“We have a lot of data” is not the same as “we have the data this product needs.” Volume cannot compensate for unclear ownership, missing context, inaccessible systems, inconsistent definitions, or weak links between inputs and outcomes.
Start with the use case
Data readiness is relative. A knowledge assistant, fraud model, forecasting tool, and document workflow each require different evidence. Define the decision or task first, then map the minimum data needed to perform and evaluate it.
Assess five dimensions
- Availability: Does the relevant data exist across the necessary period and population?
- Access: Can the team legally and technically use it?
- Quality: Are important fields complete, consistent, timely, and accurate?
- Meaning: Do teams agree on definitions, labels, and business context?
- Operations: Can the data be refreshed, monitored, corrected, and governed in production?
The goal of a readiness assessment is not to certify perfection. It is to identify which uncertainty could invalidate the product.
Prototype the data path
Trace a small representative sample from source to user-facing output. Include permissions, transformations, retrieval, and feedback. This vertical slice exposes practical obstacles faster than a broad inventory.
Make gaps part of the roadmap
Some gaps can be handled through product design: narrower scope, explicit user input, human review, or a fallback. Others require foundational work such as instrumentation, taxonomy, access control, or data contracts. Separate the two so a multi-year cleanup program does not become the default answer.
Organizations become ready for AI by learning through a bounded use case, not by declaring the data estate finished. The right first product improves the data system as it delivers value.