Data & Infrastructure Readiness

Most leaders think "data readiness for AI" means having a clean, centralized data warehouse. That's a useful long-term goal, but it's not where the actual work starts.

The harder question is: where does your important data actually live right now? Because in most organizations, the answer is "scattered." Critical knowledge sits in personal drives, individual laptops, Slack DMs, and someone's Notion sidebar that nobody else has access to. The person who owns that data can use AI tools on it. The organization can't. And once that person leaves, or moves teams, or just forgets, that knowledge disappears.

You don't need perfect data to start. You need accurate, accessible data on the things that matter most, and you need it to live somewhere your team and your tools can both reach.

It rarely looks like "we have no data." Usually it looks like one of these:

• Important business knowledge lives on individual machines or personal drives because that's where people actually work. There's no process for surfacing it back to the org.
• A key metric on an executive dashboard is wrong because the underlying variable was misnamed somewhere upstream. Nobody noticed until it became a problem. And it can't just be fixed by querying the database because the issue is in the labeling, not the data itself.
• Multiple teams have built different versions of the same data pipeline because nobody trusts the central one.
• The data exists somewhere, but nobody can tell you where it lives or who owns it.

If any of these are familiar, your readiness score is lower than it looks.

Before any AI tool gets onboarded, do this:

Find where the important data actually lives.

Not where it's supposed to live. Where it actually lives right now. This usually means asking individual contributors directly, because they know what's on their machines that shouldn't be. Set up structured spaces with the right permissions so that information can move from personal to shared without losing context.

Audit your highest-stakes dashboards.

Pick the top 3-5 reports leadership actually uses to make decisions. Trace each number back to its source. You'll find things that are wrong, mislabeled, or pulling from outdated sources. Fix those before adding AI on top, because AI will confidently surface those same wrong numbers faster.

Start with what's already accurate.

Don't try to fix everything at once. Identify one or two high-impact areas where the data is clean enough to be trusted, and start there. Build credibility through small wins before tackling the messy stuff.

• People can answer the question "where does this number come from?" without having to ask three other people.
• Important documents and decisions are being saved to shared spaces by default, not personal drives.
• When something looks wrong on a dashboard, there's a clear path to investigate it, not just a Slack thread.
• The data your AI tools touch is the same data your leadership team trusts. Not a separate, special "AI version."

• Confusing "centralized" with "ready." Just because the data is in one place doesn't mean it's accurate or usable.
• Skipping the labeling work. Variable names and field definitions matter more than people think. A misnamed field becomes a misinformed AI response.
• Believing the IT team has it covered. They might, but the data that matters most for AI often lives outside the systems IT manages. Sales notes, customer feedback, internal decisions, project context.

Data & Infrastructure is the unsexy foundation everyone wants to skip. If it's weak, Operating Model Fit has nothing real to anchor to (you can't redesign workflows around data you don't trust), and Education & Enablement falls apart (you can't train people on tools that surface wrong information).

The good news: you don't need to solve all of it before starting. You need to know honestly where you are, prioritize the highest-impact gaps, and resist the temptation to either skip this work or get stuck in it.