Knowledge graph
A knowledge graph is a way of storing information as things (people, customers, products, projects, documents) and the relationships between them: this customer owns that contract, which is managed by this account lead, who reports to that VP. Each thing is a node and each relationship is a link, so you can follow connections from one fact to the next.
Why it matters
Many useful questions at a company are about connections, not single facts. “Which open deals involve customers who filed a support ticket this month?” needs the CRM, the support tool and the calendar joined together. A knowledge graph holds those links explicitly, so a system can answer by walking from one node to another instead of hoping the right paragraphs turn up in a search.
For AI, that means fewer guesses. A model that can look up how two things are actually related doesn’t have to infer it from loosely matching text, which cuts down on confident wrong answers.
Knowledge graph vs. vector search
Vector search finds text that means something similar to your question. It’s easy to set up and works well on piles of documents, but it has no idea how things relate. A knowledge graph knows the relationships but takes real work to build and keep current: someone has to define what kinds of things and links exist, and data from each system has to be mapped onto them. Many systems use both, with the graph for structure and vector search for finding relevant text.
Where it goes wrong
Graphs go stale. If nothing updates them when the source systems change, they confidently describe last quarter’s org chart. They also tend to get scoped too broadly at the start. Teams that pick one concrete use case, model only what it needs and grow from there have a much easier time than teams that try to map the whole company first.