Show me a company’s data quality and structure, and I’ll have a much better idea of what its AI can actually do.

My rough rule: enterprise AI is probably 80% data and 20% AI. We spend a lot of attention on models and agents. The less visible constraint is whether they have useful context.

An organization can have years of documents and still struggle to answer a basic question reliably. Having information somewhere is different from knowing which information to trust.

Context has a job to do

Consider a hypothetical capital project. There’s a schedule in one system, a cost forecast in another, and a decision recorded in an email. An agent can retrieve all three and still get the answer wrong if the schedule is stale, the forecast uses a different project identifier, or the decision has been superseded.

A bigger model doesn’t settle those disagreements for us. Someone has to establish the relationships and decide which source owns which fact.

Before adding another AI tool, I want to understand a few practical things. Where does the information come from? How current is it? What does it refer to? Who owns it? And does the person asking the question have permission to see it?

Those questions sound ordinary because they are. They’re also the difference between a convincing answer and an answer a team can safely use.

Start with a bounded problem

I wouldn’t wait for every piece of company data to be perfect. That can turn into a program with no useful stopping point. Pick a workflow and identify the context it actually needs. Clean up that slice, connect it, and see whether the result holds up in real use.

For example, an assistant answering questions about a project should be able to point back to the record behind its answer. If the record is missing or outdated, that should be visible. The gap is useful information, even when it means the agent can’t finish the task.

As the workflow grows, the data responsibilities need to grow with it. Access, ownership, and updates are ongoing work. They aren’t a one-time cleanup before the interesting AI part starts.

I think there’s a lot of value in the companies and teams that make this foundation easier to build and maintain. The impressive demo gets attention. Clean, connected, trusted context is what gives it a chance of becoming useful every day.