INSIGHTS

Single source of truth: what it really takes

A single source of truth is one of the most requested things in business data and the least achieved. The reason is that most teams treat it as a product to buy, not a discipline to maintain.

One trusted, luminous source everyone refers to.

“A single source of truth” is one of the most requested outcomes in business data, and one of the least achieved. Almost every organisation wants one. Very few have one. The gap between wanting and having comes down to a misunderstanding about what the thing actually is.

It is not a database

The common assumption is that a single source of truth is a system (a warehouse, a platform, a tool) that you purchase and switch on. Buy the right product, point everything at it, and the truth follows.

It doesn’t. A central system is at most a place to keep the truth; it is not the truth itself. You can consolidate every number into one warehouse and still have three definitions of “revenue”, two ways of counting an active customer, and a metric nobody can explain. The hard part was never storage. It is agreement on what the numbers mean, and the discipline to keep that agreement current. That is a process, maintained by people, not a product you install.

The work is writing down what’s true

The actual work of a single source of truth is unglamorous: deciding what each important figure means, where it comes from, and how it is calculated, and then writing that down somewhere durable and accessible.

This sounds trivial until you attempt it. Most businesses have never written down, precisely, what they sell and how they count it. The knowledge exists, but it is scattered across people’s heads, buried in the logic of a spreadsheet, or implied by a report that has gone out unchanged for years. Pulling it into the open and committing to a single agreed definition is the real project. The system that stores it is the easy part.

AI is making this unavoidable

Organisations have tolerated this ambiguity for decades because people are good at working around it. Someone always remembers what the odd column means, or knows which of the two figures to trust. AI removes that tolerance.

An automated process will not accept “that’s how it’s always been done.” Hand it a report built on an undocumented field and it has no basis to proceed; it needs to know what the field means and where it came from. In forcing that question, adopting AI tends to deliver a benefit that has nothing to do with AI: for the first time, the business writes down how it actually operates. (This is the same realisation behind getting your data AI-ready: the documentation is most of the value.)

The buzzwords are the same idea

Much of the current technical vocabulary around AI and data (“context engineering”, “retrieval-augmented generation”, and the rest) sounds like a new discipline. Underneath, most of it restates one old requirement: write down what is true, and put it somewhere the machine can reach it. The terminology is new; the obligation is not.

Knowledge that survives

There is a second reason the writing-down matters. Knowledge held only inside a system disappears when that system is replaced. Knowledge held only inside a person leaves when that person does. Knowledge written down, in a defined and accessible form, survives both. A single source of truth, done properly, is also how an organisation stops losing what it knows every time a tool or an employee changes.

How to actually get one

In practice, building a single source of truth looks less like a purchase and more like a habit:

  • Pick the figures that matter. Start with the handful of numbers the business actually makes decisions on, not everything at once.
  • Define each one exactly: what it means, its source, and how it is calculated, agreed by the people who use it.
  • Write those definitions down somewhere they can be reached, by people and, increasingly, by machines.
  • Make one place authoritative, so that when systems disagree it is clear which one wins.
  • Review the definitions as the business changes, or they rot back into ambiguity.

The storage layer, whether a warehouse, a model, or a data warehouse or a data lake, is a decision worth getting right, but it comes after this, not instead of it. If your numbers don’t agree and no one can say which to trust, that is the problem our data solutions practice exists to fix.

Let's figure it out together.

No pitch deck, no pressure. We'll talk through your situation, share an honest perspective, and tell you if we're not the right fit.