Architecture

Knowledge Graph

A knowledge graph is a structured representation of information as interconnected entities and relationships, enabling AI systems to reason across complex, connected data in ways that isolated documents or database records do not support.
DATE UPDATED: 11 Sep 2026
|
FIRST PUBLISHED: 22 Mar 2026

What a knowledge graph is

Most organisational data exists as isolated units: a document here, a database record there, an email thread somewhere else. Each unit contains information, but the relationships between units, the connections that give that information meaning in context, are largely invisible to the systems that store it. A knowledge graph makes those relationships explicit and navigable.

In a knowledge graph, information is represented as nodes, which are the entities: clients, matters, individuals, regulations, precedents, and the edges between them, which are the relationships: this client is involved in this matter, this matter falls under this regulation, this regulation was interpreted in this precedent. The graph structure means that a query does not just retrieve a single document or record. It can traverse the relationships between entities to surface information that is connected to the query across multiple hops.

The difference in practice is significant. A conventional search for a client name retrieves documents that mention the client. A knowledge graph traversal can retrieve the client, the matters they have been involved in, the regulatory frameworks relevant to those matters, the precedents that have informed the firm’s approach to similar matters, and the colleagues who have worked on comparable cases, all in a single structured response that reflects the actual relationships between those entities rather than their co-occurrence in documents.

Knowledge graphs and AI

Language models are powerful at generating text but relatively weak at reliably navigating complex webs of specific factual relationships. A knowledge graph provides a structured foundation that an AI system can query precisely, supplementing the language model’s generative capability with a reliable source of connected factual information. The combination allows an AI system to produce responses that are both fluent and grounded in an accurate representation of the relationships between entities in a specific domain.

For organisations with complex, interconnected information, where the value lies not just in individual documents but in understanding how entities, clients, matters, regulations, risks, relate to each other, a knowledge graph is a more appropriate knowledge infrastructure than a simple document store or even a well-designed RAG system. The two approaches are often complementary rather than alternatives.

The investment required

Building a knowledge graph requires more initial investment than indexing documents into a vector database. The entities and relationships need to be defined, the existing information needs to be structured according to those definitions, and the graph needs to be maintained as new information arrives and existing information changes. This investment is justified where the connected nature of the information is itself a source of value, which in most complex professional services organisations it is.

Insight

Professional services organisations are particularly well-suited to knowledge graph architectures because their most valuable knowledge is inherently relational. The value of understanding that a particular client has interests across multiple matters, that those matters involve overlapping regulatory frameworks, and that the firm has relevant experience from analogous situations elsewhere in the portfolio is not accessible from documents in isolation. It requires a structured representation of the relationships between entities. Law firms and financial services businesses that invest in knowledge graph infrastructure are building a knowledge asset that becomes more valuable as it grows, and that is significantly harder for competitors to replicate than any individual AI tool.

Misconceptions

A knowledge graph is not the same as a database. A relational database stores records and supports queries against defined fields. A knowledge graph stores entities and relationships and supports traversal across those relationships in ways that relational databases are not designed for. The two are complementary rather than interchangeable.

Knowledge graphs do not build themselves. Defining the entities and relationships, structuring existing information according to those definitions and maintaining the graph as information changes requires sustained organisational investment. The architecture is only as good as the effort put into building and maintaining it.

A knowledge graph is not appropriate for every AI use case. For organisations whose information is primarily unstructured and whose AI needs are primarily around document analysis and generation, a well-designed RAG system may be more appropriate. A knowledge graph adds most value where the relationships between entities are as important as the content of individual documents.

BACK TO GLOSSARY