Imagine a CEO taking on a new role and discovering, in one of their first meetings, that the company lost a strategic customer two years earlier. They want to understand what happened, what was done in response, and whether other customers are now in a similar position. Getting an answer requires speaking to the head of sales, finding an old presentation, reviewing correspondence, pulling CRM data, and tracking down someone who was involved at the time and still remembers the details that never made it into the minutes. Eventually, the CEO will probably get a reasonable picture, but considerable effort will have gone into reconstructing knowledge the company already possessed.
This situation will be familiar to almost anyone who has worked in a large organization. The information exists, people remember parts of it, and the systems still hold plenty of traces. Yet whenever someone needs to understand the full story, the work of gathering, clarifying, and connecting begins again. Sometimes it takes a few hours, sometimes weeks, and sometimes there is no longer anyone who can fill in the gaps.
I think this raises a question we do not spend enough time on when we talk about learning organizations. We invest heavily in developing people, training managers, and recruiting experts, and rightly so. But the fact that people in an organization learn does not guarantee that the organization acquires a capability it can retain. A manager may learn an important lesson, a team may understand why a project failed, and a salesperson may develop a deep understanding of a customer. When they move to their next role, it is not always clear how much of that remains available to the people who follow them.
What interests me most about the idea of a Company Brain is that it offers a way to address this gap. There is, of course, plenty of technological excitement around it, and occasionally exaggerated promises. Behind them, though, lies a very old management problem: how to turn the experience people accumulate into a capability the organization can continue to use, even after the people and circumstances have changed.
Before connecting information, we need to agree on what it means
For decades, we have built systems to collect and manage organizational information. We know how to store transactions, invoices, contracts, support requests, work plans, and meeting minutes, and sometimes present them in impressive dashboards. Each system was built around a particular need and therefore reflects a particular view of the business. The difficulty begins when we try to understand something that crosses the boundaries between systems and departments.
Take a simple example. The head of sales reports that the company has 1,000 active customers, the CFO reports 850, and customer service puts the number at 1,200. There may not be a single error here: sales counts customers who purchased in the past year, finance counts paying customers, and service counts accounts that have opened a support request. Each describes something real, but uses the same term for a different population. Connecting an AI model to all three systems does not resolve the disagreement. It simply adds another participant that will have to decide which definition we meant.
This example matters to me because it is easy to skip this step when we are excited about asking questions in natural language. A convenient interface does not create shared meaning on its own. Someone still has to define an active customer, establish how to identify that customer across systems, take responsibility for the definition, and determine when it should apply. Technology teams refer to part of this infrastructure as a Semantic Layer, and may call the mapping of relationships between customers, products, people, and projects a Knowledge Graph. For leadership, the implication is quite practical: part of building an organizational brain is the work the organization must do to explain to itself how it operates.
This is also one reason I find it hard to see Company Brain as a project that can be solved by buying a product and connecting it to a document repository. Technology can help a great deal, but it will soon encounter the same gaps, contradictions, and disagreements that people in the organization already know about. A well-written answer may simply make us less likely to notice them.
What do we actually expect from an organizational brain?
Company Brain is not the name of an agreed architecture. In this article, I use it to describe a capability for connecting information, context, decisions, and outcomes in ways that help us understand the organization. An enterprise search engine can find a contract or a presentation, and an AI system can summarize it. For a more significant management capability to emerge, we need to connect what those documents say with everything else we know.
When a manager asks whether to expand the relationship with a particular customer, for example, they need more than a sales report. They need to understand the profitability of the relationship, the demands the customer places on support, commitments that remain unfulfilled, collection risks, and the customer's place in the company's plans. Some of that picture is in the data, some in contracts and correspondence, and some in the judgment of people who have worked with the customer for years. A system that brings these elements into the discussion, showing both its sources and the gaps between them, can substantially improve the starting point for a decision.
In many ways, this idea carries forward what we have been trying to achieve through knowledge management for years. What allows us to revisit it is the ability of language models to work with unstructured information, combined with better connections between systems and agents capable of taking defined actions. We can see parts of this direction in the context layers Microsoft is advancing, Glean's enterprise graph, and the approach Meta has described for improving institutional knowledge through expert feedback. These approaches differ, and they do not prove that the full vision has already been realized, but they show that the organization's unique context is taking a more central place in AI infrastructure.
To me, this is a development worth management's attention. As access to powerful models becomes more widespread, differences between companies may depend increasingly on how well they connect those models to what they know about their own business. A company can use AI to write a proposal faster, or it can use it to understand which proposal to submit, drawing on past experience, existing commitments, and delivery constraints. Both uses have value, but their effects on the company's capabilities are very different.
Keeping a record of a decision does not mean learning from it
When I try to assess what really justifies the term “organizational brain,” I come back to three capabilities: remembering, understanding, and learning. Remembering means preserving information and decisions along with their sources and circumstances; understanding means connecting them and distinguishing between fact, interpretation, and a binding decision; learning means using what happened afterward to improve our knowledge and judgment. This is not a scientific definition of intelligence, but it helps distinguish a useful information repository from a capability that improves over time.
Learning, in particular, requires us to change our habits. In most organizations, it is easier to find a record of a decision than of the assumptions behind it. We may discover that a pricing model was changed without necessarily understanding how we expected customers to respond, what alternatives were considered, or what would have counted as success. If churn rises six months later, it will be difficult to draw a serious lesson without returning to those questions. The assumption may have been wrong, execution may have been poor, or something else may have changed in the market. The fact that an outcome followed a decision is not enough to establish what we have learned.
The same difficulty arises with decisions born of compromise. Meeting minutes may record a decision to discontinue a product without explaining that the head of sales objected, operations identified a temporary constraint, and leadership preferred, at that moment, to invest resources elsewhere. Someone reading the decision two years later might interpret it as a matter of principle concerning the product or the market. An AI system can do exactly the same, only in a more authoritative voice.
Some of this context resides in what we call tacit knowledge: experience, familiarity with people, professional intuition, and understanding built over time. We cannot turn all of it into documents, and there is no reason to record every hallway conversation in an attempt to do so. But for significant decisions, we can document a little more: what we were trying to solve, why we chose this option, what we relied on, and what would justify reconsideration. That requires management attention and a culture that values explanation and correction. A system capable of reading the past well will be more useful as we learn to document the present better.
When does memory start getting in the way?
There is another aspect of institutional memory that I think deserves attention. We generally assume that more memory is a good thing, but organizations also need to know how to move beyond their own conclusions. Past experience is useful as long as we understand the conditions in which it was acquired and how far those conditions still hold.
A company that decided three years ago to stay out of a particular market may have made an excellent decision at the time. Distribution costs were high, demand was limited, and competition did not justify the investment. If costs have since fallen, demand has grown, and a major competitor has left the market, the old decision is no longer a sufficient answer. A system that retrieves it in every new discussion may give the historical position too much weight precisely because it is documented and backed by the authority of the leadership team that adopted it.
That is why we need to preserve a decision's assumptions alongside it, as well as when they were last examined. I would want a system to present the previous decision while also pointing out what has changed and what evidence warrants a fresh discussion. The original level of confidence matters too: there is a difference between a decision grounded in solid information and a choice made under pressure while much of the picture was still missing.
Without these mechanisms, an organizational brain may reinforce existing ways of thinking and make past mistakes more consistent. The risk grows if the system's outputs and recommendations return to the knowledge base and are later treated as independent evidence. The ability to learn therefore also requires room to challenge and correct, and a clear distinction between a source and a conclusion the system itself generated.
Trust, access, and the boundary between knowledge and authority
The more we want to rely on an organizational brain, the more demanding we need to be about how it presents what it knows. A useful answer should let us understand where it comes from, how current the sources are, and where uncertainty remains. When there are two conflicting accounts, the disagreement needs to stay visible until someone with the authority to resolve it does so. A smooth narrative that blends them together is not necessarily helpful to a manager who needs to make a decision.
The same applies to information that was never documented. A system may offer a plausible explanation for an old decision, but plausibility is not knowledge. I expect such a system to distinguish between a rationale recorded in the minutes and an explanation it inferred, and to allow people familiar with the event to correct it. Otherwise, we risk building an organizational history that sounds more coherent than it is faithful to what happened.
Access permissions are part of that trust. Connecting information sources to shared infrastructure does not remove the boundaries around salaries, sensitive deals, legal proceedings, or customer information. Those boundaries must also hold when a system summarizes information and combines several sources in a single answer. Who may see what needs to be part of the design from the outset, because exposure of sensitive information cannot always be remedied afterward.
Alongside access rights sits another question: what may be done on the basis of that knowledge. The fact that the company once gave a particular customer a 20% discount does not authorize an agent to offer it to the next customer. It may have been an exception approved under special circumstances, or policy may have changed since. As AI systems become able to update records, prepare proposals, and initiate processes, we will need to maintain the distinction between knowing the precedent and having the authority to act on it.
This also brings us to accountability. Technology teams can build the infrastructure, and data teams can address data quality and meaning, but deciding what constitutes binding policy and who may depart from it remains a business and management responsibility. Company Brain requires a partnership between these disciplines, with identifiable accountability for sources, definitions, and authority to act. Without it, we will be asking the system to settle questions the organization itself has yet to resolve.
How will we know whether the investment has changed anything?
It is relatively easy to measure how many documents have been connected, how many employees use the system, and how many questions they ask. These figures can help manage adoption, but to justify the investment I would also want to see changes in the work itself: a new manager understanding previous decisions more quickly, one team drawing on another's experience, or conflicting commitments being identified before they reach the customer.
What particularly interests me is whether we can see an improvement over time in the connection between experience and the next decision. When an error is discovered, is it corrected at the source and turned into knowledge others can use? When a project ends, does what we learned change anything in the planning of the next one? When we revisit a business decision, can we examine which assumptions held true without reconstructing the entire story from the participants' inboxes?
These questions are harder to measure, and we should be careful about promising better decisions simply because answers arrive faster. We need a baseline, outcome tracking, and an examination of other factors that influenced the results. Still, if we present the organizational brain as a strategic investment, we should judge it by the capability it adds to the organization.
And what remains for leadership?
I do not see this vision as a replacement for leadership. Management involves accountability, values, an understanding of people, and choices between interests that cannot always be reconciled. Even a complete picture does not necessarily tell us what to do, and certainly does not relieve anyone of responsibility for the consequences. But a better picture can support a more serious discussion and reduce the time we spend simply trying to understand where we stand.
Much of management's effort today goes into collecting data, reconciling accounts, reconstructing decisions, and bringing together people who each hold a different part of the picture. If we can reduce that effort, we can arrive at the discussion with more of the context already on the table. We will have more time to examine alternatives, ask what has changed, and exercise judgment where it is genuinely needed.
This will also raise expectations of leadership. When previous decisions are accessible, their assumptions visible, and their outcomes traceable, it becomes harder to justify repeating a mistake without examining what we already knew. To me, this is an opportunity to build a culture in which people are allowed to change their position but are expected to explain why; a culture in which experience is open to scrutiny and informs the work that follows.
Ultimately, that is why this idea interests me beyond the technology that enables it. For years, we have invested in helping our people become better at their jobs. We now have an opportunity to strengthen the way their experience stays within the organization and passes from person to person, team to team, and one decision to the next. It will not happen on its own, and not everything people know can or should be turned into digital knowledge. But a great deal of organizational experience is already being lost, and we can do a better job with it.
When I think about Company Brain, then, the question that concerns me is what will remain of all this learning when the people around the table change. If our organization learned something important yesterday, will it be wiser tomorrow even when the people who were there are no longer in the room? Finding a better answer may be one of AI's most significant contributions to management: allowing the experience we have already paid for to keep working for us.