AI-ready data
Company data an AI system can use safely: accessible by software, documented, current, with an owner and with permissions that carry over to the AI.
AI-ready data is data an AI system can read, trust and use without leaking it to the wrong people. In practice that means five things: software can reach it through an API or export, it is documented (what each field or folder means), it is current, it has an owner who fixes errors, and its access rules are known.
For document search and assistants built on RAG the problems are specific. Old versions sit next to new ones, so the model quotes last year’s price list. Duplicates and drafts outnumber final documents. Scans without text cannot be searched. Permissions get lost: if the index ignores who may see what, the assistant shows HR files to everyone, so access control has to follow the data, for example with row level security. Location matters too: where the index and model run is a data residency question for contracts and GDPR.
The common mistake is a big data cleanup program before any use case. It takes a year and cleans the wrong things. Make data ready one use case at a time: pick the first pilot, list exactly which sources it needs, fix ownership, versions and permissions for those, then move on. An AI readiness assessment usually finds that data, more than models, sets the pace.
Related terms
- RAG (retrieval-augmented generation): Finding relevant documents first and putting them into the prompt, so the model answers from them.
- Data residency (AI and cloud): Where your data is stored and processed, including where an AI model runs on it, and whether it can leave a region such as the EU.
- Row Level Security (RLS): Database rules that decide, row by row, which records each user may read or change. Missing RLS is a common leak in AI-built apps.
- AI readiness assessment: A short review of a company’s data, people, processes, governance and systems that ends with a ranked list of AI use cases, blockers and a 90-day plan.
- Chunking (RAG): Splitting documents into smaller passages before indexing them for search, so retrieval returns the relevant part instead of a whole file.
- AI use-case prioritization: Ranking candidate AI use cases by business impact and feasibility (data, risk, time to value) to pick which to pilot first and which to drop.
