The most expensive meeting I have ever sat in did not decide anything. Two teams arrived with the same question and two different answers, and both answers were built on real data, pulled honestly from real systems. We spent the hour not on the decision but on whose number was correct. By the end nobody trusted either figure, the decision slid to the next meeting, and the actual cost was a month of drift while a business waited for its own organization to agree with itself.
If you run technology across a large company, you know this meeting. It repeats in a hundred variations, and every one of them is a symptom of the same missing thing. There is no single source of truth, so the organization litigates reality instead of acting on it. Fixing that is some of the least glamorous and highest-leverage work I do, and almost none of it is technical.
The scale of the problem is easy to underestimate, and the recent numbers are worse than most leaders assume. A 2026 Financial Times study found that 74 percent of business leaders had held back a decision because they did not know what data to trust (InPublishing on the FT study). A PwC report the same year put the share of leaders who do not fully trust their own company’s data at 78 percent (The Narrative Post on PwC). And a May 2026 study of senior finance and IT executives found that nearly half, 47 percent, admitted making a material business decision on data they knew was inaccurate, incomplete or outdated (PR Newswire). Most of that damage does not arrive as one dramatic error. It arrives as the slow tax of meetings like mine, where confident people burn hours reconciling figures instead of deciding anything.
Why the numbers disagree in the first place
The instinct is to blame the systems. Bad data, broken integrations, a reporting tool that lies. Occasionally that is the cause. Far more often the systems are fine and the definitions are not.
Two teams count revenue and get different answers because one recognizes it at contract signature and the other at delivery. Two dashboards show different headcounts because one includes contractors and one does not. Neither is wrong. They are answering slightly different questions with the same word, and because the word is the same, everyone assumes the answers should match. When they do not, the argument becomes emotional, because each side can see their own numbers are correct and concludes the other side must be careless or worse.
That is the trap. The disagreement looks like a data quality problem, so people throw engineering at it, cleaner pipelines and better tooling. The engineering does not resolve it, because the real gap is that the company never agreed what the words mean. You cannot build a single source of truth on top of a hundred private definitions. You have to fix the definitions first, and that is a governance job, not a data job.
The single source of truth is a decision, not a database
A single source of truth is best understood as an agreement about what the numbers mean and who gets to define them, long before it is a place where the correct numbers live. The technology comes after, and it is the easy part.
So I stopped starting these programs with a data platform and started them with a definition. For each metric that matters, three questions have to be answered and written down. What exactly does this term mean, including every edge case people quietly disagree about. Who owns the definition and has the authority to settle a dispute about it. And where is the one system of record that produces it, so every report traces back to the same origin. This is the same conclusion the serious data-governance practitioners keep reaching, that a single source of truth begins with metric definitions and named owners and only later becomes infrastructure.
Answer those three for the numbers that actually drive decisions, and most of the disagreements evaporate. Not because the data got cleaner, but because everyone is finally counting the same thing and knows who to ask when they are unsure. The platform then becomes what it should always have been, plumbing that carries an agreed truth rather than a stage on which people argue about competing ones.
Trying to define everything is how these programs die. There are thousands of metrics in a large company and most of them do not matter enough to fight over. So I am ruthless about scope.
I start from the decisions, not the data. Which handful of numbers do senior people actually make choices on. Revenue, margin, headcount, utilization, cash. That short list is where a wrong or contested figure does real damage, and it is where I concentrate. The long tail of metrics can stay imperfect for now, because nobody is betting the business on them.
For each one on the short list, I name a single owner. Not a committee, a person. The committee model is how you get a definition nobody will defend and everybody can reopen. One accountable owner per metric, with the authority to say this is what it means and this is where it comes from, and the obligation to keep it current. When two teams disagree, they do not escalate to me, they go to the owner, and the owner decides. That single move removes an astonishing amount of friction.
Then I make the agreed number the easy one to use. This is the part people skip. If the official figure is buried in a slow, ugly report and the unofficial one is a quick export somebody built themselves, the unofficial one wins every time, and you are back to competing truths. So the sanctioned number has to be the most convenient number in the building. If the compliant path is also the fast path, compliance mostly takes care of itself. I learned that lesson on operating-model design and it applies here word for word.
This work is slow, and it is political, and there is no tool you can buy that does it for you. Vendors will happily sell you a data catalog, a governance platform, a semantic layer, and those things help once the hard part is done. None of them will tell your two most senior teams which of them owns the definition of revenue. That is a conversation a human leader has to hold, and it involves telling capable people that their version is no longer the official one. It makes you briefly unpopular, and that is precisely why it does not happen on its own and why it lands on someone like me.
I have also learned to resist the temptation to solve it perfectly. A single source of truth is never finished, because the business keeps changing and new metrics keep mattering. The goal is not a permanent state of perfect agreement. The goal is a living discipline, a small set of numbers everyone trusts, clear owners, and a fast way to settle the next disagreement when it comes. Chase perfection and you will spend three years building a cathedral nobody uses. Build the discipline and you get value in the first quarter.
When it works, you feel it in the meetings. The argument about whose number is right simply stops, and the same hour goes to the decision the number was supposed to inform. People stop hedging every figure with a caveat about which system it came from. Speed goes up, not because anything moved faster technically, but because the organization stopped arguing with itself.
That is the real prize. A single source of truth is not about tidiness or elegant architecture. It is about turning a company that debates reality into one that acts on it. The systems are the easy tenth of that job. The other nine tenths is getting people to agree what the words mean and to trust the answer when they see it, and that is leadership work wearing a data costume.
