- 01People & roles
- 02Work & resources
- 03Rules & relationships
- 04Connected context
What is a business ontology?
A business ontology is an explicit model of the concepts a business uses and the relationships between them. It might describe a location, a service, a booking, an employee role, a supplier and an approval. It establishes what those things mean in the context of the system.
For example, a service request belongs to a customer, takes place at a location and requires a resource. A particular role can approve a change. A policy applies to one type of service but not another. These relationships help distinguish a relevant answer from an action the business can actually take.
An ontology does not need to model the whole organisation before it becomes useful. Begin with the concepts and relationships required by one workflow, then expand where another capability needs them.
Search finds material. Relationships supply context.
Semantic search can retrieve text that is related in meaning to a request. This is useful when employees use different terms for the same issue or when the relevant knowledge is spread across documents. Retrieval can bring an agent the passages it needs to consider.
A structured business model answers another set of questions: which location is this request about, which version of the policy applies, who owns the task and what other work depends on it? Document retrieval and structured queries can work together.
Consider a request to change a booking. A relevant policy explains the rules, but the actual booking, current availability and the requester's authority must come from appropriate business systems. Finding a plausible passage is not evidence that the change is permitted or feasible.
Keep the source attached
Digitisation should preserve a route back to the original material. A record extracted from a paper form should retain its source and the status of uncertain fields. A statement drawn from a conversation should remain distinguishable from an approved operating rule.
Record useful provenance: where the information came from, when it was updated, what it applies to and who can approve a correction. These details help both agents and people decide whether the information is suitable for the task.
Conflicting sources need a resolution process. The newest document is not automatically the authoritative one. Define which system or owner has authority for each kind of fact, and expose uncertainty rather than silently combining incompatible versions.
Build the smallest useful model
Take a workflow such as producing staff guidance for a new service. Start with service, location, role, procedure, learning asset and approval. Define how they relate: a procedure applies to a service, a role needs the guidance, an asset depends on a procedure version and an owner approves release.
Give important records stable identifiers. Agree terms with the people doing the work. Test the model against real questions and exceptions before adding more concepts. A large diagram that nobody maintains will not provide dependable context.
The implementation may use relational records, a graph, documents or a combination. The design requirement is consistent meaning and retrievable relationships; the storage choice should follow the access patterns and operational needs.
Shared understanding still needs access controls
Knowing that two records are related does not give an agent permission to read both. Apply access rules when retrieving information and when calling tools. A model of responsibilities helps describe authority, but the surrounding system must enforce it.
Evaluate the knowledge layer with realistic questions: can it find the correct policy, distinguish two locations, identify an outdated source and decline a request outside the user's access? Check the evidence returned with the answer as well as the answer itself.
In Hello OS, ontology and shared knowledge are part of the foundation for connected agent teams. The aim is to make context reusable while keeping its meaning, ownership and limits clear.