Before the Journey: How a Board Decides on AI

This is the second of these letters. The first described a world that has reset across four dimensions at once — the technological (AI), the geopolitical, the financial, and the risk reset — four structural shifts arriving together, each a significant change in its own right, and operating in a way where each one influences the others: the cascade. Of the four, artificial intelligence is the most prominent and visible change, and the one most often misread: treated as a technology deployment when it is really a question about what your institution should become, now that intelligence itself can be manufactured in a data centre. That letter set out the doctrines for navigating the reset, with particular focus on The AI Doctrine, along with the case studies behind it.

This letter turns to an even more fundamental decision in front of the organisation: should it pursue the AI journey at all, or wait — and we offer a structured way of answering that question.

OneThe Hardest Call

Why is the decision on AI so genuinely hard for so many organisations? As you can see from the evidence below, some are rushing in too soon, while many are paralysed, waiting for clarity. But if we break the question down, it is not really about timing — are we too early, or too late? That is the surface. Underneath sit three foundational questions, and they arrive together:

• The technology is evolving so fast that by the time we have formed a view on what it can do, it has already moved to the next stage. Should we wait for it to settle so we are clear on how to implement it — or, in waiting, are we missing the benefits others are already accruing?
• The investment being asked of us is large — money, time, resources — and we are not sure of the true return.
• The proof we would normally demand is thin. For all the hype, there are remarkably few places we can point to where this has been an unambiguous, durable success — and a great many visible, expensive failures, including at companies far better resourced than us.

So we are being asked to commit serious resources, on a fast-moving technology, with little proof of success and a long list of cautionary tales — and to do it now. We are caught in a whirlwind: move too early, and we may be pouring money into something the market is already whispering is a bubble; wait too long, and a competitor may race ahead, building a moat that becomes very hard to close. From where we sit, both directions look reckless. That is what makes this decision genuinely hard — not that the right answer is difficult to execute, but that it is difficult to know, and we are expected to know.

You are not alone in this. The exact tension has now been measured, and it is present in boardrooms across the globe.

Here is my take. In this field, urgency is running ahead of understanding — which means the most valuable thing we can do is neither speed up nor slow down on instinct, but step back and decide well. That is what this letter is for. What I lay in front of you is a structure — a way to reason through this decision so that the judgement you bring to it has something solid to work with. There is no single right answer here, and no standard answer that fits every organisation. This structure exists to help you find the right decision for your company.

TwoTwo Journeys

Before we work through the decision, let us understand its consequences through two examples — because the fastest way to see what is at stake is to look at how this has actually gone for those ahead of us. And it has gone two very different ways.

The honest shape of it: most attempts to make AI pay have failed, and a small number have built something close to extraordinary. The gap between them is almost never about who had the better technology. It is about how they went about the work.

Hold the two side by side. A well-thought-through decision, patiently seen through, can deliver extraordinary results. A hurried decision, or one taken from the fear of being left behind, tends to deliver a great deal of activity and very little outcome. The difference is not luck, and it is not the technology. It is the quality of the decision and the discipline of the build. Which raises the real question — the one this letter exists to answer. Before committing to a journey like this, what should we be asking?

ThreeSeven Dimensions of One Decision

The decision about AI is not a simple yes or no. It is better understood through seven dimensions, which I will walk you through in turn. At a high level, they are these:

• 1 · Technology — Should we wait for the technology to settle, or move now? What pace and posture should we take?
• 2 · Investment — How do we judge the return? What funding model ensures the money delivers outcomes, rather than being spent without result?
• 3 · Ambition — What business outcomes do we want from AI, and over what timeline?
• 4 · Risk & Liability — What unforeseen risks and liabilities does AI carry? How do we govern a machine making decisions outside direct human control?
• 5 · Ownership — Who will own this transformation? Who is accountable for the outcomes, and who will champion the journey?
• 6 · People — What will our people feel when we deploy AI? Acceptance or resistance — and will our own leaders support the rollout, or quietly resist it?
• 7 · Readiness — Is the organisation actually ready? How do we assess that, and what should we put in place before we begin?

The culmination of all seven is what informs the decision. As I take you through each, I will bring real-world examples — successes and failures both — so you can see the opportunities and the risks clearly, and reason your own way to the answer.

1 · Technology

Let us address the core question on technology first: should we wait for it to fully evolve and settle before architecting the business around it, or make the decision now to embrace AI? To answer that, we first need to understand how this technology is actually evolving. And here, for once, we can replace anxiety with measurement.

This is not new to us. We have seen an evolution shaped like this before, with Moore's Law — and we will see the same with AI. This is not a technology that will evolve and then stabilise; it will keep evolving. Which means there is no good, still moment at which to step in. Waiting for one is waiting for something that will not arrive.

The smarter approach is to account for the change as you transform the organisation — to build the capacity and capability to move with a moving target, rather than betting on a fixed picture of it that will be obsolete soon. So: start moving, but in a deliberate, conscious way, building the core capabilities as you go.

2 · Investment

Now let us address the other big concern: investment and return. The signs of investment going wrong are already visible at the most sophisticated companies in the world.

This brings investment squarely into question, and the fundamental ask is how do we ensure we deliver business outcomes that are sustainable and deliver the return. We do not want to sign a cheque and wait three or four years to find out. This is where I propose a cycle-funding model — not something new, but a discipline that holds steady through the challenging times.

There is a base investment any start requires — to get the organisation ready, to train a meaningful proportion of people, to acquire the core licences and platform. Beyond that base, the board should approve funding only one cycle at a time — each cycle carrying specific objectives and accomplishments to be reached by its end, usually a quarter. At the end of the cycle, the sponsors come back to the board and explain what was achieved, what failed, what was learned, what they propose for the next cycle, and the funding it requires. Every quarter, the board has a real opportunity to review progress and decide whether to release the next round — keeping the whole effort moving toward measurable outcomes.

This does something important: it pushes the programme teams to pick a focused area and demonstrate the outcome there before rolling it out across the whole organisation at once. The steps are smaller, but the lessons are seen actively, at every level — and the teams have to earn the right to spend the next quarter's money. It is a far stronger discipline than a blanket cheque written up front.

3 · Ambition

This is a critical dimension, because the starting approach of a journey is so often divergent from its real goal. Most organisations say they want revenue from AI — yet the starting point is usually the deployment of AI licences across the organisation. That goal of revenue cannot be reached through a laissez-faire, sequential approach.

I have also heard many leaders talk about their goal as AI agents as a percentage of the workforce — some saying half the workforce in the future could be AI agents — and they usually frame those agents as a way of automating current human tasks to gain productivity. I think that framing and approach is wrong. Why? The way humans are organised — into functions, specialities, and capabilities — and the way we execute work, through hand-offs and processes, are themselves shaped by human limitations: limited memory, limited span of attention, focused capability. Those structures and processes are inherently inefficient because of those human limitations.

Automate them as they are, and you have not made the organisation efficient — it is inefficiency at machine scale.

You have built a more efficient version of the wrong thing. So it is always better to step back and reimagine the work in the context of AI, rather than automate the work as it stands. And that is exactly why the right measure is business outcomes, not a percentage of the workforce replaced.

With that frame, I recommend the journey begin along three parallel tracks, not in sequence:

Adopt & Automate

Deploy AI broadly across the organisation, get people using it in their own work, let them automate tasks and find efficiencies. The point is not the efficiency itself — it is to make the organisation conversant with AI, and, more importantly, to tap into what I call hinterland innovation: the unprompted use that surfaces ideas no roadmap would ever have planned, when people are simply given these tools and left to use them however they wish, with no central direction.

Reimagine the Work

This is organisation-directed: a deliberate effort to re-engineer or reimagine a particular product, business, or function toward a defined business outcome. It typically has a business and a technology leader as joint sponsors, a cross-functional team doing the work, and a report back to the board carrying both the results and the lessons — for learning and for setting direction.

Build the New

While the first two tracks are largely about efficiency and productivity, the board and chief executive should make a conscious, separate effort to launch a genuinely new, AI-native business. This needs a dedicated, named leader and specific revenue objectives, with the mandate to build from the ground up. Many treat this as something to attempt only after the first two tracks have proved themselves. I would strongly recommend the opposite: begin it on day one, in parallel with the others. It is what allows an organisation to reach real revenue from AI sooner than most — and to build a genuine distinction in the market.

The best way to reach business outcomes from AI is not to pick one of these tracks, but to run all three as one system — each feeding the others, all of them in motion together rather than in a queue.

4 · Risk & Liability

This is the biggest concern that plagues most boards, and rightly so — because managing risk is a defined, central responsibility of the board, and risk is always mitigated through controls. But for the first time, what is being deployed into the business is an intelligent machine, and the concern is fundamental: can we control it? Can we keep it within the organisation's guardrails — and, independent of whatever benefits it delivers, could it create liabilities that exceed those benefits?

These are not unfounded fears. It helps to see the risk surface not as a pile of frightening stories but as a set of distinct categories, because that is how you begin to govern it.

The risk surface — a few categories

C · Models. There are two angles we need to cover. The first is geopolitical risk: governments have started imposing controls on the model layer — the software — beyond their earlier controls on the hardware, the chips.

One line for the boardroom: the frontier-model capability your business runs on has a switch — and that switch can be controlled by a government.

The second angle is more universal — commercial dependency. If you build your business on a single provider's model and it is withdrawn, degraded, repriced, or switched off for reasons that have nothing to do with you, your business is exposed. Both angles are the same underlying risk: too much dependence on any single model.

And these are only the tip of the iceberg. A 2026 study by the vendor Sinch found that around three-quarters of organisations had been forced to roll back a live customer-facing AI agent for one reason or another.

So how do we mitigate all this — without retreating from AI, which only leads to the paralysis we have already diagnosed? Two things.

First, on the model risk specifically: build on an architecture of multiple models — local and open models alongside the frontier ones — rather than depending on any single provider. No single model is irreplaceable enough to be worth the concentration risk; they are increasingly substitutable, so treat them as interchangeable components. Done well, that one choice de-risks continuity (no single point of switch-off, whether by a government or a vendor), mitigates cost (route the cheaper models for what does not need frontier capability), and protects private and proprietary data (keep the sensitive material on models you control).

Second, and more deeply: change where in the process risk and legal sit. The organisations that scale AI safely do not treat compliance as a gate at the end that blesses or blocks finished work. They bring risk and legal in as co-authors from the first design session — so their constraints become design parameters, and the thing is built safe rather than litigated into safety afterwards. Just as important is building genuine AI fluency inside the support functions — risk, legal, compliance — so those teams can plan their mitigations while the organisation moves to capture the opportunity, rather than standing as blockers at the end.

Two further things belong on the board's agenda. There should be a dedicated working session with the board specifically on managing risk in the AI world — this is too consequential to fold into a general update. And it must be understood as an evolving space: as organisations adapt to AI, so do the regulators, who are actively reworking the rules. Any system or design built today has to be able to adapt to a changing risk and regulatory landscape — compliance here is a moving target, not a fixed one.

Let me close this dimension on the point that matters most: two things are equally true, and the board's task is to hold both at once. The first: mitigating these risks is critical — not as caution for its own sake, but so the organisation is never held liable, and never incurs losses that erase what AI was meant to deliver. The second: embracing AI is equally critical — it is what allows the organisation to leap ahead rather than be left behind. The mistake is to see these as opposing forces, where managing risk means moving slowly and moving fast means accepting danger. They are not in tension. A board that is genuinely informed — working in close coordination with its business, technology, and risk, compliance, and legal functions — does not choose between protecting the institution and advancing it. It does both, deliberately, at once.

5 · Ownership

Given the level of impact AI will have on the whole organisation, it is a no-brainer that ownership and accountability sit with the board and the chief executive. The larger and more useful question is: how should the organisation be structured to run this transformation — and who should be responsible for it? Here are four models to explore.

The chief-executive-led model puts responsibility at the top, with the executive team executing — and in practice this is often what is already happening; nearly three-quarters of chief executives say they are their company's chief decision-maker on AI. Its strength is that it treats AI as business architecture, owned where it should be. Its weakness is the paradox itself: it depends entirely on a chief executive who is truly fluent and engaged, and even then, that person cannot run it day to day — they are occupied with too many other priorities.

The technology-led model — running AI as a programme out of the CIO or CTO organisation — is being adopted widely today. The technologist has real capability and a deep understanding of the technology, but limited understanding of, and influence over, the business model. So the effort drifts toward automating existing processes — the first level of ambition. Its strength is genuine technical depth; its weakness is that the business-architecture decisions the work requires sit outside the owner's remit.

The federated, or dual, model has business and technology co-owning, with a dedicated transformation lead reporting to the chief executive, and AI embedded across the functions rather than walled into one. It is the structure closest to what the work actually needs.

Its strength is that the ownership matches the shape of the problem: AI is horizontal, so its ownership is horizontal. Its weakness is that the transformation-lead role can easily fall into the coordination-overhead trap — becoming a programme office with no real responsibility, rather than genuinely driving the AI in close conjunction with the technology and business leaders.

The chief-AI-officer model is where the language gets confusing, so let me be careful. The market has begun calling the transformation lead described above a "Chief AI Officer." So the title itself tells you little; what matters is the mandate. The version I want to put in front of you is not a coordinator but a builder. Picture the role as a seeding ground: a senior owner who works across the existing businesses to transform them, in close conjunction with the technology and business leaders, and is also given the mandate to build the new AI-native business itself. That turns the role from integrative to genuinely future-building. Its strength is clear senior ownership joined to a builder's mandate. Its weakness is real, and I will name it: it demands an exceptional individual and a true mandate, and done weakly it recreates the very silo the title is usually criticised for. I will also be candid that, this early, I cannot point you to a named institution running precisely this reframed version with a documented result.

There is no right model in the abstract. There is only a fit — to the organisation, the culture, the company's track record, and the depth of its business and technical talent. That said, I will give you my own view plainly: I would urge you to consider a dedicated Chief AI Officer charged with building the AI-native business in parallel, because it gives the company a head start on real revenue and creates a genuine distinction in the marketplace.

6 · People

The earlier dimensions were all about defining the AI transformation. This one determines whether the board and chief executive actually achieve the ambition they have set — because history is clear that no transformation succeeds without the active participation of its leaders and its people. Change management has always existed for exactly this reason: to bring people along, to win their support and participation in organisational change.

But AI transformation presents a far bigger challenge. Almost every article and headline carries the message that AI will eliminate jobs — and that fear is real, widespread, and certainly present among your own people. Alongside it runs the talk of delayering, and of the changing role of the leader and manager in the new organisation. So the very leaders and managers who are meant to motivate and carry the associates are themselves uncertain about their own roles and futures. The people you would rely on to drive the change are unsettled by it.

Addressing the concerns of leaders and people properly is a large subject in its own right, and I will devote a future letter to it — not at the surface, but dissecting the root of these fears and building the response to them. This letter stays at the level of the initial conversation the board and chief executive must have.

Before embarking on the AI initiative, it is critical that the board and chief executive spend real time on how they will carry their people along the journey. Many leaders are saying they will reskill everyone, that there will be no job cuts, that almost everyone will have a place. But assurances given before the outcome is known are hard for people to believe, however sincerely they are meant. I would strongly recommend the board address this honestly, in three parts. First, honest communication: tell people the truth — that it is early, that we do not yet know the true extent of the change until we are further into the journey, and that we commit to keeping everyone updated, openly and regularly, telling the truth at each stage as it becomes clear, including the parts that are hard to hear. That is a commitment about how we will behave, which people can believe, rather than a promise about an outcome we cannot yet see. Second, reskilling that means something: not generic retraining, but active participation in the AI transformation that will make our people genuinely more valuable in the market — more in demand — whether or not there turns out to be a place for them here. That is a promise we can actually keep, and one a person can verify; it is the opposite of asking someone to train their own replacement. Third, a humane approach to exits: if exits do have to come, they will be handled with the most humane approach we can manage, and in a way that ensures the people affected genuinely share in the benefits the transformation creates.

It feels frightening to communicate any of this so early — before we even know whether AI will deliver for us. But being honest and engaging people at the very beginning is precisely what builds the trust that carries a transformation far. The alternative — masking the uncertainty with assurances that may not hold — buys quiet for a while, and then breaks exactly when you need people most. Trust is what turns people from wary observers into active participants, and their participation is what determines whether you reach the ambition you set. The honesty is not only the decent choice; it is the strategic one.

What is equally critical is effective communication — how you present the information, and the words you use.

So AI transformation requires a dedicated communication architecture — defined deliberately, to ensure the message is received correctly by all stakeholders: employees, leaders, investors, regulators, the press. There is as much focus on using the right words as on avoiding the wrong ones. I will cover that architecture in detail in a future letter. For now, it is enough to see that how you carry your people is not the soft wrapping around the decision. It is a major part of the decision.

7 · Readiness

Now the focus shifts to a pertinent question: is the organisation ready to begin this journey? The general norm is to conduct a multi-week organisational readiness check across many parameters to test true readiness. This section is something different — a board-level scan: a set of questions the board and the CXO team can work through together. Where they are answered well and agreed upon, the organisation is standing on firm ground. Where the answers are difficult, or there is disagreement in the room, that is even more valuable — because those are precisely the questions the CXO team should carry into the deeper organisational readiness study.

Below is a sample set of questions. As a first step, I would ask you to add any others you think belong on the list. The intent is never to get a simple yes-or-no — that is a generic response, and it does not force a conversation. These questions are built to generate discussion in the room, because the real insight and value live in the conversations, not in the questions.

These questions are not a scorecard — they are conversation generators, because there are no right or wrong answers here. You will know yourself whether you have arrived at the right answer or not.

FourThe Decision Before the Journey

We have come to the end of the letter, and by now you can infer that this is no longer a yes-or-no decision on AI transformation. Every organisation has to plan its own journey on this road — and working through the dimensions above, and the readiness scan, will have given you a real sense of what it takes to frame that path and begin. As I said in the readiness section, the value is more in the divergence in the room than in the agreement — because those conversations are what bring the real opportunities and the real issues to the table, and that is what eventually helps the organisation succeed.

Making the decision to begin is critical. But seeing the journey through asks far more than a decision — it asks courage, patience, and perseverance, sustained over years and through the troughs where results have not yet arrived, through the false starts, and through the moments where your people need to be held steady against their fears and their anxiety.

And here is the part that is rarely said aloud. Everything this letter asks you to do for your people — to be honest before the outcome is known, to hold steady through the troughs, to lead from ahead of where the organisation is — you are asked to do while standing in the same uncertainty they are, and under a pressure they cannot fully see. You are being asked to steady an organisation through a change you did not choose the timing of, cannot fully predict, and will be held accountable for — and to do it without the certainty everyone assumes you must have. There is no version of this where the pressure is delegated upward and away from you. It stops with you.

Which is exactly why it can only be led, not managed. A transformation this uncertain cannot be run from a plan or held together by a programme office; it holds together only because someone at the top has decided to carry it — to go ahead of where the organisation is, and carry its leaders, its associates, its investors, and its regulators through the change, including on the days the results have not yet come and the doubt is loudest.

That, in the end, is what this decision commits you to. Not a technology. A journey that only leadership can carry — and only because leadership chooses to.

A note on the evidence. Every case and figure in this letter is drawn from primary sources or the institutions' own disclosures, and where a claim rests on a settlement or a survey rather than a court ruling or an audited number, that distinction is kept. In a field moving this fast, figures are dated and refreshed before each use; where a verified example could not be found, that is said plainly rather than filled with invention.