Every failed AI initiative we review traces back to data: not model selection, not talent, not budget. The model is downstream of a decision about whether the data was ever positioned to be consumed. Readiness is not a data-quality project that finishes; it is an operating condition that must hold before AI scales.
What “AI-ready data” actually means
- Ownership: a named domain owner accountable for quality, not a committee.
- Lineage: every field traceable to its source system and transformation.
- Access: governed self-service, so AI teams do not wait in queues.
- Freshness: the cadence the use case needs, not the cadence the warehouse inherited.
Notice what is missing: volume. Most enterprises have more data than they can govern. Readiness is about the narrow slices AI will actually consume, and making those slices trustworthy first.
Assessing readiness in weeks, not quarters
A focused assessment interviews the domain owners, samples the target datasets, and scores them against the four conditions above. It does not attempt an enterprise-wide data census, which is why it can run in three to six weeks and produce a decision-ready readout.
READINESS SCORECARD (per domain)
Ownership ████████░░ 8/10 named owner
Lineage █████░░░░░ 5/10 partial, manual
Access ████░░░░░░ 4/10 queue-gated
Freshness ██████░░░░ 6/10 daily batch
→ Blocking for: customer-service AI (lineage, access)The output is a short list of what blocks each candidate use case, and the sequence to unblock them. That is the baseline every funded AI roadmap should start from.