Hi Gabriel,
This is a very common challenge in large Copilot deployments.
Even with separated KBs, semantic similarity can still cause cross-brand retrieval if:
- terminology overlaps
- agents share queues/workspaces
- articles have similar operational language
The best results I've seen come from combining:
- queue segmentation
- strict KB scoping
- metadata/tag standardization
- minimizing duplicated operational wording
Another important factor is reducing "generic" article titles/content that semantically match multiple brands.
Some teams also isolate:
- intents
- routing
- and Copilot context
As early as possible in Architect before the interaction reaches the agent.
Pure KB separation alone is often not enough in shared-agent environments.
Regards!
------------------------------
Lilian Lira
Services and Developer Manager
------------------------------