By Jaap van Duijvenbode
Co-Founder and VP Product Strategy & Customer Experience
Summary: Connecting an AI assistant to enterprise files does not give it organizational knowledge.
What makes information useful to AI is context: meaning, permissions, provenance, relationships and relevance. Organizations that build that context layer get answers they can trust; those that skip it get confident answers built on the wrong document.
More access, same blind spots
Most enterprise AI programs start with a connection. An assistant is pointed at SharePoint, a file share or an object store, and the expectation is that everything in those locations becomes usable knowledge. It rarely works that way. The assistant finds files, but it cannot tell the approved contract from the draft, the current policy from the one replaced in 2019, or the drawing that was built from the one that was abandoned.
The problem is not the volume of data. Gartner has predicted that through 2026, organizations will abandon 60% of AI projects that are not supported by AI-ready data. Most of those organizations have plenty of data. What they lack is the context that tells an AI system which information matters, who is allowed to see it, and whether it can be trusted.
What turns information into context
Context is the set of signals around a document that an experienced employee uses without thinking about it:
- Discoverability: the AI can find the information at all, including content in scans, recordings and legacy formats.
- Meaning: text, tables and entities are extracted and indexed, so a question about a counterparty finds every agreement that names it, not just files with a matching word.
- Permissions: the AI sees only what the person asking is already allowed to open, based on the access rules the files carry today.
- Provenance and version history: every answer points to the file, page or passage it came from, and older versions are recognized as older.
- Entities and relationships: people, organizations, locations and agreements are linked, so the AI can follow a thread across documents.
- Relevance: the AI receives the passages that answer the question, not every document that mentions a keyword.
- Governance: every query is recorded, sensitive fields can be redacted, and sources can be excluded from AI entirely.
Why this matters now
Three shifts make context urgent. First, organizations now run several assistants and agents side by side, and each one needs the same trusted view of enterprise information. Rebuilding context separately for every AI tool is expensive and inconsistent. Second, agents act on what they retrieve. An assistant that surfaces the wrong document produces a bad answer; an agent that acts on it produces a bad outcome.
Third, context changes the economics of enterprise AI. When retrieval isn't precise, an AI system processes far more information than it needs, increasing token consumption and cost. A well-built context layer returns the right passages the first time.
Where to start
Begin with a question the business cares about and trace what a correct answer requires: which sources, which permissions, which versions, which relationships. That exercise exposes the gaps faster than any inventory. From there, build context once, close to where the data already lives, and make it available to every assistant and agent through open standards rather than one-off integrations.
Organizations that treat context as infrastructure get AI answers they can defend. Those that treat it as an afterthought keep adding data and wondering why the answers do not improve.
Want to know how ready your enterprise information is to serve as trusted AI context? Talk to our specialists.
Additional sources:
