The Three AI Questions That Get Boards the Wrong Answers

Board conversations about AI follow a predictable pattern. Someone asks a question, an executive gives a reassuring answer, the board nods, and the conversation moves on. Three questions come up again and again. All three sound reasonable. All three tend to produce answers that feel credible and turn out to be wrong.

The problem sits in the framing. The questions are too blunt to surface what a board actually needs to know, and most organisations have learned to answer them in ways that satisfy the room without creating any real accountability.

Here's what's being asked, what's being said in response, and what I'd ask instead.

Question one: "Are we using AI?"

The answer boards usually get: yes, several initiatives are underway, we've invested in a platform, the innovation team is running pilots.

That answer is technically true and operationally useless. It says nothing about whether AI is being used well, badly, safely, or at all in any meaningful sense.

The harder version of the question is: do we know what AI is being used for, by whom, with what governance, and producing what outcomes?

Most organisations don't know the answer. Shadow AI, the use of AI tools outside approved channels, is widespread. Employees are using ChatGPT, Claude, Gemini, and dozens of other tools on personal accounts to get through their work faster, without anyone's knowledge or permission. They're putting customer data, commercially sensitive documents, and internal processes into systems that have never been assessed, contracted, or governed.

The approved AI programme an executive team reports on is often the smaller part of the picture. The larger part is invisible, and finding it means auditing actual AI activity across the organisation rather than the workstreams that have a project code.

Understanding that activity in isolation isn't enough either. A board needs to know whether the organisation has the data foundations, governance structures, and people capability to support the AI it's deploying, or whether it's building on soft ground.

Question two: "How much are we spending on AI?"

The answer boards usually get: a budget figure, the AI programme's cost, sometimes broken out by year or workstream, occasionally benchmarked against peers.

Again, technically accurate and structurally misleading. It tells the board what the approved programme costs. It doesn't tell them what AI costs.

The better version of the question is: what is the total cost of AI in this organisation, including shadow AI subscriptions, departmental tools procured outside the programme, integration and maintenance work, and the ongoing cost of keeping AI systems running and governed?

The gap between these two numbers is routinely large. Departmental leads buy tools on procurement cards. IT teams fold AI features into existing systems as part of routine upgrades. Customer service licenses conversational AI without a formal business case. Legal and compliance run their own AI tools for contract review. None of this shows up in the AI programme budget, and much of it doesn't show up anywhere coherent.

That matters for three reasons. The board is approving a budget number that represents only part of the actual spend. The organisation is making investment decisions without a full picture of total commitment. And return-on-investment calculations made against the programme budget alone look very different once total cost is included.

A complete AI cost model captures direct costs, embedded costs, operational costs, and the cost of governance and compliance. Building it isn't complicated. It requires someone to ask the question across the whole organisation, not just the programme team. The AI Transformation Playbook at transformationplaybook.ai has templates for building exactly this.

Question three: "What's the ROI?"

This is the question that produces the most sophisticated-sounding wrong answers.

The usual response involves a business case: projected savings from automation, efficiency gains expressed as FTE equivalents, revenue opportunity from new AI-enabled capabilities. The numbers are often large. The confidence intervals are often invisible. The assumptions are often optimistic.

Business case projections built during the experiment phase are hypotheses, not results, and boards are frequently handed a hypothesis dressed up as one.

I'd replace "what's the ROI?" with: what has this programme learned, what has it proven, and what decisions has it enabled that we couldn't have made otherwise? That reframes the conversation from a financial justification exercise into a learning and accountability exercise. It asks the executive team to show their working: not the projected outcome, but the actual outcomes so far, the assumptions tested, and the decisions that now rest on evidence rather than projection.

Early and mid-phase AI programmes shouldn't be held to a simple ROI standard. They should be held to a learning standard. Is the organisation building genuine capability? Is it distinguishing between experiments that work and experiments that don't? Is adoption happening because people find the tools useful, or because the programme is reporting usage numbers that don't reflect real value?

Knowing how many people have accessed a tool is not the same as knowing whether they're using it productively, trusting its outputs, and changing how they work as a result. Those are different questions, and they produce different answers.

What the board should ask instead

If I were framing the AI agenda item for a board pack, I'd replace the three standard questions with five better ones:

• What AI activity is happening across this organisation that isn't being managed through the programme?
• What is the full cost of AI, including departmental and shadow spend, not the programme budget alone?
• What has the programme learned in the last quarter that changed how we're approaching this?
• Which experiments have we stopped, and why?
• What governance exists for AI decisions, and who is accountable when something goes wrong?

None of these require the board to understand AI technically. They require the executive team to demonstrate that it does, which is the right division of labour.

The fourth question tells you the most. Organisations learning from AI experiments stop things: they kill initiatives that don't work, reallocate resource, and change direction based on evidence. Programmes that are performing for the board keep everything alive, because stopping something feels like failure to whoever's presenting. Asking what's been stopped, and getting a clear answer, is a better signal of programme health than any business case projection.

The board's role

A board's job on AI is to ask questions that force the executive team to demonstrate it understands the technology.

The three questions boards typically ask are too easy to answer reassuringly. They create the appearance of oversight without the substance, the organisation learns to give the right-sounding answer, the board marks it done, and the real questions about governance, total cost, genuine learning, and accountability go unasked.

The organisations doing this well are the ones where the board asks uncomfortable questions and the executive team builds the systems and disciplines needed to answer them honestly, regardless of the size of the AI budget or the sophistication of the platform. That's a governance capability as much as an AI one, and the board is the one body positioned to drive it.