Six boardroom conversations for turning AI into Enterprise Value

How should we change the way we manage the company?

About this article

A CFO recently asked me a question about AI that most boards are thinking but few are answering well: where is the ROI? People are faster. Automated processes are working. But the productivity gains are not translating into growth, margins or cash.

Part of the answer is tactical: wrong use cases, inadequate architecture. But the deeper reason is that many leadership teams have not yet recognised that AI is changing how the management system itself should operate.

AI is becoming part of how organisations sense, analyse, decide and act. It changes some of the constraints around which our current management systems and organisational structures were designed. Those systems were built partly for a world in which information was expensive, coordination was slow, decisions moved through hierarchies and functional boundaries helped make complexity manageable.

Capturing the value therefore requires a two-way adaptation: we need to change how we manage through the emerging Enterprise Brain, and become much more deliberate about how we design and steer that brain.

This article identifies six boardroom conversations that leadership teams need to have, from decision speed and workflow ownership to judgement, system control and what we are actually teaching the machine to pursue.

Introduction

A CFO recently challenged me with a simple question.

"Nice, all the AI tools. Where is the ROI? In reality, what will be fundamentally different a year from now?"

Good questions, because much of the AI conversation still centres on individual productivity and on finding process use cases that deliver actual positive ROI. People analyse faster, code faster, write faster and processes get automated. Useful, but those gains do not automatically translate into higher growth, better margins or more cash, let alone shareholder value.

Yes, part of the answer is that individual productivity is probably insufficiently focused on improving the customer proposition or simply selling more, and that companies still pick the wrong processes to automate or do this with inadequate scaffolding.

But it goes deeper. Think about what we now have in our hands. In a recent conversation with another CFO of one of the world's leading digital companies, she made two observations. First, AI is becoming a multiplier of decision-making: it changes the speed and economics of information, analysis, coordination and execution. Second, it is far less constrained by the functional boundaries through which organisations have traditionally divided complex work.

A simple fact with deep implications: 

"We have a new machine. We people need to learn how to work with it."

That means a two-way adaptation. We need to redesign how we manage so that we can make use of what I call the Enterprise Brain. And we need to design and steer the Enterprise Brain so that it actually works for the organisation.

Wait, Enterprise Brain? Think of it as the connected system of enterprise data, definitions, models, agents, rules and human interfaces through which information is interpreted, decisions are supported and, increasingly, actions are executed. It is an emerging intelligence layer connecting functions, workflows and decisions across the enterprise.

As that layer becomes more capable, boardroom conversations need to move beyond "Where can we use AI?" towards "How should the way we manage the company change now that this new intelligence exists?" I see six conversations that leadership teams increasingly need to have. Three concern how the organisation needs to adapt. Three concern how the Enterprise Brain itself needs to be steered.

Part I: How the management system needs to adapt

By management system, I mean the mechanisms through which a company sets direction, allocates resources, makes decisions, coordinates across boundaries, manages performance, controls risk and develops people. 

1. Decision velocity: is our management system operating at the wrong speed?

One of the most striking observations in my recent CFO conversation was that AI is becoming a multiplier of decision-making.

Historically, many management practices were built around the time and cost required to collect information. Monthly reports. Quarterly reviews. Annual budgets. Layers of approvals. Steering committees. Extensive preparation before decisions reached senior management. Those cadences made sense when information was expensive to assemble.

But what happens when information that took weeks to collect can increasingly be assembled in seconds? Making the analyst faster captures only a fraction of that opportunity. The bigger implication is that the speed of management itself can change.

That should force leadership teams to reconsider decision rights, delegation, meeting cadence, planning and forecasting cycles, budgeting, approval levels and how much information is required before acting.

It also changes how we think about risk. Finance has traditionally been very good at protecting organisations against the cost of making the wrong decision. But in a faster-moving environment, leadership teams also need to make the cost of waiting explicit.

Some decisions rightly require careful deliberation. Others become more dangerous the longer they remain unresolved. A useful principle here is to consider both reversibility and enterprise impact. A reversible, low-impact decision can move extremely quickly. An irreversible, high-impact decision deserves considerably greater scrutiny. Between those extremes, stage-gated decision-making becomes increasingly powerful: make an initial commitment, learn quickly, and increase investment as uncertainty falls.

Questions for the board:

If AI compresses the information cycle from weeks or months to minutes or seconds, which parts of our management system are still operating at yesterday's speed?

Have we made the cost of waiting as explicit as the risk of acting?

2. End-to-end value: why are we still managing horizontally created value through vertical organisations?

This takes us directly to the ROI problem. Much of the first wave of AI has focused on individual productivity. One person saves two hours here. Another team produces a report 40 per cent faster there. Someone claims a task is now 58 per cent more efficient. But a 58 per cent productivity improvement in one activity or process can create almost zero enterprise value if the bottleneck simply moves somewhere else.

Value tends to flow horizontally through an organisation. A customer journey crosses marketing, sales, service, operations, technology and finance. Pricing may depend on commercial data, market information, supply constraints and financial objectives. Order-to-cash, procure-to-pay, product development and customer service all cut across functional boundaries.

Yet most organisations are still managed primarily through vertical structures: functional leaders, functional budgets, functional targets, functional teams, functional technology ownership. There are good reasons for this. Functional expertise, accountability and scale matter. But functional boundaries also reflect the practical limits of what humans have historically been able to oversee and coordinate. AI may therefore do more than make the existing organisation more efficient. It may weaken some of the reasons why we organised work that way in the first place.

AI changes that constraint. An Enterprise Brain can increasingly connect data, logic and decisions across functions without requiring every connection to travel manually through layers of meetings, reports and coordination. That creates the possibility of managing more around complete value flows while retaining the specialist capabilities of functions.

It also exposes a structural obstacle to AI ROI. The biggest value may increasingly sit in redesigning a relatively small number of critical cross-functional workflows, rather than accumulating thousands of isolated use cases. The sequence becomes: What outcome genuinely matters? Which workflow creates it? Where are the bottlenecks, handovers and critical decisions? Where can AI materially change the economics or performance of the whole flow?

And then comes the organisational question: who owns it? In many companies, end-to-end ownership remains fragmented, while the real authority over people, budgets and systems still sits within functions. But ownership alone is insufficient. Targets, incentives and resource allocation also need to support the end-to-end outcome. Otherwise the organisation continues rewarding local optimisation while asking AI to improve the whole.

Questions for the board:

Which handful of cross-functional workflows have the greatest impact on enterprise value, and where could AI materially change their economics?

Who owns these workflows, and do our targets, incentives and resource-allocation mechanisms support the end-to-end outcome or pull people back towards functional optimisation?

3. Judgement through AI: how do we develop people when AI removes the work through which they learned?

AI will increasingly remove analytical and preparatory work traditionally performed by junior employees. That creates obvious productivity benefits. But that work served another purpose. It was how many of us learned how the organisation is wired.

We prepared analyses. Reconciled numbers. Chased information across departments. Sat in meetings. Built models. Discovered that what looked straightforward on paper rarely was. A lot of it was inefficient, but buried inside that inefficiency was apprenticeship.

This is why I think the choice between human judgement and AI is the wrong framing. The more interesting opportunity is judgement through AI. AI can help younger professionals interrogate assumptions, explore scenarios, compare alternatives, surface patterns and challenge their reasoning far faster than previous generations could. It can become part of how judgement is developed.

But this requires learning to be deliberately designed. That could include deeper onboarding that builds a genuine business lens (understanding the economics, customers, business model, workflows and dependencies), greater cross-functional exposure and rotations, and more decision-based learning where younger employees are given real business choices and asked what they would do and why.

AI itself can play a role here, for instance by testing assumptions and simulating alternatives before comparing their judgement with that of experienced leaders and with real-world outcomes. And there is real value in asking senior executives to make tacit judgement explicit: which signals did you notice, what did you ignore, and what made you act before all the information was available?

The point is to understand which learning experiences disappear when work is automated, and deliberately replace or improve them.

Questions for the board:

How do we redesign work and learning so that AI accelerates the development of judgement?

And a deliberately uncomfortable follow-up: which work should we retain, redesign or replace because its developmental value may exceed the efficiency gained by automating it?

Part II: How we need to steer the Enterprise Brain

4. Objectives and systems thinking: what exactly are we teaching the Enterprise Brain to pursue?

Once embedded in workflows, AI can pursue specified objectives at unprecedented speed, consistency and scale. That makes badly specified objectives more consequential.

Imagine an AI system driving sales purely for revenue. Another driving procurement purely for cost. Another driving operations purely for utilisation. Another driving finance purely for working capital. Each could improve its own KPI while reducing the value of the enterprise as a whole. AI can accelerate one-sidedness.

This is where systems thinking becomes important again. The first challenge is to prevent local optimisation from destroying system value. The second, more interesting opportunity is to use AI to redesign the system so that competing outcomes increasingly reinforce one another. 

AI can help organisations reconcile competing outcomes that previously appeared difficult to achieve simultaneously. Real-time automated monitoring can allow decisions to move faster because risk is continuously assessed (speed through control). Shared data, rules and intelligence can give local teams greater freedom while maintaining alignment (autonomy through coherence).

Automation can reduce cost while personalisation improves service (efficiency through customer intimacy). A common platform can maintain consistency while adapting execution to local circumstances (standardisation through localisation).

The Enterprise Brain therefore needs more than individual KPIs. It needs to understand which outcomes the organisation needs to hold together. This is where through-through thinking becomes particularly relevant, rather than either/or or and/and thinking. The objective is to design AI around the combination the business needs to achieve, and to understand what happens elsewhere in the system as the machine pursues it.

Questions for the board:

Which outcomes must our Enterprise Brain pursue together rather than maximising one at the expense of another?

What indirect consequences elsewhere in the system could an apparently successful AI intervention create?

5. Trust and control: what happens when managers increasingly supervise systems rather than work?

A CFO recently put another aspect of the change very clearly: "Before I had a human doing it. Now I've got an agent doing it. What are the checks and balances?"

That changes the nature of control. Historically, managers reviewed work produced by other humans. Increasingly, they will need to assure the systems that produce analysis, recommendations and actions. Trust becomes distributed across people, data, models, rules, controls and architecture.

The questions change accordingly. What can the system decide? What requires approval? When does it escalate? Which data may it use? How can its reasoning and actions be reconstructed? Where is human judgement required? Who remains accountable when a recommendation or action came from a machine?Management control remains essential, but more of it moves into the design and assurance of the system itself. Managers increasingly supervise the architecture that produces the work, alongside the people who work with it.

There is also an accountability question at the enterprise level: who is responsible for the coherence of the Enterprise Brain across technology, risk, data, operations and business ownership?

Question for the board:

What should our control model become when managers increasingly supervise systems rather than individual transactions, analyses and decisions?

6. Data and architecture: does our Enterprise Brain actually understand the company?

There is one uncomfortable reality beneath all of this. An Enterprise Brain cannot reason coherently if different parts of the organisation mean different things by basic concepts like customer, margin, volume, risk, forecast, product or revenue.

If definitions differ, systems do not connect, data ownership is unclear or the AI does not know which source is authoritative, the Enterprise Brain cannot form a coherent view of the organisation. This is why AI architecture, data architecture and semantic layers matter so much. They determine whether information can be understood consistently across functional boundaries.

This matters even more as AI begins to connect those boundaries. A human working inside one function may know that "margin" means something slightly different locally. An agent operating across sales, supply chain and finance needs to know exactly which definition applies, to what, and when. Otherwise organisations risk placing sophisticated intelligence on top of fragmented plumbing.

That also means boards should be careful about judging AI progress primarily by the number of visible tools and use cases. Some of the most important investments may be almost invisible: common definitions, trusted data, interoperability, permissions, architecture, authoritative sources.

Question for the board:

Do our data, definitions, architecture and sources of truth allow the Enterprise Brain to reason reliably across the enterprise, or are we placing intelligence on top of fragmented plumbing?

Taken together, the six conversations cover both sides of the challenge. 

On the organisation side: 

• Are our decision cycles still operating at yesterday's speed? 

• Are we managing vertically what creates value horizontally? 

• Are we developing judgement in people when AI removes the work through which they learned?

On the Enterprise Brain side: 

• Are we teaching it to pursue the right combination of outcomes? 

• Do we have a control model for supervising systems rather than work? 

• Does the machine actually understand the company it is reasoning about?

So where does the ROI actually come from?

The six conversations above concern the management system and the relationship between the organisation and its Enterprise Brain. But CFOs still need a disciplined way to translate those conversations into measurable value.

The six conversations define what needs to change in the management system. The six steps below provide a discipline for deciding where to invest and proving whether those changes actually create value.

1. Enterprise-value outcome. Start with the business. What do we expect to change in growth, margin, cash, risk or customer value? Avoid making "increase AI adoption" the objective. Think instead about outcomes such as improving conversion, reducing customer resolution time, increasing service quality, improving pricing, lowering working capital or shortening financial decision cycles.

2. The workflow that produces it. Which workflow creates that outcome? Where are the critical decisions, bottlenecks, handovers and sources of friction? This keeps the discussion focused on enterprise value rather than allowing it to fragment into disconnected use cases.

3. Foundation. Can the data, architecture, definitions, permissions and controls support what we want the system to do? If the foundation is weak, adding more AI will rarely solve the underlying problem.

4. Redesign around AI. Ask a more fundamental question than where to insert a copilot or an agent: if information retrieval, intelligence, analysis, coordination and execution suddenly became radically cheaper and faster, how would we design this workflow today? That may lead to fewer approvals, different decision rights, new roles, different controls, new interfaces between functions, or greater ownership of complete value flows.

5. Baseline and measure. Before implementation, establish what performance looks like today: cost, throughput, cycle time, quality, conversion, errors, customer experience, cash. Then trace the expected improvement from AI through workflow performance into a business outcome and ultimately into P&L, cash, risk or enterprise value. Otherwise impressive productivity percentages can remain economically meaningless.

6. Stage-gate and scale. Experiment cheaply. Measure. Learn. Increase investment as evidence improves. This brings disciplined capital allocation into AI investment and reduces the temptation to choose between endless experimentation and very large upfront programmes.

Where the boardroom conversation goes next

The deeper implication is that AI may reduce some of the constraints around which our current management systems and organisational structures were designed. If information becomes cheaper, decisions faster, coordination more cross-functional and execution increasingly machine-enabled, then some of the practices we built around scarcity and human limits should change with them.

Capturing the value therefore requires us both to change how we manage through the Enterprise Brain and to become much more deliberate about how we steer that brain. 

Perhaps that is also the answer to the CFO who asked where the ROI is. The boardroom question worth asking is: What has to change in the way we manage the company, and where will that difference show up in enterprise value?

***

Three related conversations, at three different levels

Seen together, there are three lines of thinking that belong to the same transformation. They reinforce each other, but they answer different questions:

1. AI adoption concerns how people learn and choose to work effectively with AI. https://www.intothenxt.com/post/leading-ai-transformation-human-behaviour-before-technology

2. Agentic Transformation concerns how work and process execution change as agents take over meaningful activities. https://www.intothenxt.com/post/agentic-transformation-what-boards-and-senior-executives-need-to-know-now

3. Management-system transformation, the subject of this article, concerns how the organisation itself should be managed once AI becomes part of how information flows, decisions are made and actions are executed.

About the author

Bas Kemme advises executive teams on how AI changes the way companies should be managed, not just the work inside them. His focus is on the management systems, decision practices and organisational designs that determine whether AI investment translates into enterprise value.

He has spent 25 years as a boardroom adviser across strategy, M&A, post-merger integration and transformation, working with companies including IKEA, Rabobank, ING, Bayer and BASF. That background informs his current work: helping leadership teams redesign how they manage as AI reshapes information, decisions and execution across functional boundaries.

Bas is the founder of IntotheNXT Consultancy and is co-authoring Tension to Traction with Fons Trompenaars, on how leaders can reconcile competing demands and turn organisational tensions into stronger performance.