Twenty years of network drives become a searchable archive with retention that runs itself

The target is designed before anything moves, because the failure mode of every archive project is copying the folder tree into SharePoint and calling that a migration. The architecture is a few sites by business domain, libraries with content types the business recognises, and columns people actually search on: counterparty, policy or claim number, document type, effective date, records class.

Born-digital files are classified and described inside SharePoint by document processing for Microsoft 365, which supplies prebuilt and custom models plus autofill columns that write metadata from the document itself, consumed pay-as-you-go. Legacy scans need a different engine: UiPath Document Understanding handles what a 2008 fax looks like after two photocopies. Robots log every file with its hash, collapse exact duplicates, and group near-duplicates by type and number so one version survives as operative and the rest are linked as superseded.

Retention is applied as the document is filed. Each library and content type carries a default retention label, Microsoft Purview auto-apply policies cover what location alone cannot decide, and disposition review goes to a named records owner. People appear only where judgement is required: what the models cannot place, version conflicts and misplaced personal data become tasks in Microsoft Teams, and new arrivals from the scanner library and the shared mailbox run the same rules.