The unglamorous work that makes AI possible

By Erika Fille T. Legara

THERE IS a category of organizational investment that rarely generates enthusiasm in the boardroom. It does not carry the promise of artificial intelligence (AI), the urgency of cybersecurity, or the narrative appeal of digital transformation. It does not make for compelling slides. And yet, without it, virtually every ambitious technology initiative an organization pursues will underperform, produce unreliable outputs, or fall short of its promises.

That investment is data governance. And the case for taking it seriously at the board level has never been stronger.

WHEN THE CUSTOMER KNOWS MORE THAN THE COMPANY
Consider a scenario that will feel familiar to many. A customer updates their address with one business unit of a financial institution. Months later, a statement from a different unit of the same institution arrives at the old address. A card renewal follows it there. From the customer’s vantage point, this is baffling; they are dealing with one company, the same name, the same logo.

From the institution’s internal perspective, however, these are separate business lines with separate systems, separate data repositories, and no shared mechanism for propagating a simple change. The customer knows something the company does not: that they have moved.

What this reveals is that the organization has no reliable single view of its customer. Data updates are siloed, and the various parts of the enterprise are operating on different versions of the same reality. For a board that has approved investments in customer experience, digital channels, and personalization, this is precisely the kind of gap those investments were meant to close, and precisely what data governance is meant to prevent.

THE 360° VIEW IS A DATA GOVERNANCE PROBLEM
Many organizations today aspire to what practitioners call a “360° view” of their customers. The idea is intuitive. Aggregate all available data about a customer, including transaction history, service interactions, product holdings, preferences, and risk profile, into a single coherent picture that enables better decisions, better service, and better outcomes. It is an aspiration that boards readily endorse, and the business case is well understood.

What is less often discussed in the boardroom is the operational reality of achieving it, and how frequently organizations discover they are further from it than they assumed.

A bank may have a customer who holds a savings account, a credit card, and a home loan, three products potentially managed across three different systems with three different teams. When that customer calls to report a change in income, the information may be updated in one system but not the others. The loan officer reviewing a restructuring request may be working from a different picture of the customer than the one the credit card team holds.

In healthcare, the gap can be even more consequential. A pharmaceutical company that manufactures medicines, runs patient support programs, and engages directly with physicians may be touching the same patient through multiple channels. Yet, each of these functions often operates with its own data infrastructure. The patient enrolled in a chronic disease management program may be invisible to the pharmacovigilance team tracking adverse events. A physician who is both a prescriber and a program participant may appear in multiple databases with no common identifier linking them. The organization’s ability to understand outcomes, adjust programs, or respond to safety signals depends entirely on whether these data streams can be connected, and that connection requires governance, not just technology.

Energy companies offer a version of this problem that extends beyond customers entirely. A power distributor managing transmission infrastructure generates data from sensor readings, maintenance logs, outage records, and real-time telemetry across its grid assets. These streams are typically produced by different systems, managed by different teams, and stored in different formats. When a transformer fails, the data needed to understand why, including its maintenance history, prior anomaly readings, and load patterns, may sit across multiple systems with no straightforward way to bring it together. Predictive maintenance depends entirely on the ability to integrate and trust these streams; a model trained on incomplete or misaligned data will produce predictions that engineers cannot rely on, and in infrastructure management, that carries consequences well beyond inconvenience. The governance challenge here is not always about knowing the customer. Sometimes, it is about knowing the grid.

In each of these cases, the ambition is real, and the investments are substantial; what is consistently underestimated is the governance layer that makes the data trustworthy enough to actually use: common identifiers, shared standards, defined ownership, and the organizational discipline to maintain all of it over time. Without that, the 360° view remains an aspiration rather than a capability.

BEYOND ‘GARBAGE IN, GARBAGE OUT’
The computing maxim “Garbage In, Garbage Out” has been around for decades, and it remains directionally correct. Feed a model bad data, and it will produce bad outputs. But for a board seeking to exercise meaningful oversight, the phrase is too blunt an instrument. It tells you that data quality matters without telling you what to ask, what to monitor, or what governance actually entails.

A more useful framing is to think of data quality as having several distinct dimensions, each with different risks if neglected.

Accuracy refers to whether data reflects reality. Is the address on file the customer’s actual address? Completeness asks whether critical fields are populated. Are there customers in the system with no risk classification, no contact information, no transaction history? Consistency concerns whether the same entity is represented the same way across systems. Does the customer appearing on three different platforms share a common identifier, or are they three separate records that no one has reconciled? Timeliness captures whether data is current enough for the decisions it informs. Is the credit score driving a lending decision based on information from last month, or last year?

Each dimension has a corresponding governance mechanism: data standards, validation rules, master data management, refresh cycles, and ownership accountability. Each failure mode also carries a corresponding business consequence: mispriced risk, failed compliance checks, poor customer service, and increasingly, AI models that produce confident predictions from flawed inputs.

This last point deserves emphasis. AI amplifies data failures at scale. A human analyst reviewing a spreadsheet might notice an anomaly; a credit officer with experience might sense that something is off. An AI model trained on and fed by compromised data will process it at volume without hesitation, embedding the error into thousands of decisions before anyone identifies the pattern. The same scale that makes AI valuable is what makes bad data so consequential.

WHAT DATA GOVERNANCE INVESTMENT ACTUALLY MEANS
Boards that approve data governance investments are sometimes presented with proposals for data platforms, cloud migration, or master data management systems, all of which are necessary, but which represent only the technical layer of a much broader requirement. The platforms matter, but the governance challenge is as much organizational as it is architectural.

Effective data governance requires investment across at least three areas.

The first is institutional, defining who owns which data, who is accountable for its quality, and who has authority to establish and enforce standards. This sounds straightforward, but is often contentious in practice. Business units that have long managed their own data with their own definitions resist centralized standards. Functions that have built workflows around local data conventions resist harmonization. Without clear ownership and accountability, governance frameworks tend to become committees without authority.

The second is technical, encompassing the systems, tools, and architecture needed to implement governance at scale. This includes data catalogues that document what data exists and where, lineage tools that trace how data moves and transforms, quality monitoring systems that flag anomalies, and integration layers that allow data to flow across business units without being corrupted in transit.

The third, and perhaps most underestimated, is cultural. Data governance fails not because organizations lack policies, but because those policies are not followed, not enforced, and not embedded in day-to-day workflows. People enter incomplete records because they are measured on speed, not accuracy. Systems are integrated hastily because project timelines do not allow for proper data mapping. Exceptions are granted because enforcing standards is inconvenient. Boards should ask not only whether governance frameworks exist, but whether the organization’s culture treats data as an asset that requires active stewardship.

WHAT BOARDS SHOULD ASK
Asking whether a data governance policy exists is a reasonable starting point, but it tells a board relatively little. More revealing is whether accountability is real, whether standards are enforced, and whether quality is actually being measured. Boards should expect management to be able to answer several basic questions. What is the organization’s data quality posture across its most material data assets, those that drive credit decisions, risk models, regulatory reporting, and customer interactions? Where are the known gaps, and what is being done to close them? How are data quality issues identified and escalated? Who is accountable when a data failure produces a material business or compliance consequence?

Equally important, boards should look for the connection between data governance investments and the strategic ambitions they have approved. If the board has endorsed an AI strategy, it should expect management to explain how data infrastructure supports that strategy, not in aspirational terms, but in concrete ones: what data is available, how trustworthy it is, and what gaps remain between the current state and what the strategy requires.

The address that was never updated is a small thing in isolation. But as a signal, it points to an organization that has not yet treated data as a governed asset, and that has not yet connected data quality to the customer experiences and analytical capabilities it is trying to build. In a world where AI is being embedded into consequential decisions at scale, the cost of that gap is rising. Boards that understand this, and ask accordingly, will be better positioned to distinguish organizations that are genuinely AI-ready from those that are merely AI-aspiring.

Dr. Erika Fille T. Legara, FICD is a physicist, educator, and data science and AI practitioner working across government, academia, and industry. She is the inaugural managing director and chief AI and data officer of the Philippine Center for AI Research, and an associate professor and Aboitiz chair in Data Science at the Asian Institute of Management. She serves on corporate boards, is a fellow of the Institute of Corporate Directors, an IAPP Certified AI Governance Professional, and a co-founder of CorteX Innovations, Corp., a technology company.