How to Tell the Board the Truth About AI Without Losing the Room

Boards across every industry are arriving at the same position: AI is a strategic priority. The mandate has come from somewhere, market pressure, competitor activity, investor expectations, a compelling conference keynote, and it has landed on the technology leadership as a brief to deliver.

The problem is what happens next.

In most organisations, the honest assessment of AI readiness would be uncomfortable to deliver and uncomfortable to receive. The data is messy. The processes aren't documented well enough to automate. The team doesn't have the skills yet. The quick wins are smaller than the pitch decks suggest and the foundations required for the bigger wins are more expensive and time-consuming than anyone wants to acknowledge.

So the message gets softened. Not through dishonesty, exactly, but through the entirely human instinct to tell a room full of powerful people what they want to hear. The roadmap gets presented with confident timelines. The dependencies get mentioned briefly and not dwelt on. The risks get a slide that nobody spends much time on.

And then, twelve months later, the board asks why the AI initiative hasn't delivered what was promised, and the real conversation finally happens, at much greater cost than it would have had at the start.

This post is about how to create the conditions where that first conversation is honest. It's written for board members, because the conditions are largely yours to set.

What "We Need AI" Actually Means

When a board mandates AI, it's almost always expressing one of three underlying concerns, and which one it is matters enormously for what the honest response looks like.

The first is cost reduction: the belief that AI can do work currently being done by people, at lower cost. This is the most immediately testable proposition, and in many cases there's genuine substance to it. But the realistic scope of near-term automation is usually narrower than the headline number in the McKinsey report, and the transition costs, retraining, process redesign, change management, are consistently underweighted.

The second is revenue growth: the belief that AI can help the business grow faster, serve customers better, or create new products. This is where the most exciting possibilities live and also where the most unrealistic expectations cluster. The path from "AI can personalise customer experiences" to "AI is generating measurable revenue uplift" is considerably longer and more dependent on foundational data work than most board conversations acknowledge.

The third is competitive anxiety: the fear that if we don't do something, a competitor will, and we'll be left behind. This is the most psychologically powerful driver and the least useful basis for a technology investment decision. Doing AI because everyone else appears to be doing AI is how organisations end up with expensive pilots that don't connect to any strategic outcome.

The technology leader who can diagnose which of these three is driving the board conversation, and respond to the actual concern rather than the stated brief, is doing their most valuable work before they've written a single line of strategy.

The Conversation That Doesn't Happen Often Enough

There is a slide that belongs in almost every AI strategy presentation that almost never appears.

Call it the readiness slide. It covers, honestly, the current state of the organisation's data, its processes, its technical infrastructure, and its capability. Not what these things could be with investment, but what they actually are right now.

The reason this slide rarely appears is that it's uncomfortable to present and uncomfortable to receive. Acknowledging that your CRM contains five years of inconsistent, partially duplicated customer data is not a message any technology leader looks forward to delivering to a board that just asked them to build an AI layer on top of it. But the alternative, building the AI layer without fixing the data, produces an AI application grounded in unreliable information, which produces outputs that undermine confidence in the whole programme.

The data quality problem is the most common hidden dependency in AI initiatives, and it is almost always more expensive to fix than people expect. Not impossible, not a reason not to proceed, but a dependency that needs to be named clearly and sequenced into the plan honestly, or it becomes the reason the initiative stalls twelve months in.

The same applies to process documentation. AI systems that automate processes need the processes to be defined well enough to automate. In most organisations, a significant proportion of operational processes live in people's heads rather than in any documented form. The work of making those processes explicit is a prerequisite for automating them, and it is unglamorous, time-consuming, and not something that appears in the capability demonstration.

The Short, Medium, and Long Game

One of the most useful things a board can ask for, and rarely does, is a clear articulation of the AI roadmap across three time horizons, with an honest account of what each one requires.

The short term is what can be done in the next six to twelve months with the organisation as it currently is. Not what could be done with perfect data and a fully capable team. What can actually be done now. For most organisations, this is a narrower set of applications than the ambition suggests: targeted automation of specific high-volume processes, productivity tools for knowledge workers, early-stage experimentation in one or two domains. Meaningful, but bounded.

The medium term, twelve to thirty-six months, is where the foundational investments start to pay off. Improved data infrastructure, capability building in the team, refined processes that are now documented well enough to scale. The AI applications that become possible here are more significant than what's available in the short term, but they are dependent on the foundational work being done, and that work needs to start now even though the payoff isn't immediate.

The long term is genuinely uncertain, and a technology leader who presents a confident thirty-six-plus month AI roadmap is either unusually prescient or not being entirely honest about how fast this technology is moving. What's knowable is the direction: organisations that build strong data foundations, genuine AI capability in their teams, and the organisational discipline to evaluate and adopt new tools as they emerge will be significantly better positioned than those that don't. The specific applications that will matter in three years are harder to call.

This framing gives the board something it rarely gets: a realistic sense of sequencing, a clear picture of what success looks like at each stage, and an honest account of the dependencies.

The Governance Questions Worth Asking

Beyond the strategy, there are governance questions that boards have a specific responsibility to ask and that often don't get asked clearly enough.

Who owns the output of AI systems? When an AI model produces a recommendation, an analysis, or a decision, who is accountable for it? This is not a hypothetical question. In regulated industries it has specific legal dimensions. In every industry it has accountability dimensions. The answer needs to be thought through before the system is in production, not after something goes wrong.

What happens when it's wrong? AI systems make mistakes. The more consequential the decisions they're informing, the more important it is to have a clear process for catching errors, understanding why they occurred, and correcting them. This is the failure mode conversation that most AI strategy presentations treat as a footnote.

What data are we training on and what are we giving away? The data privacy and IP questions around AI tools are genuinely complex and evolving. Many off-the-shelf AI products have terms of service that are unfavourable from a data ownership perspective, and organisations are often not reading them carefully before deployment. This is a governance exposure that boards are right to inquire about directly.

What is our dependency risk? If the AI system that's been embedded in your operations is suddenly unavailable, degraded, or significantly more expensive, what is the impact on the business? Vendor dependency is not a new risk, but the pace at which organisations are embedding AI into core operations is creating new concentrations of dependency that deserve board-level visibility.

Creating the Conditions for Honesty

If you're a board member reading this and recognising that you're probably getting a softened version of the AI reality in your organisation, the question is what to do about it.

The most useful thing is to make honesty safe. Explicitly. Boards that respond to uncomfortable assessments with frustration or impatience teach their leadership teams to deliver comfortable assessments instead. The board that says "we want to hear the real constraints before we commit to the plan" will get better information than the board that says "we need an AI strategy by next quarter."

Asking specifically for the readiness assessment, not as a sidebar but as a central part of the strategy conversation, sends a signal about what kind of information you value.

And extending meaningful patience for the foundational work that doesn't produce visible AI outputs. Data quality, process documentation, capability building: these are not glamorous investments and they don't generate compelling quarterly updates. But they are the prerequisite for everything else, and an organisation that rushes past them to get to the exciting part will find itself rebuilding them under pressure later.

The technology leaders who will give you the most honest assessment of AI in your organisation are the ones who trust that honesty is what you're actually asking for. Creating that trust is, in the end, the board's job.

The AI strategy conversation at board level is one of the more consequential discussions organisations are having right now. If you're preparing for it, or trying to work out how to structure it more productively, I'm happy to think it through with you.