Most organisations seem to be racing to deploy AI right now. But can their data actually support it?

With businesses shifting from experimentation to enterprise-wide deployment, data readiness is becoming one of the biggest factors in whether an AI project succeeds or stalls.

Before you get into picking models or briefing a team to build a copilot, it’s worth answering five fairly unglamorous questions about the data you already have.

1. Do you know where your data lives?

If the information that matters is scattered across email inboxes, SharePoint folders, a CRM, and a handful of legacy systems nobody’s fully documented, an AI tool isn’t going to piece that information together for you. The first step is simply knowing where that information is.

2. Can you trust the data?

Duplicate records, gaps, fields that haven’t been updated in years. If that's the information your AI is working from, don't be surprised when the outputs sound convincing but are wrong. It’s worth checking accuracy, consistency, and completeness properly.

3. Is your data organised?

AI tools can handle unstructured content reasonably well, but they still need some context to work with. That means having metadata in place, clarity on who owns what, and some sense of how different pieces of information relate to each other. Without that context, the model has to make assumptions.

4. Is governance keeping pace?

Who owns this data? Who’s allowed to access it? And just as importantly, what should never be fed into a model in the first place? This is the difference between an AI deployment that’s safe and one that quietly leaks something it shouldn’t.

5. Are you solving a business problem?

The projects that tend to work are the ones that start with a specific outcome in mind – time saved, a decision made faster, an error avoided – rather than a demo built to show off the technology. If you can’t point to the business problem, that’s worth pausing on before you go further.

 

What AI is really exposing

Most organisations don’t really have a model problem. They have a data problem, and AI just makes it visible: the messy records, the inconsistent processes, the reports nobody’s owned in years were always there. AI just doesn’t let you ignore them anymore.

None of this means chasing perfect data, which doesn’t exist anyway. It means finding the gaps that matter most, fixing those first, and then giving whatever system you deploy something reliable to work from.

Getting your data in order should be the first step, not something to circle back to later. And whatever AI project follows has a much better shot at delivering something real.