Skip to content

Conversational Policy Advisor

Questions or comments?

Post them as a comment on the tracking issue -- requires a GitHub account with access to this repo.

Scope

A thread-based, conversational compliance advisor used internally by HR Operations (policy owners, compliance analysts, other HR decision-makers) to ask questions about existing policy and applicable regulation in plain language. Delivered inside the Nova portal on the ServiceNow Service Portal as a chat-style surface, scoped to HR Operations users only — this is deliberately not an employee- or people-manager-facing self-service tool, since published policy content also determines how other Nova agents behave at runtime.

The agent retrieves grounded content from the organization's published Knowledge Base and answers with a citation reference to the specific policy/KB article. Every turn is persisted back to ServiceNow so the conversation survives interruptions and is available for internal audit (see the auditable-conversation-history Feature for the persistence contract itself).

This feature is AI-powered end to end for the query path: the agent autonomously retrieves, reasons over, and answers policy questions without a human in the loop for a routine query. It answers only from what's already published — it cannot edit or publish a policy itself. A roadmap item ("policy edit suggestions in chat mode") would let the agent propose an edit, but that's a design in progress, not shipped, and out of scope here; any such suggestion would still have to route through the human-publish gate owned by native-policy-editor-collaborative-authoring.

Known, confirmed gap (per July 9 sync with Ravindra Kumar, carried over from the source PRD): when the advisor cites a government regulation rather than an internal policy (e.g. FLSA, FMLA), it surfaces a reference number in the response text today, but not a clickable link back to the source regulation text. That capability depends on the Codex Curator data layer tracking stable source URLs per regulation — see ../../nova-codex-curator/concept.md and its versioned-auditable-source-tracking Feature. Not fixable from this Feature Set alone.

Value

Gives HR a fast way to sanity-check a policy question, or an interaction between two policies, before making a change or publishing — rather than manually cross-referencing every related document — and produces a defensible, auditable record of that reasoning for internal audit.

Acceptance criteria

  • An HR Operations user (policy owner, compliance analyst, or other HR decision-maker) can open the advisor chat surface and type a plain-language policy or compliance question.
  • The agent creates a new thread or resumes an existing one, retrieves grounded content from the published Knowledge Base, and returns an answer inline in the chat thread with a citation reference to the specific policy/KB article it drew from.
  • The user can ask a follow-up question in the same thread; there is no explicit accept/reject control on the answer, since the advisor only answers and never takes action on this path.
  • Access to the advisor chat surface is scoped to HR Operations roles only — not exposed to general employees or people managers.
  • When the advisor's grounding for a jurisdiction falls to Tier 2 (configured-source fallback) or Tier 3 (open web), per the three-tier grounding model defined in the Codex Curator PRD, the answer is still produced, citing the web source it actually used — this Feature Set does not require Tier 1 coverage to answer, only to answer well-grounded.
  • Citing a government regulation surfaces a reference number in the response text; a clickable link to the source text is explicitly out of scope for this Feature Set until Codex Curator ships stable source-URL tracking.
  • Every question and answer in a thread is persisted to ServiceNow per the auditable-conversation-history Feature's contract, with no gap in coverage (i.e. no advisor turn should exist only in the chat UI's local state and not in ServiceNow).