What does 'data readiness for AI' mean?

Whether the data a specific AI workflow needs is accessible, good enough quality, structured enough to use, properly governed, and available in sufficient volume. It's the most common reason AI projects stall.

Do I need a data warehouse before starting AI?

Usually not. You need the specific data for your target workflow to be accessible and reasonably clean — not a perfect enterprise-wide platform. Over-investing in data infrastructure before shipping anything is a common trap.

How much historical data do I need?

It depends on the approach. Retrieval and prompting can work with modest data; fine-tuning needs more. The assessment flags whether your volume is a likely blocker for the use case.

Data Readiness Assessment for AI — free checklist & score

How to use it.

1. Assess for a specific workflow

Data readiness is per use case, not company-wide. Answer for the data the AI project actually needs — not your entire data estate.

2. Check honestly across five dimensions

Access, quality, structure, governance, and volume. Inflating your score only hides the problem that will stall the project later.

3. Read your band and weakest dimension

You get a readiness percentage, a band (Foundational / Developing / Ready), and the single weakest area — which is usually what to fix first.

4. Fix the blocker, then re-score

Most AI projects stall on data access or quality, not the model. Address the flagged dimension and re-run to track progress.

Frequently asked questions.

What does 'data readiness for AI' mean?: Whether the data a specific AI workflow needs is accessible, good enough quality, structured enough to use, properly governed, and available in sufficient volume. It's the most common reason AI projects stall.
Do I need a data warehouse before starting AI?: Usually not. You need the specific data for your target workflow to be accessible and reasonably clean — not a perfect enterprise-wide platform. Over-investing in data infrastructure before shipping anything is a common trap.
How much historical data do I need?: It depends on the approach. Retrieval and prompting can work with modest data; fine-tuning needs more. The assessment flags whether your volume is a likely blocker for the use case.