Company know-how lived in senior staff's heads, and one question had three answers
The company's know-how lived in senior staff's heads and in chat histories, and it left when they did. A new hire with one question had to interrupt three people, and the same question had three different answers. Buying an AI tool wouldn't fix it: if the knowledge itself isn't organized, AI can't make sense of it either.
- Client
- Advertising agency · internal operations
- Scope
- Company knowledge base
- Key result
- Rules first · groundwork for AI Q&A
The problems we found
- Knowledge existed only in people's heads and chat logs, and it left when they did
- New staff had to interrupt several senior colleagues for an answer, and the answers often contradicted each other
- Just buy an AI tool? If the knowledge itself isn't organized, AI can't make sense of it either
What we changed
- Rules first: each fact lives on exactly one card. Duplication = future contradiction, so reference instead of copying
- Secrets are stored as pointers, never the secret itself: the knowledge base never contains passwords, only where they are kept and who to ask for access
- Knowledge cards follow a structured format with automatic validation (schema checks, broken-link scans), so the system enforces quality
- The sections of the knowledge tree are the future permission boundaries: once AI Q&A is connected, who can ask about what is already governed by the structure
Results
- Rules and the first set of knowledge cards are being rolled out
- Groundwork laid for an internal assistant where you ask AI and get an answer
For AI to answer questions about your company, the company's knowledge first has to be organized into a shape AI can read. Set the rules first, then bring in the tools.
Does your company's know-how live in a few people's heads?
Start with a free 30-minute call about how the work runs today, and we'll judge whether a closer look is worthwhile. Finding where it really gets stuck means looking at the actual workflow.