
How ranking works
- Audit events first. IrisAgent reads Zendesk
MacroReferenceandAgentMacroReferenceevents on tickets it already pulls, and ranks macros that were actually applied on similar tickets. - Tags as fallback. For older tickets without audit events, IrisAgent matches tags. Unique per-macro tags help but are not required.
- No extra LLM call. Ranking adds no model request.
Optional: add unique tags
To improve ranking on tickets from before audit capture, give each Zendesk macro an action that adds a unique tag, one tag per macro.