Bringing a small practice's client knowledge into one system

A two-partner professional-services practice needed a reliable way to answer questions about a client's history without digging through years of documents and correspondence each time, and a way to keep follow-ups from depending on memory. I designed a dossier system in NotebookLM: one notebook per client, each holding only that client's source pack, all of them running on the same shared instructions and the same shared knowledge base. Around it I set up the firm's working tools on its own accounts: Claude for project work and documentation, ChatGPT for research, Attio for contacts and relationships, Linear for tasks and follow-ups, and an AI-ready file system for the whole firm. Then I trained the team to run it day to day.

• Shared instructions. One set of chat instructions used by every client notebook, so every notebook behaves the same way: cite the document, separate facts from inferences from open questions, and recommend professional review when the evidence is thin.
• A shared knowledge base with a written source hierarchy. When sources disagree, the order is fixed: the official current source first, then the client's own source pack, then the shared knowledge base, then general heuristics. The shared layer is written as a maintenance layer with refresh and retirement rules, not as a frozen authority.
• A reference layer. A compact reference on the recurring topics the practice handles, each anchored to its official source, with a refresh cadence.
• One client per notebook, on a firm-wide file system. An AI-ready folder structure for the whole practice, with a source pack per client, and the first client's source area set up to show the pattern.
• The tools around it. Claude for project workflows and document work, ChatGPT for research, Attio for contact and relationship tracking, Linear for tasks and follow-ups, all set up inside the firm on its own accounts. I documented the structure and trained the two partners on daily use.