Accuracy shows up in metrics. Trust shows up in behaviour.

I’ve lost count of how many times someone has told me the data is wrong — not because it was incorrect, but because it wasn’t trusted.

The number was too high, too low, or simply not what last month looked like. The conclusion followed quickly: the data must be wrong.

This shows up more often than we expect. A forecast doesn’t align with last year. A model output contradicts a familiar report. A new metric breaks a pattern people recognise. And the conversation shifts away from the result towards the reliability of the data itself.

At first, this looks like a data quality problem. Sometimes it is. But often, something else is happening.

Decisions don’t run on accuracy. They run on trust.

Trust doesn’t come directly from accuracy. It comes from alignment — how well something fits the way the system is already understood. Familiar reports, stable numbers, and intuitions built from experience become anchors — not because they are perfect, but because they are trusted.

New data doesn’t arrive into a vacuum. It arrives into that system. If it doesn’t fit, it creates friction.

This is where a lot of technical work quietly breaks down. Improving the data — cleaner pipelines, better features, stronger models — doesn’t guarantee that the output will be used. It only strengthens the case.

Accuracy shows up in metrics. Trust shows up in behaviour. We tend to optimise for the first, and then question the result when it doesn’t translate.

You can point to validation results, backtests, and error metrics. But if the output disrupts trust, it struggles to take hold.

The reverse is also true. When data aligns with what is already trusted, it often moves without resistance — accepted quickly, shared widely, and used to reinforce the existing view. The same scrutiny is rarely applied in both directions.

This creates an asymmetry. Contradicting data is interrogated because it disrupts trust. Confirming data is absorbed because it reinforces it. Over time, the system doesn’t just resist change; it stabilises around what it already believes.

The fastest analysis is often the least useful — the one that confirms what was already assumed. A more valuable question is asked less often: what would need to be true for this to be wrong?

This isn’t irrational. We rely on compressed representations of the world — patterns, heuristics, mental models — because the underlying system is too complex to hold in full. Those representations are hard-earned. They don’t shift because of a single new number.

If we want better decisions, this matters. Building better models is only part of the problem. The harder part is understanding what those models are entering into.

Most of the time, the data isn’t wrong. It’s arriving into a system where trust has already been allocated — and the new information isn’t part of it