Data governance fails more often as a political problem than a technical one. Without someone with real authority, no amount of tooling resolves conflicting data definitions.
Every large organization I have read about has, at some point, tried to answer the question: what is our revenue this quarter? The answer almost always arrives as multiple numbers, one from the finance system, a different one from the sales CRM, a third from the customer success platform, and a fourth from a business analyst who built a custom SQL query two years ago and stopped updating it. Each number is defensible by the team that produced it. None of them agree. And in the meeting where all four numbers appear on the same slide, somebody senior has to decide which one counts. That decision is not a technical problem. It is a governance problem.
Data governance is the set of policies, processes, and organizational structures that determine how data is defined, collected, maintained, and used across an organization. DAMA International, the professional organization for data management practitioners, has published the DMBOK (Data Management Body of Knowledge) as a reference framework covering areas including data quality, data architecture, metadata management, and data security. The framework is comprehensive and, in my experience reading about it, rarely the reason governance fails. The reason governance fails is that enforcing the framework requires authority that most data teams do not have.
The political structure of data in large organizations is relatively consistent. Business units treat their data as their data. Finance owns its revenue recognition methodology. Sales owns its pipeline definitions. Marketing owns its attribution model. These are not arbitrary territorial instincts. Each unit built its definitions to answer questions that were important to it, in contexts that the other units did not fully understand. Finance's revenue number is the auditable one. Sales' pipeline number is the one that gets shared with investors on roadshows. Marketing's attribution model is the one the marketing team believes in, because it is the only one that makes their campaigns look effective.
When these numbers conflict, as they reliably do, resolving the conflict requires someone who has authority over all three units. In most organizations, that is the CFO or the CEO. If those people are not actively engaged in data governance, the conflict stays unresolved. Teams learn to use whichever number suits the audience they are talking to. Analysts learn to hedge all data questions with caveats about source systems. Executives learn to discount any data point that challenges a decision they have already made. The organization operates on data theater instead of data governance.
The Chief Data Officer role was invented partly to address this problem. By some accounts, the CDO role started appearing in major companies in the 2000s and has grown considerably since, though I would hedge specific headcount figures from industry reports. The theory was that a dedicated executive with data in the title would have the authority and mandate to enforce standards. In practice, CDOs who lack budget authority, who cannot hire and fire in the data function across business units, and who do not have direct C-suite access tend to become expensive data quality advocates rather than governance enforcers. They can write policies. They cannot make the VP of Sales change how they define a qualified lead. Only the CEO can do that, and the CEO is usually focused on other things.
The "single source of truth" language is instructive here. Every few years, an organization announces an initiative to create a single source of truth for some important data domain. Customer data, product data, financial data. The announcement tends to arrive with a new platform, a new data warehouse, or a new master data management system. The data engineers work hard. The platform gets built. Then the business units continue using their existing systems, their existing definitions, and their existing reports. The single source of truth is populated but not used, because using it would mean giving up the definitions that each unit built its workflows around. Nobody with authority forces the transition. The platform sits there as a theoretically correct but practically irrelevant system. I have seen this described in so many forms that I think of it as one of the standard failure modes of enterprise data initiatives.
The technical infrastructure for data governance has improved considerably. Modern data catalogs can track lineage, define glossaries, and alert on quality violations. Data quality tools can score data against defined rules automatically. Metadata management platforms can make the ownership of every dataset explicit. But a data catalog that nobody refers to does not resolve a dispute about revenue definitions. A quality score that nobody acts on does not clean the data. The tooling can make governance easier once the governance decisions are made. It cannot make the governance decisions.
I wrote about a related pattern in why knowledge management systems usually fail. The knowledge management literature found that systems for capturing and sharing knowledge fail when the organizational incentives and power structures do not support knowledge sharing. Data governance has the same problem. The technical infrastructure for sharing and standardizing data is solvable. The organizational will to actually standardize, to accept that Finance's revenue number is the authoritative one and that Sales will report based on it even when they disagree, requires leadership that cannot be delegated to a CDO who lacks the authority to enforce it.
The organizations where I see data governance actually working have a few things in common. First, a business problem so visible and painful that senior leadership cannot ignore it. A regulatory requirement, a public embarrassment from conflicting numbers, a merger that requires integrating two datasets that disagree about basic facts. Something that forced the CFO or CEO into the room. Second, an executive who was willing to make the call about which definition wins and then enforce it, which means telling one business unit that its historical data practices are now wrong and will change. Third, a CDO or equivalent who had the authority to back that call with structural changes to reporting lines and systems.
That combination is not common. When it is absent, data governance produces policies that nobody enforces, platforms that nobody uses, and annual reports from the CDO function about how much progress is being made on data quality while the revenue number still comes in four versions every quarter.
About the author
Share
More notes
Related notes