The Decision Before the Decision
Boards make far fewer decisions than management. Our role is more to oversee the people making them, the conditions under which those decisions are made, and their consequences.
That is a key boundary—the board should not move into management.
Yet as AI becomes part of how companies analyse information, develop alternatives and recommend action, directors may need greater visibility into how those decisions are being formed. Not to make the decision for management, but to exercise judgment over what eventually reaches the board.
That is the decision before the decision.
Freedom to choose — but who built the choice?
Management brings a recommendation. Directors review it, challenge it, test the assumptions, debate the alternatives and decide whether to approve, reject, defer or ask for something different—that is how the process is supposed to work.
But I have been thinking more about what happens before the board ever gets there.
By the time a major decision reaches the board—whether about strategy, an investment, an acquisition or a technology implementation—the problem has already been framed. Data has been selected, assumptions have been made and alternatives have been narrowed.
Some options may have been developed in detail while others have not. That is not a flaw. Management has to do that work; the board cannot start every important decision from a blank sheet of paper.
What is key is understanding how much of the eventual choice has already been shaped along the way.
That becomes more important as AI moves further into the process. AI can help management search more broadly, compare more information, model scenarios, identify patterns and test assumptions faster than before. It can improve the quality of the work that comes to the board.
But it can also make a recommendation look complete very quickly, as the analysis becomes cleaner, alternatives become easier to compare and one path begins to look more obvious than the others.
At that point, the board may still have full authority to decide. The harder question is how much room remains to exercise independent judgment over the way the choice itself was constructed.
That is the distinction I want to explore here. Not whether AI will make decisions for boards, but whether boards can preserve enough visibility into how decisions are being formed to know when they are truly choosing—and when they are mainly choosing among options that have already been shaped for them.
How the decision gets shaped before it reaches the board
I have worked on the other side of this process.
In corporate strategy, I led a team of more than 20 strategists selected from across the company. Our role was to bring together what the CEO needed to make major enterprise decisions and, ultimately, what the board would need to understand and approve.
That meant doing a great deal of work before a decision ever reached either room. We tested assumptions, looked across businesses and geographies, compared alternatives, challenged the numbers, considered what it would take to execute, and tried to understand interdependencies—where one decision would affect another part of the enterprise.
That work was important. Neither the CEO nor the board could work through every piece of analysis behind a major recommendation. But it also made something very clear to me: by the time a decision reaches the board, much of its shape is already there.
The planning horizon has been chosen. Return thresholds have been applied. Certain assumptions are built into the model. Some alternatives have been developed further than others. What appears practical has often been influenced by the capabilities, incentives and constraints of the organisation doing the analysis.
And decisions rarely stand alone.
A capital investment can make complete sense for one business unit and still be the wrong investment for the enterprise. I have seen that happen when a larger strategic decision made elsewhere changed the logic underneath an otherwise sound proposal.
This is what I mean by decision architecture. It is broader than governance structure or formal authority. It includes the information, incentives, assumptions, decision rights,
timing and organisational constraints that shape what eventually reaches the board as a choice—with technology now part of that architecture too.
The board does not need to recreate management’s analysis, but it does need to understand enough of the architecture to ask a different set of questions: Why did these become the alternatives? What assumptions matter most? What has already been ruled out? What other enterprise decisions should change how we look at this one?
AI raises the stakes because it can make this shaping faster and less visible. It can expand the analysis, but it can also accelerate convergence around an answer. That is where board judgment has to re-enter.
The business case before the business case
AI proposals usually arrive with a number attached: a projected return, a revenue lift, productivity gain, cost reduction.
Important data, but before the board evaluates the return, it needs to understand how the return was constructed.
What is being counted as value? What assumptions were made underneath it? My favourite question: what has to change in the operating model for that value to show up in the P&L? How much depends on execution, labour reduction, faster cycle times, higher conversion, better pricing, lower error rates, or something else entirely?
The math can be correct and the business case can still be weak, because the board is not simply evaluating a projected return—it is evaluating the logic that produced it.
This becomes harder with AI because the economics are moving quickly. The cost of capability can fall quickly. What seems expensive today may become commodity capability tomorrow. A custom build that appears differentiated now may have a much shorter useful life than the organisation understood.
That changes the way the board should look at capital allocation.
The relevant question is not only, “What does the AI cost?”
It is, “What outcome are we buying, and what has to be true for that outcome to create economic value?”
Depending on the enterprise, it may mean a faster quote, a resolved customer request, a better forecast, a lower error rate, higher throughput, improved pricing, or a better decision—because efficiency does not automatically become value.
Saving an employee 20 percent of their time does not translate to 20 percent back into earnings. The company has to decide what happens to that released capacity. Does it
reduce cost? Increase throughput? Improve service? Support growth? Or simply allow more activity without changing the economics?
The same applies to build-versus-buy decisions. The economic life of an AI investment may be much shorter than anticipated as capabilities outside the company continue to improve. That means useful life, switching costs, ownership of data and models, integration costs, and the ability to exit all become part of the investment case.
The board has to consider three different costs: the cost of moving too fast, waiting too long, and not understanding the economics well enough to know the difference.
Moving too fast could lock the company into an immature architecture or a very expensive capability. Waiting allows competitive gaps to widen. And ignorance, as always, can be expensive both ways.
AI capital allocation is not simply a technology decision. The board will not select the model or negotiate the contract. But it must understand how management is defining value, what assumptions are driving the return, what could make the investment obsolete, and what evidence would justify scaling, stopping or changing course.
The business case is not separate from the decision architecture. It is one of the things that shapes the decision before the board ever makes it.
Recognition capability
There is another side to this. Sometimes the problem is not how a decision was shaped before it reached the board. Sometimes the more important decision has not reached the board at all.
Boards spend a lot of time evaluating what is on the agenda. The harder question may be whether we are recognizing early enough when something in the environment has changed enough that it should be on the agenda.
I think of this as recognition capability. It is related to foresight, but not quite the same. Foresight helps us think about what may be coming. Recognition is knowing when something already on the radar has moved close enough, when the surrounding conditions have changed enough, that it now deserves a different level of attention.
Technology is full of examples of ideas that were visible well before they became viable. For me, the issue is not whether someone could imagine the future. It is recognizing when the technology, economics, customers, infrastructure or competitive environment have come together enough to make the opportunity real.
My own experience in innovation and applied technology shaped how I think about this. Technologies are often created for one purpose and later find much greater value in a different application, industry or business problem. I have seen that shift happen. The opportunity was not necessarily where the technology started, but where the conditions finally made it useful.
That is part of what I mean by recognition capability. Not simply seeing what is coming, but recognizing when something that has been around for some time suddenly matters in a different way.
AI can help here too. It can find signals across much more information than a board or management team could ever absorb, connect patterns and surface changes earlier. But seeing a signal is not the same as understanding its significance.
That still requires judgment. Is this noise, or has something changed? Is the timing right? Is the application different from what we first expected? And if the opportunity is real, is the organization actually ready to act?
Good governance is therefore not only about evaluating the decision that comes to the board. Sometimes it is recognizing that a different decision should exist before the organization has fully formulated it.
The board needs its own decision architecture
If management has an architecture through which a recommendation gets built, the board has one too. We may not always think of it that way. We think about the agenda, board materials, presentations, discussion time, committee work, perhaps an executive session. Yet all of those affect how directors pay attention, what gets challenged and ultimately how a decision gets made.
I was reminded of this recently in a discussion on the neuroscience of board decision-making. The practical ideas were fairly simple: interrupt autopilot, ask what problem we are really solving, surface the assumptions and evidence, use a premortem to imagine that the decision failed and ask why. Put the hardest issues where the board still has the energy and time to deal with them. Separate discussion from decision when the issue deserves it. Invite dissent instead of waiting for someone to create it.
None of this moves the board into management. It does something different. It creates the conditions in which the board can actually use the experience and judgment sitting around the table.
The way an issue is sequenced matters. So does whether directors receive thirty pages explaining a recommendation or enough information to understand the trade-offs and what remains uncertain. A consequential decision placed at the end of a crowded agenda does not get the same board as one given the time to be challenged.
AI makes this more important because the material coming to boards will become easier to synthesise and, probably, increasingly persuasive. That is mostly a good thing.
But a very coherent answer can also move a group toward agreement before the assumptions underneath it have received enough attention.
The balance is not easy. The board needs enough visibility to challenge how the decision was shaped, but not so much process that governance itself starts to slow the innovation the company needs.
The mix of lenses, not the AI seat
As boards look at AI, one response is understandable: do we need an AI expert around the table? Maybe. In some companies the answer may clearly be yes. But I would be careful not to turn this into a search for one person who can somehow “cover” AI for the rest of the board. We have seen versions of this before with technology and cyber, where expertise was added or delegated and the rest of the board could too easily step back from building its own understanding.
AI is broader than that. It cuts across strategy, capital allocation, operations, talent, customers, risk and, increasingly, the way decisions themselves are made.
I have used the idea of lenses before because that is how I tend to see a board. No single lens is enough. Strategy, finance, operations, technology, people, customers, regulation, risk—each can change how the same decision looks.
A director with deep technical knowledge may see limitations or opportunities others will miss. An operator may recognise that the economics look attractive but the organisation is not ready to execute. Someone with capital allocation experience may ask whether the capability being built today will still be differentiated by the time it is deployed. Another director may see the customer, workforce or regulatory consequence that has not yet entered the model.
The value is in what happens when those perspectives work together.
This is also where I think the board should be a counterbalance to management, not simply a mirror of it. If management already has strong technical capability, the board may not need more of exactly the same. The gap may be somewhere else—commercial judgment, operating experience, capital discipline, international exposure, talent, risk, or simply someone who has lived through a transformation that did not go according to plan.
That does not make technical expertise less important. It means the question is broader than whether there is an AI specialist on the board. What matters is whether the board has enough collective understanding to know what it is looking at, where to challenge, and when it needs deeper expertise from outside.
The same applies to board refreshment. A board can have very accomplished directors and still see the world through similar experiences. Independence on paper does not necessarily mean independence of thought.
So perhaps the better question for nominating and governance committees is not only, “Do we have AI expertise?” It is, “What are we likely to miss together?”
That is where the mix of lenses becomes part of the decision architecture itself.
Stewardship of capability: board and CEO together
At some point the discussion moves beyond whether to invest in AI. The company has decided to move, capital has been committed and management now has to make it work.
The CEO and management own the implementation. They decide how the organisation changes, where the technology sits, what processes need to be redesigned, who owns what and how people will work differently. The board should not run any of that. But neither should it approve the investment and simply wait for the promised return to appear.
This is where the original assumptions matter.
A pilot can work. Productivity can improve. People may use the technology and the technical milestones may all be met, yet the value may still not reach the P&L. Maybe the workflow did not change, incentives stayed the same, another part of the organisation became the constraint, or the capability was deployed where the economics were never strong enough in the first place.
That does not automatically mean the original decision was wrong. It may mean the conditions required to capture the value were not there, or changed along the way.
For me, this is where stewardship matters. The board and CEO need a common understanding of what value was expected, what had to change for that value to be captured and what evidence would make them revisit the original thesis.
And there has to be room to learn. Innovation rarely moves in a straight line. Some assumptions will be wrong, the application may change and what looked like the best use of the technology at the beginning may not be where the value finally appears.
The balance is not easy. The board needs enough visibility to understand whether the company is learning, whether the economics are holding and whether implementation is still moving toward enterprise value, without creating so much process that governance itself becomes an obstacle to the innovation the company needs.
That is not managing the implementation. It is making sure management is still governing against the value case that justified the investment in the first place.
Can you still explain the investment thesis?
That leads to another question I think boards should be able to answer sometime after the decision has been made: can we still explain why we made it?
Not the presentation or the original ROI number. The thesis underneath it.
What problem were we solving? Where did we believe the value would come from? Which assumptions mattered most? What did we expect to change in the operating model? What would have made us decide not to invest?
And now that we know more, which of those assumptions still hold?
This matters because AI investments will not all produce value on the same timetable. Some may show fairly direct cost or productivity gains. Others may create capability, customer value, better decisions or strategic options that take longer to show up in earnings.
The board needs to distinguish between a thesis that is taking time to mature and one that is no longer working.
That is harder than looking back at the original business case. It requires understanding which assumptions made the decision reasonable at the time and being willing to reopen them when the evidence changes.
The same question eventually has to stand up outside the boardroom, whether the audience is investors, owners, a sponsor or lenders. But the external explanation is really only a test of something more basic: do the board and management themselves still understand why the capital was committed and what would cause them to change course?
If they do not, the problem started long before the external question was asked.
AI is the current test, not the last one
AI is the technology forcing this conversation now, but it will not be the last one.
Quantum computing is already moving closer to commercial relevance. Other technologies will follow, some we can see and others we probably cannot. Each will arrive with its own promises, economics, risks and specialists telling boards why this one is different.
We cannot rebuild the board every time that happens.
That is why I come back to governance architecture and judgment rather than expertise alone. Directors need enough understanding of the technology to know what questions matter and when they need help. But the more enduring capability is being able to
understand what the technology changes in the business: the strategy, economics, operating model, people, risk and ultimately the decisions management has to make.
The tools will change. So will the speed and amount of information available to us. What should not change is the discipline around how important decisions are framed, challenged and revisited.
And perhaps that is the larger point.
The opportunity with AI is not simply to add another subject to the board agenda. It is to use this moment to become better at governing in an environment where technology will increasingly shape how the enterprise sees, evaluates and acts on its choices.
Because the next technology may be different. The decision before the decision will still be there.
For the latest research and opportunities to exchange ideas with fellow board directors, visit the INSEAD Corporate Governance Centre.
