Your data is ready for AI when three things are true: the metrics that matter have one agreed definition, the data behind them arrives complete and on time, and someone is accountable when either breaks. Most companies fail the first test before the technology is ever involved. The good news is that you can find out where you stand in about a week, without buying anything.
Start with the one-meeting test
The fastest readiness check I know takes a single meeting. Pick your most important metric — revenue per customer, active users, fill rate, whatever runs the business. Ask three different teams to pull it independently, then compare the numbers in the same room.
If they match, you are ahead of most companies I've worked with. If they don't — and at one large enterprise I worked with, three teams produced three materially different answers to the same question — you haven't failed the test so much as found your first project. Most AI readiness problems are not AI problems. They are definition problems. If Finance and Marketing compute "active customer" differently, no model fixes that — it just automates the disagreement.
The five checks that actually predict AI success
After the one-meeting test, run these five checks against the two or three data sources your first AI use case would depend on. Score each honestly: solid, shaky, or unknown.
- Definitions. Is there one written definition per key metric, with an owner? Not a data dictionary nobody reads — a definition someone will defend in a meeting.
- Completeness. What percentage of records have the fields your use case needs? Pull 100 random records and count. Below roughly 90%, a model learns your gaps instead of your business.
- Freshness. How stale is the data when decisions get made from it? Daily decisions on weekly data is a readiness gap no algorithm closes.
- Lineage. Can anyone trace a number on a dashboard back to its source system in under an hour? If the answer involves a person who left the company, write that down — it is your biggest risk.
- Accountability. When a pipeline breaks or a number looks wrong, is it clear whose job it is to fix it? Data quality without an owner decays back to baseline within a quarter. I've watched it happen on teams with excellent tooling.
Two or more "shaky" scores on a source means fix before you model. All five solid means you are more ready than most of the market — pick a narrow use case and move.
What readiness does not require
Teams routinely over-scope this. You do not need a new data warehouse, a governance committee, a catalog tool, or a year of cleanup before starting AI work. Readiness is scoped to a use case, not to the company. A team I led shipped a useful forecasting workflow on exactly two well-governed tables while the rest of the platform was, charitably, a work in progress. The scope of cleanup is the scope of the use case — that is what makes this tractable in weeks instead of quarters.
The inverse trap is just as common: buying an AI tool first and hoping the data sorts itself out. It won't. The tool will demo beautifully on the vendor's data and then inherit every unresolved definition fight you already have.
What to do with your score
Write the scores down, even roughly — a table in a doc beats a feeling in a hallway. Then sequence the gaps: definitions first (they are political, not technical, and take the longest), completeness and freshness second, lineage as you go. Re-run the one-meeting test after a month. When three teams pull the same number, you're ready for the interesting work.
If you want a second set of eyes on the assessment — or the meeting where three numbers don't match needs a referee — that is exactly the kind of engagement I do. Book a free strategy call and bring your worst metric. No prep needed.