When an AI agent is asked why profit fell last quarter, it can report the correct figure and still blame the wrong cause. Governed semantics make the number accurate, but the explanation depends on how the business works, and that knowledge is rarely recorded anywhere an agent can read. A context graph makes it explicit, so AI agents can explain why a metric moved.
A context graph connects business entities, metrics and concepts through typed relationships that carry rules, conditions and metadata. Some enterprise implementations, such as Kyvos’ business context graph, also model causal links and weights, so an agent can tell which relationships drive a result and which matter most. At query time, AI agents use these connections to identify relevant data and see how the pieces fit together.
The term context graph is still evolving, and vendors use it for different kinds of information, including data lineage, permissions and decision history. In analytics, a context graph focuses on business measures: which KPIs drive others, in which direction and under which conditions.
Enterprise AI needs a context graph because explaining a business result depends on relationships that raw data and metric definitions do not record.
Consider a retailer whose profit falls while sales rise. An AI agent can read both metrics correctly and still miss that discounts are reducing profit, that mass-market customers respond to discounts and that premium customers respond little. Without these relationships, the agent can recommend extending the promotion to every segment, against pricing policy.
A context graph records each relationship with its conditions attached, so the agent limits the promotion to the segment that responds.
A context graph comprises nodes, which represent what matters to a task, decision or action, and the relationships that connect them. For enterprise analytics, nodes fall into four types:
Relationships are the core of the graph. They record how one node influences, owns, uses or depends on another, and each one carries the rules and conditions that decide when it applies. Here’s how:
Nodes and relationships also carry:
A retail example: profitability depends on sales, profit, discounts and costs, and each arrow shows how one metric affects another
| Ontology | Knowledge graph | Context graph | |
|---|---|---|---|
| What it is | A formal model of a domain’s concepts, properties and constraints | A graph of real entities and facts, structured by an ontology | A graph of relationships with the rules and conditions that apply to them |
| Question it answers | What is a customer, and how can it relate to other concepts? | What is connected to what? | What drives this result, and what applies here? |
| How it reasons | Formal logical inference from defined rules and constraints | Traversal across connected entities and facts | Following drivers and conditions from a result back to its causes |
| Example | Every customer belongs to one segment, and a discount applies to a product | Customer 1042 is in the premium segment and bought product B at a 10% discount | Discount amount reduces profit, and its effect on sales depends on customer segment |
| Where is it applied | Implemented through a knowledge graph or a context graph | Record-level analysis, such as linking shared accounts to expose a fraud ring | Enterprise AI analytics on governed metrics, such as explaining why profit fell and what to do next |
Kyvos’ business context graph builds on Kyvos semantic models. It connects governed metrics to business entities, concepts and external entities through typed relationships, such as ‘has primary’, ‘decomposes into’, ‘segmented by’, ‘positively influenced by’, ‘negatively influenced by’, ‘belongs to’ and semantic relationships that teams name with verbs, such as ‘buys or sold in’. Relationships carry rules and weights, and influence relationships form causal paths across multiple semantic models.
AI agents assemble context for each query from semantic models, the business context graph and integrated policies, procedures and documentation. Governance is built into every component, and every metric in a causal path carries one approved definition, so AI answers stay accurate and consistent.
To see how business context graph works for enterprise AI, talk to a Kyvos expert.