What it means
Missing data is not always missing in the obvious sense. It can be data that exists but is unstructured (notes in a text field), data that was never captured (the reason a customer cancelled), data that exists on paper, data that lives in WhatsApp screenshots, or data that was once captured but has since been deleted.
An AI deployment is constrained by what data actually exists, not what you think you have. The first audit always finds at least one critical field that lives only in someone's head and needs to start being captured before the AI can be useful.
Why it matters
Discovering missing data three weeks into a build is the most expensive way to discover it. Discovering it during a 90-minute data audit at the start is cheap. Either you find a workaround (proxy data, manual capture for a few weeks, schema change in the source system) or you cut the use case until the data exists.
The honest answer to 'can AI do this?' is sometimes 'not yet, because the data is not there'. That is useful information. It tells you what to start recording today so AI can use it in six months.
Example
A wealth-advisory firm wants AI to draft follow-up notes after client meetings. The audit finds that nobody has been recording meeting agendas or action items in any structured form, just in advisors' notebooks. The AI deployment pauses; the firm spends six weeks getting advisors to log meetings in the CRM, then the AI ships with real data to work with.