Is your data ready for AI? 7 signs it isn't

By Barry Middlebrook · Middlebrook Data & AI Governance

Every finance team is racing to put AI on its reporting. Most haven't asked the only question that decides whether any of it can be trusted: is the data underneath governed? Here are seven tells that the foundation isn't ready — each one a place where AI will quietly produce a confident, wrong answer.

1. Three teams define "revenue" three ways. If finance, sales, and the warehouse don't agree on a metric, the AI will pick one at random — and it won't tell you which.
2. No one can trace a number back to its source. If a figure in a report can't be followed to the system of record, neither you nor an auditor can defend it.
3. Your AI queries raw tables, not a governed layer. Pointed at raw data, AI improvises definitions. That's the single biggest cause of wrong answers.
4. "Customer" means different things in different systems. Without one golden record per core entity, the AI reasons over contradictions.
5. You discover data-quality problems in board decks. If issues surface downstream instead of being monitored upstream, AI is already amplifying them.
6. AI answers arrive with no provenance. No sources, no definitions, no as-of date — nothing to verify, nothing to audit.
7. Nobody actually owns the data. Governance fails as a technology project and succeeds as an accountability structure. No owner, no trust.

Each sign is a place AI will fail silently — confidently, at machine speed, with no analyst in the loop to catch it.

What to do about it

Don't boil the ocean. Start with the highest-leverage fix: define your key metrics once, in a governed semantic layer, and point the AI only at that layer. Add lineage so numbers can be traced, name an owner for the domain, and require provenance on every AI answer. Fix one high-value reporting domain end-to-end, then expand.

The honest first step is simply knowing how many of these seven apply to you today.