Data governance is the set of agreements about who defines each data item, who answers for its quality and who may use it. It is not a tool and not a committee: it is the answer to a concrete question — whose figure is this? — which in most organisations has no owner at all.
Below: why it fails without owners, the decisions that have to be made, the committee that works and the one that dies, how to introduce it without stalling, and what to expect in the first year.
The scene repeats in committees across the region: two teams present the same metric with different figures and the meeting is spent establishing which is right. Nobody is lying; each calculated with a different definition and both are defensible.
That is the problem data governance solves, and it is why it is rarely solved by buying software. What is missing is not a tool: it is an agreement about which of the two figures is the one that counts, and who gets to say so.
Why it fails without named owners
A data item without an owner degrades through an accumulation of small decisions. Somebody adds a field for a one-off case, somebody else changes a filter criterion, a third copies the report and adjusts it. None of those decisions is wrong on its own and the aggregate ends up inconsistent.
When there is a named owner, those decisions pass through a single point and get recorded. That owner does not need to be technical: what is required is somebody from the business who can say what the data means and what it does not.
The decisions that have to be made
What each term means. "Active customer" seems obvious until somebody asks whether it includes a buyer from eleven months ago. A written, accepted definition is worth more than any dashboard.
Who answers for each domain. Sales, receivables, inventory. One name per domain, not a department. Departments do not answer questions.
What level of quality is required. Not every data item deserves the same effort. Demanding perfection everywhere guarantees achieving it nowhere.
Who may see what. Access is part of governance, and in Colombia and Mexico that decision is conditioned by data protection law rather than by internal preference alone.
The committee that dies, and the one that works
The most common form of data governance is a monthly committee reviewing initiatives. It almost always dies the same way: it becomes a reporting forum where nothing is decided, and people stop attending without anybody declaring it over.
The one that survives has three characteristics. It meets to settle concrete disputes — this definition against that one — rather than to review progress. Its decisions are written where anybody can find them. And it has the authority to close an argument, which means somebody with organisational weight stands behind it.
Without that third condition governance is a recommendation, and recommendations lose to each team's deadlines.
How to introduce it without stalling the operation
The most common mistake is starting with the full catalogue. Documenting every data item in the organisation before using any of it produces an out-of-date inventory and zero behaviour change.
The sequence that works runs the other way: take one business decision currently made blind, govern only the data that decision needs, and leave the agreement written. It is a small scope with a visible result, and it creates the precedent that makes the next one possible.
From there governance grows on demand rather than as a project, which is the only way it tends to survive a change in priorities.
The two roles that are always confused
Governance conversations stall on a distinction worth making early: the person who owns the meaning of a data item is rarely the person who maintains it.
The owner is from the business and answers questions of definition: what counts as a sale, when a customer stops being active, which exclusions apply. The steward is usually technical and answers questions of state: where it comes from, how often it refreshes, what broke last night.
Collapsing the two into one role is the most common way governance quietly fails. Give it to the technical side and definitions get decided by whoever writes the query, which is how the organisation ends up with three defensible versions of the same metric. Give it to the business side alone and nobody can answer why the figure was stale on Tuesday.
Both names should exist for every domain that matters, and both should be findable by anybody who has a question. A governance model where you have to ask around to discover who to ask has not started yet.
What to expect in the first year
Not a complete catalogue. What is reasonable is two or three domains with owners, written definitions for the metrics that reach committee, and a clear rule about who authorises a change.
The sign that it is working is not the volume of documentation: it is that committee discussions stop being about whether the figure is right and start being about what to do with it.
A second sign appears slightly later and is worth watching for, because it is easy to mistake for a problem: people start bringing disputes to the owner instead of resolving them privately in their own reports. That looks like more friction and is the opposite — it means the disagreements that were always there have become visible and now get settled once rather than silently and differently by each team.
What it costs when nobody does it
The visible consequence is committee time spent reconciling figures, and it is the least serious one. The serious consequence is that the organisation starts deciding by instinct, because the data stopped being an authority: when every figure is arguable, the better arguer wins.
The second consequence is duplicated work. Each team builds its own version of the truth rather than depend on a number it does not trust, and the organisation ends up paying three times for the same analysis while believing it has one reporting function.
Neither appears in a budget line, which is why a data governance initiative competes at a permanent disadvantage against work whose benefit can be invoiced. Naming both costs explicitly is usually what gets it funded.
Where it sits in the analytics capability
Governance is the base everything else rests on. A dashboard built on data without owners produces fast, badly founded decisions, which is worse than having no dashboard.
So it pays to measure the starting point before investing. A data and analytics maturity assessment walks the full path of the data — origin, quality, governance, model, consumption — and says which stretch holds the weakness, which is almost never where it is being looked for. From there comes the analytics plan that actually holds.
Frequently asked questions
What is data governance?
The set of agreements about who defines each data item, who answers for its quality and who may use it. It is an organisational decision before a technological one: what is missing when two teams bring different figures is an agreement, not a tool.
Do I need to buy a tool to start?
No. The first definitions and owners can be documented in any shared medium. A tool helps once there are agreements to administer; buying it first usually produces an empty catalogue.
Who should own a data domain?
Somebody from the business who can say what the data means and what it does not, named individually rather than as a department. Departments do not answer questions; people do.
Why do governance committees fail?
Because they become reporting forums where nothing is decided. The ones that survive meet to settle concrete disputes, write their decisions where anybody can find them, and have the authority to close an argument.
Where do you start?
With one business decision currently made without reliable data. Govern only what that decision needs and write the agreement down. Starting with the full catalogue produces an out-of-date inventory and no behaviour change.
What does success look like after a year?
Two or three domains with owners, written definitions for the metrics that reach committee, and a rule about who authorises changes. The real sign is that committees stop arguing about whether a figure is right.
