Find the right document. Keep the right context.

Scanned records, searchable PDFs, spreadsheets and office documents behave differently. Review representative files, their quality and how they are currently organised. The project should identify which document types are in scope and who can decide whether a record is current or superseded.

Folders, metadata and document categories should reflect the team’s work. Define identifiers and version relationships before adding automated classification. Poor naming and duplicate files need a resolution process so the new system does not simply make an inconsistent archive easier to search.

Text extraction can help make records searchable, but scans, tables and unusual layouts may produce errors. Establish sample checks and a route for correcting important fields. Extracted information should not silently replace an authoritative record without review.

Search and assisted answers need to respect the user’s permission to view the source. A summary should not expose restricted content that the same person could not open directly. Source references also help staff inspect the original wording before relying on an answer.

Agree how new versions are ingested, how removed files leave search results and how long records remain available. Hosting, storage, model processing and backup responsibilities need to be specified. A pilot using representative documents is a better basis for rollout than a demonstration on a few clean files.