Sia Partners With Neo4j and Lettria to Build Enterprise Knowledge Layers

AI agents are moving beyond answering isolated questions. Organizations increasingly expect them to support decisions and take part in business processes. To do this reliably, agents need more than access to information: they need to understand how the business works, how its entities connect and which rules apply. 

That knowledge is often scattered across structured systems and unstructured sources such as contracts, research reports, technical documentation, procedures, emails and employees’ expertise. Retrieval-augmented generation (RAG) can find relevant passages, but questions that cross systems or require several connected facts can be harder to answer and verify. Different teams may use the same term in different ways, and reviewers may not be able to trace an answer to its source and the rule behind it. 

An Enterprise Knowledge Layer connects structured data and unstructured content on top of the systems an organization already uses. It does not require companies to migrate or replace their existing sources. 

Ontologies define the shared business vocabulary: concepts, entities, relationships and rules. Knowledge graphs connect those concepts to one another and to the data and documents where the knowledge lives. Domain specialists validate and maintain the model, grounding it in how the business actually works. Powered by Neo4j, knowledge graphs provide the connected business context that underpins a Context Engine, helping AI agents access and reason over relevant enterprise knowledge. 

This is especially valuable for unstructured data. Documents can be analyzed to identify entities, facts and relationships, then linked to relevant business concepts and structured data. AI assistants can use those connections to answer questions across sources and point reviewers to supporting documents and rules. 

Structuring knowledge also helps control token use and inference costs. Rather than sending broad collections of text to a model for every question, an agent can retrieve the relevant part of the graph and use more targeted context. 

The partnership brings together distinct capabilities to build and scale the layer. Lettria helps structure documents and data into ontologies and knowledge graphs. Neo4j provides the graph technology to run those graphs in production. Sia leads use-case framing, architecture and adoption, ensuring the knowledge layer supports real business needs.  

This approach is already taking shape at L’Oréal Research & Innovation, where Sia has been working for two years to make 116 years of scientific knowledge usable by experts and AI. 

Building an Enterprise Knowledge Layer is a long journey. Our approach helps organizations build it step by step, starting with a focused use case. To get started, a four-week assessment maps relevant data and governance, prioritizes use cases and checks whether a graph-based approach fits. An eight-to-twelve-week pilot then builds the layer for one domain and delivers an assistant evaluated with business teams against agreed questions and target results before a go-live decision. 

Organizations retain their ontologies, documentation and operating method in open formats. As new use cases emerge, domain specialists can extend the model and connect it to other domains instead of rebuilding context from scratch. 

The goal is to make enterprise knowledge a shared, governed asset that every use case can build on—so people and AI can act on what the organization knows.