How can we constrain Now Assist Virtual Agent to business-domain conversations only
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
2 hours ago - last edited 2 hours ago
We're evaluating Now Assist Virtual Agent for enterprise use cases (HR, IT, CSM, etc.) and are trying to understand the best practice for what I'd call conversation governance.
From what we've observed, the underlying LLM will happily respond to prompts that are unrelated to the configured business domain. For example, users can ask questions like:
- "Tell me a joke."
- General knowledge questions.
- Upload a random image and ask questions about it.
- Other non-business conversations that have nothing to do with the configured HR/IT/CSM experience.
While these responses are technically valid LLM behavior, our concern is around enterprise governance rather than safety.
We understand that ServiceNow provides Guardian / Guardrail capabilities (Prompt Injection, Offensiveness, Sensitive Topics, etc.), but those appear to be focused on AI safety and protection, not on limiting conversations to an intended business domain.
Our goal is something along the lines of:
"This assistant is intended only for HR / IT/CSM) related conversations. If a user asks an unrelated question, politely refuse the request and redirect them back to supported business topics."
The reasons include:
- Reducing unnecessary AI Assist/token consumption.
- Keeping conversations focused on the intended business purpose.
- Preventing Now Assist from becoming a general-purpose chatbot.
- Providing a more governed enterprise AI experience.
Are there any recommended OOB mechanism / patterns or best practices for implementing this type of conversation governance?
I'd appreciate any guidance, architectural recommendations.
#otto #NowAssist #VirtualAgent
