What I Tell Every Board Before We Start Their AI Strategy

The first thing I tell boards is that their competitors are probably ahead of them on AI, and that their first instinct about what that means is probably wrong.

When executives discover they are behind on AI, their instinct is often to accelerate: to make decisions more quickly, to skip governance steps that feel like they are slowing things down, and to approve AI investments they have not fully evaluated because the competitive pressure feels urgent.

This is exactly the wrong response. The organisations that are genuinely ahead on AI are not ahead because they moved carelessly. They are ahead because they made a small number of good strategic decisions, early, and executed against them with discipline. The gap they have opened up is not primarily a technology gap. It is a capability gap: data infrastructure, governance frameworks, workforce AI literacy, and management muscle for AI delivery. These capabilities take time to build and cannot be shortcut.

The appropriate response to competitive lag is not acceleration without governance. It is a clear-eyed assessment of the two or three capability investments that will do most to close the gap, and disciplined execution against those investments.

The second thing I tell boards is that AI is very good at specific things, and those things are often not the things that boards expect.

Boards often come to AI strategy conversations with the expectation that AI will solve business model problems, competitive positioning problems, or management effectiveness problems. It will not. AI is extraordinarily good at processing information at scale, finding patterns in data, generating content from templates and examples, and automating repetitive cognitive tasks. It is not good at originating strategy, building relationships, making genuinely novel judgments, or compensating for management processes that are fundamentally broken.

The organisations that get the most from AI are those that deploy it against problems that are genuinely amenable to AI approaches: large volumes of structured or semi-structured information to process, repetitive cognitive tasks that benefit from speed and scale, pattern recognition in large datasets, and generation of first-draft content that experts then refine. Deploying AI against problems that require human judgment, relationship, or genuine strategic creativity will produce disappointment.

Almost every board I work with has a more optimistic view of their organisation's data quality than the evidence warrants. They believe their data is reasonably complete, reasonably accurate, and reasonably accessible. When we conduct a data readiness assessment as part of AI strategy development, the reality is usually considerably less positive.

This is not a criticism of the organisations involved. Data quality degradation is the natural outcome of years of system changes, mergers, manual workarounds, and governance gaps. It is also invisible until you try to do something with the data that exposes its limitations.

I tell boards to expect to discover data problems when they start AI programmes seriously, and to build time and budget into their AI strategy to address those problems. The organisations that plan for this discovery do not treat it as a failure when it happens. Those that do not plan for it treat every data quality issue as an unexpected obstacle, which slows the programme and undermines confidence in its viability.

The fourth thing I tell boards is that effective AI governance will sometimes feel like it is slowing things down, and that is not a bug but a feature.

The organisations that have moved fastest on AI have not skipped governance. They have designed governance that is proportionate to risk: lightweight for low-risk productivity AI, rigorous for consequential AI that affects customers, employees, or regulated activities. The governance feels fast to practitioners because it is proportionate. It feels like friction when governance is applied uniformly regardless of risk, which is what happens when governance is designed by compliance teams rather than by people who understand AI risk.

I help boards design governance frameworks that are proportionate rather than uniform. The output is governance that actually gets used, because it does not impose the same overhead on a Copilot productivity deployment that it imposes on an AI credit decisioning system. Both need governance. They do not need the same governance.

The last thing I tell every board is this: the most valuable thing I can do is tell you what is true rather than what is comfortable. The AI strategy that is most comfortable is often not the AI strategy that will deliver the most value.

A comfortable AI strategy confirms existing plans, validates existing investments, and suggests that the problems are manageable with more of what the organisation is already doing. An honest AI strategy often challenges existing plans, identifies where previous investments have not delivered the promised value, and requires difficult decisions about what to stop doing as well as what to start.

Boards that are willing to hear the honest assessment, and to act on it, are the ones that build AI programmes that genuinely transform their competitive position. Those that are primarily seeking validation of decisions already made will find consultants who will provide it. I am not one of them.