The number everyone uses and nobody signs
There is a scene that repeats itself in companies of different sizes. Someone puts margin by product line on the screen. Someone on the leadership team says the figure is wrong. Twenty minutes of debate follow about where it comes from. Everyone agrees to review it. The meeting moves on. And a month later the exact same thing happens with the exact same number.
The interesting part is not that the number is wrong. It is that nobody is obliged to fix it. It sits in a report prepared by finance, with data entered by sales, on prices set by the leadership team, using costs calculated by operations. Four areas touch that number. None of them answers for it.
The industry calls this data governance, which sounds like a committee and a seventy-page policy. In a mid-sized company it is far simpler and considerably more uncomfortable. It means saying out loud who answers when a number is wrong.
What it costs when a number belongs to no one
IBM published an analysis in January 2026 on the cost of poor data quality with figures worth reading slowly. More than a quarter of organizations estimate they lose over USD 5 million a year to it, and 7% go above USD 25 million. And 43% of chief operations officers rank data quality as their single most significant data priority.
But the detail that interests me most in that report is not a figure. It is how it describes one of the most common warning signs: datasets that cannot be traced back to an identifiable owner. It does not say incomplete data. It says orphaned data.
In Spain the context makes it worse. The national observatory ONTSI puts 74.2% of small and mid-sized companies at basic digital intensity or above, while only 5.6% reach high levels. Translated into what you actually see inside these businesses: most already have tools, very few have order. And a company with five systems and no owner per number produces contradictions faster than it resolves them.
A data committee is the elegant way for nobody to answer
When a mid-sized company decides to sort out its numbers, the usual reflex is to set up a committee. Six people are convened once a month, minutes are taken, tasks are handed out, and four months later the committee dissolves without anyone announcing it.
It does not work because a committee dilutes responsibility instead of assigning it. If margin by customer is the committee's responsibility, it is nobody's responsibility. Six people staring at a wrong number do not fix it faster than one person. They fix it slower, because each one waits for another to move first.
A number without an owner is not data. It is a rumour formatted as a table.
The owner of a number is whoever decides with it, not IT
In almost every company I walk into, the first answer to "whose number is this?" is "IT" or "whoever runs the ERP". That is the wrong answer, and it is the one that stalls the whole thing for years.
IT answers for the number existing, travelling and not getting lost. Not for it meaning the right thing. The definition of active customer, of fulfilled order, of contribution margin is not a technical decision. It is a business decision, and it has to be signed by whoever makes decisions with that number.
If the sales director pays commissions based on margin by customer, the owner of margin by customer is the sales director. Even if they cannot open the ERP. Even if they have never touched a table in their life. Owning a number does not mean maintaining the calculation. It means the definition is theirs, and if the number is wrong, the phone call goes to them.
How you put a name next to a number
This requires no budget and no new tool. It requires a spreadsheet and two hours of the leadership team. Here is how I set it up.
- List the numbers decisions are made with. Not every piece of data in the company. The ones that show up in leadership meetings and area reviews. In a mid-sized business that is usually between fifteen and thirty. If you end up with two hundred, you have listed reports, not decisions.
- Write one person's name next to each. One only. The moment "sales and finance" appears, that number is back to having no owner.
- Ask every owner for their definition in one sentence. "An active customer is one who has bought in the last twelve months" is a definition. "An active customer is one who is active" is not. This step surfaces disagreements that have been buried for years.
- Set what happens when the number is wrong. Who flags it, to whom, and how long the fix takes. Without this, the previous three steps are a nice document.
The first step alone changes conversations. When the managing director asks who signs a number and there are three seconds of silence in the room, everybody understands the problem without needing a presentation.
With AI in the mix, this stops being something you can postpone
Gartner presented its data and analytics predictions in March 2026. One says that by 2030 half of AI agent deployment failures will come from insufficient governance enforcement. Another elevates the semantic layer, which is the technical translation of "what each number means", to critical infrastructure, on a par with data platforms and cybersecurity.
The logic is straightforward. A report with a doubtful number gets looked at by five people and someone raises a hand. An agent querying that same number uses it hundreds of times a week and never raises a hand. AI does not fix orphaned data. It distributes it faster.
Before you ask a model for anything, it pays to know the real state of the house. If you are not sure, start by getting a single version of every number and by defining the indicators that genuinely support decisions.
How I see it
When I walk into a company that wants to sort out its data, I do not start with architecture or tooling. I start with the list of numbers and the names beside them. It is the least impressive part of the work and the one that meets the most resistance, because it forces people to sign for things that until now belonged to everyone and to no one. It is also the part that keeps the project from collapsing six months in. That is how I approach every project.
You do not need a committee. You need a list, one name per line, and the discipline to keep it alive when someone changes role.
If you want to try one thing this week, try this. Take the five numbers you look at most and ask who signs them. If you get five clear names, your house is in better shape than you think. If you get five silences, you know where to start, and it does not involve buying anything.