Is your data ready for AI? Five foundations to get right first

Many AI initiatives stall not because the model is wrong but because the data is. Documents are duplicated, records conflict, permissions are unclear and nobody is sure which source is authoritative. Fixing these issues after an AI system is built is slow and expensive. Addressing them first makes everything that follows easier.

1. Know what you have

Start with an inventory of the data sources relevant to your priority use cases. For each one, record the owner, how current it is, how it can be accessed and how sensitive it is. A lightweight catalogue is far better than none.

2. Decide what is authoritative

When the same customer, product or policy appears in several systems, decide which one is the source of truth. AI systems that retrieve from conflicting sources produce conflicting answers, and users lose confidence quickly.

3. Carry permissions with the data

An AI assistant must not reveal information a user is not entitled to see. That means access rules from source systems need to travel with documents and records into search indexes and data platforms, and be enforced at query time.

4. Test data like you test code

Automated checks for freshness, completeness, uniqueness and valid values should run on every pipeline. When a load fails or data drifts, the right people should know before an AI system or report uses it.

5. Clean the unstructured content

Document collections accumulate drafts, outdated versions and near-duplicates. Archiving obsolete material and adding basic metadata such as document type, owner and effective date measurably improves retrieval quality.

Start with the use case, not the platform

You do not need to fix all of your data before starting with AI. Pick one or two valuable use cases, identify the data they depend on and bring that data up to standard. Each project then leaves behind cleaner, better-governed data for the next.