Knowledge Graph (for Business)

An institutional knowledge graph is a structured, queryable representation of a firm's accumulated documents, decisions, and tacit memory — turning the most valuable asset most mid-market firms own (eight years of operational history) into something both humans and AI systems can retrieve from. It changes the access model from social — "ask the longest-tenured person in the room" — to systematic.

The architecture has three layers. At the bottom, the firm's source documents — HR policies, engineering artifacts, customer interactions, contracts, prior decisions, internal memos — are normalized into a consistent format. In the middle, the graph captures the relationships between those documents: which client this contract belongs to, which decision this memo recorded, which engineer authored this artifact, which compliance period this audit covers. At the top, retrieval tools (RAG, search, dashboards) read the graph to answer questions for humans, LLMs, and agents.

Why operators care: the cost of the social retrieval model is invisible but real. A senior employee leaves and takes years of context with them. An onboarding takes six months because the new hire has to discover the firm's playbook by accident. A customer escalation hits and the team scrambles to reconstruct the history. A knowledge graph is the operating answer to all of these — the asset compounds while the firm's tenure does not.

A pragmatic starting point is markdown plus a structured frontmatter convention (Obsidian-class tooling) plus a RAG layer on top. That gets a working firm to a queryable knowledge surface in weeks, not quarters, and it scales into a more formal graph database later if the volume justifies it. The architecture is incremental. The discipline of writing decisions down so the graph has something to ingest is the part that has to start now.