Ask what an assistant would need to answer a real operational question, and data readiness stops being abstract.
"Is our data AI-ready?" is hard to answer in the abstract and easy to answer against a specific question. Take one the business actually asks — which supplier contracts expire this quarter and who approves renewal — and trace what answering it requires.
Four things the question exposes
It needs the contracts to be retrievable, not just stored. It needs the vendor master to agree with them. It needs an approval matrix that reflects who actually signs. And it needs all three to be reachable under the permissions of the person asking.
Any one of those missing produces a confident, wrong answer — the failure mode that erodes trust fastest.
What readiness looks like
Modelled, so relationships are explicit. Documented, so meaning survives staff turnover. Access-controlled, so retrieval respects existing rules. Refreshed on a known cadence, so answers reflect the current state.
That is ordinary data engineering. It is also the difference between an assistant that is used and one that is quietly abandoned.