Can You Trust Your Data? Why the Question Hasn't Changed

The question that outlived four platform generations

We have watched reporting move from data warehouses to self-service BI, from on-premise cubes to cloud platforms, from dashboards to AI assistants that answer in prose. Every generation arrived promising that this time, the business would finally see the truth.

Every generation then rediscovered the same constraint. The value of any reporting solution is not in the tools or the graphics. It is in the trust users place in the data underneath. Roll out the most beautiful dashboard in the company and it is worthless the day a regional director says, quietly, that the numbers do not match his spreadsheet.

In two decades of migration and analytics delivery we have heard the same sentence in almost every engagement, in different accents and industries:

“We knew our data might be bad, but we didn't realise how bad it really was.”

What trusted actually means

Trust is not a feeling about a dashboard. It decomposes into questions that can be answered with evidence. Is there an agreed source for this number, or three competing ones? Is the definition shared, so that margin means the same thing in every country? Is the quality measured against rules the business wrote, not rules a tool shipped with? Can the lineage be walked from the figure on the screen back to the transaction that produced it?

When any of those answers is no, users do what users have always done: they build their own extract, keep their own spreadsheet, and trust that instead. Data silos are not an IT failure. They are a rational response to unmeasured quality.

The fix has never been technical alone. Agreed sources, shared definitions, business-owned rules and visible measurement: this is governance, and the best moment to start it is any project that already forces the business to look hard at its data. A migration is the prime example. There is no better time to begin than when the data is already on the operating table.

Why AI raises the stakes

A human analyst squints at a suspicious number. An AI system does not. Retrieval systems and assistants consume enterprise data at face value and repeat it back fluently, with confidence, to people who have no reason to doubt a well-written answer.

That changes the economics of bad data. A wrong figure in a report misleads the person who reads it. A wrong figure in the data an AI system draws on misleads everyone who asks, in every phrasing, indefinitely. Agents that act on enterprise data extend the same problem from bad answers to bad actions.

So the old question has a sharper modern form: not only can the business trust the data, but can the business afford what its AI systems will do with data it cannot trust? The organisations investing in data quality and governance now are not being cautious. They are reading the direction of travel correctly.

What we do about it

Our approach has been consistent across every platform generation. Establish the trusted source and retire the competitors. Write the quality rules in business language, with named owners. Measure continuously, not annually, and put the measurement where management can see it. Fix in the source, so the improvement holds.

The tooling for this is better than it has ever been. Our own platform profiles data, executes business-defined rules, detects duplicates and gives executives a live quality score. But the tooling only automates the discipline. It cannot substitute for it. That was true in the data warehouse era, and it is just as true now that the consumer of the data might be a machine.