Is your data actually AI-ready? Or is it tripping you up?

I said this in an interview in April, but I will continue to keep harping on about needing context! Most "is your data AI-ready?" conversations I see floating around focus on the visible stuff; duplicates, messy fields, inconsistent naming. But there's also a layer sitting above that which is harder to spot: data that's technically usable, but not actually self-explanatory.

You might have tables where users, transactions, and fees are represented by coded values, and there might not be a reference table or piece of documentation sitting alongside them. There may be status codes that don't mean what they look like. There could be filters you have to apply before a count is trustworthy. Usually none of it is written down anywhere the data itself can point to...

I might know that context because I work with the data every day, so when I'm prompting, I can say "include these categories, exclude those ones, treat this field differently here", and sense-check the result if something looks off. But someone else could access exactly the same data, ask a perfectly reasonable question, get a perfectly plausible answer, and still be wrong because they didn't know the unwritten rules.

In that case the “fix” is less a cleanup effort and more around documentation: the semantic layer, business definitions, filter logic, edge cases, all written down somewhere the model (and the person) can actually see. Otherwise AI just makes it easier to produce a really confident interpretation of the wrong thing.

The silver lining is that AI readiness has quietly turned data maturity from optional into essential, which is good for any business whether or not they end up doing much with AI (and, selfishly, for my own sanity)!