoided losses cannot be quantified in any defensible way: the decision taken better because the history was available, the unplanned downtime you could have anticipated, the measurement campaign you did not have to redo. These effects are real and often larger than the visible time saving, but putting them in pounds in a table would amount to inventing them. A good case cites them as expected benefits left unquantified, accepting that they do not count in the calculation.

Likewise, effects spread over several years, on unplanned downtime or on the overall budget, are measured after the fact, not before. Announcing a one-year percentage saving on those items would be dishonest, and an experienced board knows it. Better a case that says "here is what I can quantify, here is what I expect without quantifying it" than a case that gives the illusion of controlling everything.

Frame the risk, not just the saving

A board does not decide on return alone, it decides on risk. A case that only addresses the saving looks naive; a case that addresses the risk looks serious.

Two points are particularly reassuring. First, the fact that AI prepares but does not decide: no regulatory conclusion is reached by a system, responsibility stays human and identified. Second, the narrow scope of the first project, which caps the financial risk: you are not asking for a transformation budget, you are asking for enough to prove the point on one unit, as described in the roadmap for a first project in 90 days. A project that can be stopped cleanly after a quarter, on clear data, is a project a board dares to launch.

• Turning up with a conviction and no numbers.Enthusiasm does not get funded. The current cost of the task, in hours, is the first number to put down.
• Announcing a percentage saving with no origin.It discredits the whole case at the first question. Better an owned share than an imposed percentage.
• Forgetting the entry cost.Migrating the history, setting up, training: ignoring them is paid for in lost credibility when they surface.
• Not counting verification.AI shifts the work, it does not remove it. A saving that forgets verification will not materialise.
• Quantifying avoided losses.The better decision or the anticipated shutdown are real but cannot be put in pounds without inventing. Citing them as unquantified benefits is more robust.
• Talking only about the saving.A board decides on risk as much as on return. A bounded scope and a clear AI-human boundary reassure more than a flattering projection.