Organizations Outsources Ambiguity Handling to People

A Monday Morning Question

Imagine it is Monday morning. A product manager opens an AI assistant connected to the organization’s data platform and asks what appears to be a straightforward question: “Who owns Customer?” The assistant searches the enterprise glossary and quickly finds three definitions. One describes a customer as anyone who has ever opened an account. Another limits the definition to people with at least one active product. A third defines a customer as anyone who has interacted with the bank within the last five years. The AI does exactly what it was designed to do—it retrieves information, summarizes it, and confidently presents an answer. Yet none of those answers is universally correct.

An experienced employee would immediately recognize the problem. Rather than accepting one of the definitions, they would ask a follow-up question: “For which business process?” Without consciously realizing it, they have performed a task that the AI cannot. They have recognized ambiguity before attempting to resolve it. For decades, organizations have relied on people to perform this invisible work instinctively. AI has not created this problem; it has simply made it impossible to ignore.

This raises a deeper question: What if one of the most important capabilities inside an organization has never appeared on an organizational chart because humans quietly performed it every day?

We Think AI Has a Data Problem

When AI systems produce inconsistent or unreliable answers, our first instinct is to blame the technology. We assume the metadata is incomplete, the data quality is poor, governance has failed, or the model lacks sufficient context. Sometimes these explanations are correct. However, imagine an organization where every dataset is documented, every column has a business definition, every table has a clearly assigned owner, and every pipeline is carefully monitored. Even in such an environment, the original question remains unresolved: Who owns Customer?

The difficulty is not identifying who owns a table or maintains a database. The difficulty lies in determining who owns the business meaning of the word itself. Different parts of the organization legitimately require different interpretations depending on the context in which the term is used. The disagreement is therefore not technical—it is organizational. Technology has merely exposed a coordination problem that has always existed beneath the surface.

Looking Back

This was far less visible before organizations became deeply digital. In smaller organizations, people sat close together, shared context naturally, and clarified misunderstandings through conversation. Knowledge lived inside relationships rather than repositories. As organizations expanded, they invested in reporting structures, standard operating procedures, enterprise systems, data warehouses, governance forums, business glossaries, metadata catalogs, and countless other mechanisms intended to improve coordination.

Each innovation reduced a particular coordination cost, yet one fundamental assumption remained remarkably consistent. Whenever definitions became ambiguous or business meaning changed, people would resolve the uncertainty through discussion. Meetings, email threads, and hallway conversations quietly repaired shared understanding whenever formal documentation fell behind reality. Because humans were remarkably adaptable, organizations never considered this activity to be infrastructure. It was simply regarded as part of normal work.

The Invisible Infrastructure

Consider the countless situations that occur inside a large enterprise every day. An analyst notices that two executive dashboards report different customer numbers. A governance lead asks whether a new regulation changes the definition of a critical business metric. An architect pauses an integration because two systems use the same term differently. A product owner explains why one business unit treats inactive customers differently from another.

None of these activities produces a new product feature. None appears on an architecture diagram or technology roadmap. None is typically measured as organizational capability. Yet if these conversations disappeared overnight, confusion would spread rapidly throughout the organization. Decisions would diverge, reports would conflict, and trust in information would gradually erode. What we have traditionally dismissed as organizational overhead may, in reality, be one of the most important forms of infrastructure an enterprise possesses.

What Existing Thinking Explains

This observation builds upon rather than replaces existing organizational theory. Ronald Coase argued that organizations exist because coordinating work within firms is often less costly than coordinating through markets. Later scholars demonstrated how routines, institutions, shared knowledge, and organizational learning enable people to collaborate effectively despite increasing complexity. Within the data profession, governance, metadata, stewardship, lineage, and quality management have all sought to improve organizational coordination by creating more consistent information.

These perspectives remain valuable, but the arrival of AI introduces a new constraint. Machines cannot participate in the informal conversations through which humans routinely resolve ambiguity. They cannot notice hesitation during a meeting, recognize that two departments use the same word differently, or interrupt a discussion to ask, “When you say customer, do you mean active customer or any customer?” The organizational flexibility that humans have always supplied cannot simply be assumed when one of the participants is a machine.

A Different Lens

Perhaps organizations have unknowingly outsourced ambiguity handling to people for decades. Whenever business definitions conflicted, humans negotiated meaning. Whenever ownership became unclear, humans escalated decisions to colleagues with more context. Whenever business processes evolved, people adapted their understanding without requiring every change to be formally documented. Their ability to absorb ambiguity became part of the organization’s invisible operating system.

Because people performed this work naturally, organizations rarely attempted to measure it. AI changes that assumption. Machines cannot compensate for ambiguity unless organizations first make their knowledge explicit and operational. What appears to be a limitation of AI may therefore be revealing a limitation of the organization itself—one that has remained hidden because humans quietly compensated for it every day.

Why Data Management Suddenly Matters

Viewed through this lens, the renewed importance of data management becomes easier to understand. Metadata is no longer merely documentation describing technical assets; it becomes the shared context through which organizations preserve institutional meaning. Governance is no longer simply a compliance function; it becomes the mechanism by which organizations negotiate durable agreement about business concepts. Data quality extends beyond correctness and completeness to include confidence that shared meaning survives across teams, systems, and intelligent machines.

The technology itself has not fundamentally changed the importance of these disciplines. Instead, AI has exposed how much invisible organizational work they were quietly supporting all along.

Beyond Banking

Although the examples here come from banking, the underlying pattern extends far beyond financial services. A hospital may ask whether a patient is considered discharged. Does that mean medically cleared, financially settled, or physically absent from the hospital? A manufacturer may ask whether a product is available. Does availability mean manufactured, shipped, reserved, or ready for sale? A government agency may attempt to define a household, while a retailer may debate what qualifies someone as a customer.

Humans navigate these contextual differences with remarkable ease because they possess shared experience and situational awareness. Machines require organizations to make that context explicit before they can reason reliably.

An Open Question

Perhaps the defining challenge of the AI era is not making organizations more intelligent but making them more coherent. If that is true, then one of the largest organizational costs has never appeared on any balance sheet. It is not poor data quality, weak governance, or inadequate technology. It is the continuous, invisible effort through which people translate ambiguity into shared understanding.

For decades, organizations benefited from this capability without consciously recognizing it because humans supplied it instinctively. As AI increasingly becomes another participant in organizational work, that invisible infrastructure can no longer remain invisible. The question is no longer whether AI can understand our organizations. The more important question may be whether our organizations understand themselves well enough to explain their own knowledge to anyone—or anything—that must rely upon it.

Checkout my new book here: https://ankit-rathi.github.io/store/