Insight · AI readiness

12 AI Readiness Gaps Boards Are Beginning to Question

The questions have changed. A year ago boards asked what you were doing about AI. Now they ask what it returned, who owns it, and what happens when it is wrong.

For chief executives All sectors 14 min read

In brief

Boards have moved from asking whether the firm is doing something about AI to asking what it produced and who is accountable for it.
The twelve gaps below are the ones most likely to be exposed by that shift. Most are answerable within a quarter if the work starts now.
The position worth holding is not “we have an AI strategy”. It is “we can answer any of these twelve in the meeting, without preparing”.

The questions boards ask about AI have changed, and most leadership teams have not yet noticed how much.

A year or two ago the question was whether the firm was doing something. Being able to describe pilots and enthusiasm was a sufficient answer. It demonstrated the business was paying attention.

That is no longer where the conversation sits. Boards are asking what the spend produced, who owns each thing that is running, what happens when it gets something wrong, and whether the firm could explain any of it to a regulator or a customer.

The twelve gaps below are the ones that shift exposes. None of them require a programme to close. Most require a decision, an owner and a quarter of steady work. The uncomfortable part is that they are easier to close before the question arrives than during the meeting where it does.

01

Nobody can say what the AI spend has been, in total

Individual projects have budgets. What rarely exists is one number covering everything: build, licences, infrastructure, vendor features, internal time, the consultancy support.

Those sit in different budgets across different periods, so producing the total takes days and arrives with caveats. A board that asks the question expects an answer in the meeting, and the delay is itself the finding.

02

The strategy is a list of use cases rather than a position

Most mid-market AI strategies are inventories. Here are the things we are trying, here is the roadmap, here is what each might do.

A position is different. It says where AI matters most in this business and why, what the firm will not do, what has to be true for it to work, and how the money will be allocated between competing claims. Boards increasingly ask the second question and receive an answer to the first.

03

No single executive is accountable for the whole thing

Each use case has a sponsor. Each function has its own initiatives. The technology sits with IT. The models sit with data.

What is usually missing is one executive accountable for the portfolio — where it goes, what it returns, whether it should stop. Without that the board is being asked to fund a collection of departmental decisions and hold nobody in particular responsible for the total.

Boards have stopped asking whether you are doing something about AI. They are asking what it returned and who is answerable for it.

04

Nothing was measured before, so nothing can be proved after

The capability works and people use it. But if the process was never measured before the change, there is no comparison to make now.

So the answer to what it returned becomes an estimate, a general impression, or a comparison against a period that differed for other reasons. Each is arguable, and a finance director will argue with all three.

05

Nobody could list what is actually running

AI arrives through many routes. Built in-house, bought as a product, added by a vendor to something already licensed, adopted by a team without a purchase order.

No single route triggers a record, so no complete list exists. Every question that follows — about risk, cost, ownership or compliance — depends on a list nobody has, and assembling it is usually more revealing than anyone expects.

06

The data position was never assessed against what the use cases need

Firms know their data is imperfect. What they usually cannot say is whether it is good enough for a specific thing they intend to build.

Does it exist, can it be trusted, can it be reached automatically, who owns it, should AI be allowed to see it? Those five questions get answered during the build, when the answers are expensive, rather than before it, when they are nearly free.

Where do you sit?

Answering these before the board asks is a different exercise from answering them during the meeting.

A free Executive AI Opportunity Assessment. Forty-five minutes, online, with your leadership team. We arrive with a model of how firms in your sector make money and where AI has moved those numbers — and you tell us where it is wrong.

Book a free assessment

07

Nobody decided where a person must still make the call

Some outputs go straight through and some get reviewed. That split was usually decided by what was convenient to build rather than by what the consequences justify.

A board asking what happens when AI gets something wrong is asking about this. The answer should name the threshold, who set it, and why. In most firms it names whoever wrote the code.

What the questions have shifted to — a pattern, not a measurement

Was

Are we doing something?

Pilots and enthusiasm were a sufficient answer. Attention was the thing being demonstrated.

Now

What did it return?

Against what baseline, at what total cost, owned by whom.

To an investor, an auditor, an insurer or a corporate customer who asks in procurement.

Gap

Where most firms are

Answering the first question well while the board has moved to the second.

This describes the shift we see rather than a measured dataset. The test worth running is your own: pick three of the twelve at random and see whether you could answer them in a board meeting without preparing.

08

The work never changed, so the benefit could not arrive

The tools went in on top of the existing process. Nothing about how the job is done was redesigned. Adoption gets reported and the numbers do not move.

This is the most common pattern in the mid-market and the hardest to raise, because everything technically succeeded. The board approved a transformation and the firm bought a faster version of what it already did.

09

Nobody is tracking what the regulators are doing

Rules are moving across jurisdictions and sectors. Legal watches legal. Compliance watches its own regulator. IT watches technology. AI sits across all three and belongs to none.

So a board asking about the regulatory position gets an answer assembled on the spot from three partial views. The fix is naming someone accountable for that specifically, reporting periodically on what has changed.

For the chief executive

Three things worth doing before the next board meeting

Ask for the total AI spend to date, in one number. However long it takes, that duration tells you something.
Ask for a list of everything running with a named business owner against each. Most firms find things nobody reviewed.
Pick your largest AI investment and ask what it returned, against what baseline. If the answer needs building, so does the measurement.

10

The people whose jobs changed were never properly brought in

Training happened. What usually did not is the harder work: new decision rights, changed measures, jobs that look different, managers coached on running a team that works another way.

So adoption looks acceptable in the reporting and shallow in practice. People use the tool for the parts that suit them and work around the rest, which is a rational response to a change nobody explained properly.

11

The firm cannot show its working if someone asks

A customer asks why a decision about them went the way it did. An auditor asks how a model was validated. A corporate buyer asks about AI in procurement.

In most mid-market firms each of those takes weeks to answer, because nothing was recorded at the time. The information exists in fragments and nobody has ever assembled it. That gap becomes visible at the least convenient moment.

12

The board is being briefed on activity rather than on outcomes

The most structural of the twelve, because it hides the other eleven. The AI update describes pilots launched, tools deployed, people trained, adoption rates.

All of that is activity. None of it is outcome. A board reading activity reports has no way of knowing whether any of the other gaps exist, which is comfortable until the first person outside the room asks a harder question.

The fix is a different paper: total spend, portfolio position, what each thing returned against its case, what is running and who owns it, what stopped and why. It is a more uncomfortable document to write and a far better one to have written when the questions change.

What this means

None of these twelve require a programme. Most require a decision, an owner, and a quarter of unglamorous work. What makes them urgent is not that they are difficult but that the questions have already changed and most leadership teams are still answering the previous ones.

The pattern behind them is consistent. AI arrived through many routes, each locally sensible, none triggering the governance that a decision of that size would normally attract. So spend is unaggregated, ownership is unassigned, measurement was never designed, and the board is briefed on the only thing anyone can easily report — activity.

The economics favour acting now by a wide margin. Producing the total spend takes a week. Building the register takes a fortnight. Appointing an owner takes a meeting. Doing all of it while an investor, auditor or regulator waits is a different exercise entirely.

Which is the argument in one line. Change the work before you change the technology — and answer these twelve before someone else asks them, because the same answers are worth far less when they arrive under pressure.

The next step

Could you answer these in the meeting, without preparing?

Forty-five minutes with your leadership team, online, at no cost. We bring a working model of how firms in your sector make money and where AI has moved those numbers. You tell us where it is wrong. Then we tell you what we think — including “don’t invest”, if that is the honest answer.