Write Down Where Your Data Came From

Every AI system is a mirror of its data. So when an auditor asks about your model, they're really asking about your data — and "it came from somewhere in the system" is not an answer.

Keep a simple data inventory. For each dataset your AI touches, write down five things: where it came from, when you got it, what's in it, who can access it, and what you know is wrong or missing from it. That last one matters most — every dataset has gaps, and naming them is a sign of maturity, not weakness.

I learned this building databases: if you don't know what's in the tables, you don't know what your queries mean. The same is true tenfold for AI. A model trained on last year's customer data will make last year's assumptions. If nobody wrote that down, nobody will remember to question it.

You don't need data lineage software to start. A spreadsheet works. What matters is that someone, somewhere, can answer the question "where did this data come from?" without going on an archaeological dig.

Data you can't trace is data you can't defend. Write it down.

I explain frameworks; I don't give legal counsel.