Being data-informed, not data-driven

9 July 2026

We're a data-driven organisation. It's said with pride, and the intention is good: decisions based on evidence instead of the loudest opinion in the room. Who could argue?

But listen to the phrase literally. Driven, by data. The data steers and you're the passenger. And data, for all its virtues, has never met your customers, doesn't know a redesign is coming, and cannot tell the difference between a genuine problem and a marketing campaign that changed who's visiting.

The ambition worth having is slightly different. Be data-informed instead. Evidence on the table every time, but context and judgment in the chair. That's the difference between using numbers and obeying them, and it matters once you start monitoring your website seriously.

What data is brilliant at

Measurement does things human attention simply can't:

• It notices. No one re-checks four hundred article pages after a template change. An automated test does, every time, without fatigue or optimism.
• It removes flattery. Everyone in the building experiences the site on fast hardware and a good connection. The data speaks for the visitors who don't.
• It settles taste arguments. The page feels fine versus p75 LCP is 4.1 seconds is not a fair fight, and it shouldn't be.
• It gives early warning. A score that slips this week is a small fix. The same slip discovered in next year's audit is a project.

What follows argues with none of that. Measure continuously, trust the measurements. The question is what happens after the number arrives.

Where driven goes wrong

A number arrives without its story. Accessibility findings doubled on Tuesday. Data-driven says regression, escalate. The context a human adds is that we migrated 200 archived pages onto the new site on Tuesday, so the findings are old problems finally being seen, which is the system working. Same number, opposite meaning. Every metric needs someone who knows what the organisation did that week.

The metric becomes the target. There's an old observation, Goodhart's law, that when a measure becomes the target it stops being a good measure. Chasing a score can genuinely improve a site, or it can teach a team to satisfy the test. Think of technically passing alt texts that describe nothing, or contrast fixed to the exact decimal that passes while remaining unpleasant to read. The score was always a proxy. The target is people, the visitors who can find, read and use what they came for.

Whatever is measurable steers. Speed produces numbers by the thousand; does the checkout make anxious people trust us? produces none. A strictly data-driven roadmap drifts toward the quantifiable, not the important. Sometimes the right call is the one no dashboard voted for.

What data-informed looks like in practice

This needs no workshop. It comes down to a few habits.

• Annotate your data with what you did. Keep a simple log of releases, campaigns and migrations next to your test history. Half of all mysterious dips stop being mysterious.
• Ask what would change our mind? before acting on a number. If the answer is nothing, you're not being data-driven, you're decorating a decision already made. That can be fine, but say so.
• Pair every score with a person. For each metric you track, be able to name who feels it. LCP is the visitor on the train, and the missing label is the screen-reader user at the checkout. If no one can say who a number represents, question why it's on the dashboard.
• Let humans overrule, in writing. Sometimes you'll ship something the numbers dislike, for reasons the numbers can't see. Good. Write the reason down, so future-you can check whether the judgment held up.

Owning the judgment is, at bottom, the same principle as owning your data. The measurement should be transparent, the data should be yours, and the conclusions should be drawn by you rather than handed down by a dashboard, a vendor, or a model you can't inspect.

Tools that respect that boundary make you sharper. Tools that blur it make you obedient. Collect the best evidence you can, then have the nerve to think about it yourself.

We build the evidence half, transparent and continuous, and you can inspect and export all of it. The judgment half stays yours. See the plans.