Tabular data answers 'what is the value of X'. A knowledge graph answers 'how is X connected to everything else' — which is the question analysts, risk managers and AI agents actually ask. Contagion, supply chains, ownership webs, fund exposures: all of these are graph queries wearing spreadsheet costumes.
A financial knowledge graph does not need exotic technology to start. It needs typed entities (company, security, market, fund, country), typed edges (lists_on, domiciled_in, managed_by, issues, regulates) and stable identifiers on the nodes. A relational database with a well-designed relationships table is a knowledge graph in the way that matters.
The compounding effect is what makes graphs worth it: each new entity type multiplies the value of existing ones. Add funds to a company graph and you can suddenly answer exposure questions; add countries and you get jurisdiction risk; add banks and you get counterparty maps. The marginal dataset gets cheaper to add and more valuable, the opposite of siloed tables.
For AI, the graph is the difference between retrieval and reasoning. A language model given linked entities can traverse — 'this fund is managed by this company, listed on this exchange, in this jurisdiction, whose central bank just moved rates' — instead of pattern-matching over disconnected text. Structured context is the highest-leverage input an agent can be given.