McKinsey connects enterprise data through a knowledge graph for AI

McKinsey is making a familiar point with a new wrapper: enterprise data gets a lot more useful when you connect it as a graph instead of leaving it in disconnected tables. James Kaplan, a distinguished partner at McKinsey & Company, argued that graphs are already part of daily life for anyone using LinkedIn, Wikipedia, or social media. The enterprise, he said, is just late to the party.

His case is simple. Relational databases are excellent for transactional work, but they struggle when the data is messy, ambiguous, or tangled up in business relationships. A graph, by contrast, can show how a customer, product, process, or step in a process relates to everything else. That context is what AI applications often need if they’re going to be more than fancy autocomplete.

Kaplan said AI can now help turn unstructured information into structured data and business rules that can be stored in a knowledge graph. In the past, that kind of work relied on business analysts or data scientists, and it was slow, expensive, and never quite clean. Now, he said, AI can interrogate complicated processes programmatically and create deterministic business rules. That’s a big shift in how companies can handle messy data.

McKinsey is using this idea through EcliptOS, an AI operating system that links C-suite strategy with everyday execution through agentic workflows. The system uses a semantic data layer to organize data and relationships for generative AI applications. Kaplan described it as a graph of databases, built with flexible schemas and a virtual layer that can connect many different systems.

He also pushed the point that not every AI effort should be aimed at productivity. Customer experience might matter more, depending on the business. That’s the useful part of the argument here: knowledge graphs are being sold not as another data toy, but as a way to decide what the machine should actually know before it starts acting clever.