McKinsey’s Knowledge Graph Pitch for Enterprise AI: Architecture, Not Evidence

McKinsey describes EcliptOS as using a semantic data layer and a knowledge graph to give AI applications business context. The SiliconANGLE report does not provide measured evidence that this approach improves decisions, reduces maintenance work, or outperforms other ways of connecting business data. ¹ That makes EcliptOS worth examining as a design pattern, not as a demonstrated outcome.

What McKinsey described about its knowledge graph for AI

A SiliconANGLE report published October 9, 2026, describes EcliptOS as an AI operating system designed to connect C-suite strategy with everyday execution through agentic workflows. It says the system uses a semantic data layer to organize data and relationships for generative AI applications. ¹

McKinsey distinguished partner James Kaplan described a graph of databases and said it was easier to create a “virtual graph” connecting many databases. He also described graphs as having more flexible data schemas. ¹ The report does not include an architecture diagram, identify production data sources, or explain how the graph is built. ¹

That gap matters. “Knowledge graph” describes a way of representing entities and their relationships; by itself, it does not specify what data is included, how relationships are defined, or how their accuracy is assessed.

What the enterprise AI knowledge graph represents

Kaplan gave customers, products, processes, and process steps as examples of things whose relationships can provide context. The report does not establish that these are the data types used in EcliptOS. ¹

The report does not specify which databases or documents feed the graph, whether it spans operational, financial, or customer data, or which kinds of relationships are modeled. ¹

Analysis: A graph could make enterprise data relationships explicit, for example by distinguishing a process step from the process it belongs to. That potential benefit depends on accurate definitions and links. The report does not explain how EcliptOS handles incomplete, conflicting, or outdated source data. ¹

From unstructured information to AI business context

Kaplan said AI can help turn unstructured information into structured data and business rules that can be stored in a knowledge graph. The report does not say how much of that pipeline is built into EcliptOS. ¹

One way to interpret the described architecture is that source information becomes entities, relationships, and rules, and an AI application then uses those connections as context for a task. Analysis: This is a possible model of how the approach could work, not an outcome reported by McKinsey.

For an enterprise buyer, practical questions include whether each fact has a source and owner, whether uncertainty is recorded, and whether effective dates distinguish current information from stale information. The report does not discuss provenance, confidence scoring, freshness, or ownership controls in EcliptOS. ¹

What the source says, and what it does not measure

The report describes an approach but supplies no performance comparison or measured outcome. It does not report whether EcliptOS improves decision quality, increases productivity, or reduces the effort of maintaining enterprise context. ¹

Question	What the report supports	What remains unknown
Is there a semantic layer?	EcliptOS is described as using one to organize data and relationships. 1	Its schema, scale, and production coverage.
Can it connect multiple databases?	Kaplan described a graph of databases and a virtual graph that can connect many databases. 1	Which databases are connected in practice.
Can unstructured data become graph content?	Kaplan said AI can help create structured data and business rules for storage in a graph. 1	How much of that process is live in EcliptOS. 1
Are outcomes better?	The report supplies no measured results. 1	Effects on decision quality, productivity, or maintenance effort.

The EcliptOS discussion reports no figures for decision accuracy, hallucination rates, time saved, cost, adoption, data quality, or graph-maintenance load. ¹ It therefore does not show whether this approach outperforms a company’s existing way of connecting and presenting data. ¹ The same gap between a described capability and a deployment decision shows up elsewhere; see Sakana AI’s 73% core-claim result is a benchmark, not a deployment decision.

The report discloses that theCUBE was a paid media partner for the interview series and that Neo4j sponsored the event coverage, without editorial control over the content. ¹ That disclosure does not answer the implementation or outcome questions; it gives readers context for the account.

Governance questions for knowledge graphs for enterprise data

The report does not explain how EcliptOS creates, validates, updates, versions, or retires graph data, relationships, or business rules. It also does not describe who governs those definitions, how access is controlled, or how schema changes are managed. ¹

Kaplan said a virtual graph can connect many databases and that graphs have flexible data schemas. ¹ Analysis: Such an approach could reduce the need for full data consolidation in some cases, but it would not by itself resolve inconsistent definitions, data quality, or responsibility for changing business rules.

Governance is an evaluation criterion for any buyer. If an AI application relies on a relationship to support a decision, the buyer should establish where that relationship came from, whether it is current, and who is accountable for its definition. These are questions to investigate, not claims about EcliptOS. For a broader framing of evidence requirements, see AI agent governance: what evidence is enough before an agent can take action.

What enterprise AI teams should do

Treat the McKinsey example as a hypothesis to test, not proof that knowledge graphs improve enterprise decisions. The report describes EcliptOS as using a semantic layer and discusses a graph approach, but offers no outcome evidence. ¹

A bounded pilot can start with one decision process rather than a broad mandate to “add context.” Before choosing graph technology, define:

• The decision or workflow the application must support.
• The source systems and documents it may use.
• The entities, relationships, and business rules the task requires.
• How fresh the data must be and who owns each definition.
• How the team will handle missing, contradictory, or stale information.

Then compare the pilot with the status quo. Analysis: Potential measures include decision quality, review effort, response usefulness, and the ongoing cost of maintaining the graph. Set the baseline and evaluation method before deployment; the report gives no EcliptOS results on these measures. ¹

Ask any vendor or internal team proposing this pattern to identify the graph’s sources, the use cases in production, who governs the model, how changes are audited, and how contradictions are resolved. The report does not answer those questions for EcliptOS. ¹ The same production-controls lens applies to other orchestration pitches; see IBM’s AI orchestration pitch makes production controls the buying test.

The takeaway is not that every enterprise AI system needs a knowledge graph. It is that AI context depends not only on retrieving related records but also on how a company defines and maintains relationships between them. A graph is worth testing when those relationships matter to a specific decision and can be governed well enough to trust. Outcome data from an EcliptOS deployment would help test the approach; the report provides none. ¹

Source

¹ “McKinsey connects enterprise data through a knowledge graph for AI.” SiliconANGLE AI, October 9, 2026. https://siliconangle.com/2026/10/09/mckinsey-connects-enterprise-data-using-knowledge-graph-ai-neo4jdatatoknowledge/