Define a dedicated AI-facing knowledge manager role
The work of curating AI's inputs needs an owner; treating it as a side activity reliably loses to the primary role's deadlines.
The notes Start with knowledge management, not tools and Make tacit knowledge explicit, or AI cannot use it both presuppose that someone is doing the work of structuring, curating and updating the inputs an AI deployment depends on. That presupposition is usually wrong. Knowledge management ends up a side activity for someone whose primary role is something else, and it loses every contest with the primary role’s deadlines. The result is the failure mode A document store is not a knowledge management system describes: documents accumulate, structure does not.
The heuristic is to give the work an owner. A dedicated knowledge manager, whose role is curating the firm’s information for AI consumption rather than for human retrieval, captures three functions nobody otherwise owns.
Audit and surface what exists. Most firms do not know what their own information sources contain. The first activity is a working census: where things live, what is current, what is missing.
Convert and structure. Word documents, PowerPoint decks, scanned PDFs and tacit-knowledge interviews each need a different treatment to land usefully in an AI’s accessible context. The conversion is craft work, not routine.
Maintain. Documents go stale; staff leave with knowledge in their heads; new policies arrive without integration into existing material. Without an owner, the maintenance does not happen and the knowledge base degrades faster than it grows.
The role does not need to be senior. It does need to be defined. The pattern that has worked at the engagements visible to this wiki is to identify a capable analyst already adjacent to the work and reframe their role rather than recruit externally. External hires arrive without the institutional context the role needs.
The role’s success measure is unconventional. Useful AI output is the downstream signal, but the operational measure is more concrete: how much of the firm’s actual practice is reflected in structured form that an AI can use, and how quickly that material is updated when reality shifts.
The recommendation has now arrived independently in multiple advisor engagements at organisations of broadly comparable scale, with the same structural reasoning each time. The reason it keeps recurring is described in AI treats documentation as authoritative: once AI is the primary access route to organisational knowledge, the maintenance work on the underlying documentation goes from optional to load-bearing, and an owner who does only that work — at full attention rather than as a side activity — becomes the structural fix. The role exists because the work exists, and the work exists because of how AI consumes documentation.