Skip to main content

Room assignment-source tracking gap (needed for AI agent metrics)

Context​

Data/BI needs to classify how a room participant got assigned (e.g. Get Chat vs manual pick from the room page). Today no dedicated field exists. The urgency is the AI agent metrics report — in particular the unintended room metric, which detects rooms that became assigned through actions outside the AI agent's configuration (e.g. get chat, manual chat). Discussion thread Jul 16 → Wed, involving Puji Triwibowo, Lingga Mahesa Riella, Dandi Pangestu, Burhanudin Hakim, Dimas Fauzi Hidayat.

Source thread: https://chat.google.com/room/AAAAEkQek1I/_jRzcHog2Bw/_jRzcHog2Bw?cls=10

Notes​

  • Puji Triwibowo (Jul 16): asked whether manual assignment (via get chat or direct pick on the room page) leaves any log / flag recording "assigned via what". Needed by the data team for classification.
  • Dandi Pangestu: no dedicated assignment-source data exists — it is currently only appended as a message; the usual way to check is the system-type message bubble in the room.
  • Burhanudin Hakim: confirmed "belum ada" — adjustment needed first. Round-robin has its own log but in a different table; manual assignment is only visible via Grafana etc.
  • Granularity question (Burhanudin): if a room has >1 assignee, is the marker on the room or the participant? Puji clarified the need is per participant — "Participant A was assigned via what".
  • Chat Panel side also wants this data for internal analysis and debugging (Puji asked, Burhanudin confirmed "iya").
  • Lingga Mahesa Riella listed the expected assignment types:
    • AAA
    • Round robin
    • Assign by bot (bot can target a specific user)
    • Get chat
    • Handover
    • Assign manual dr room
    • Contact assignee (including the new one coming)
  • Dimas Fauzi Hidayat (Jul 16, 4:12 PM): BI is building the report now; it will be used for a weekly Monday checkpoint with stakeholders including C-level (scheduled ~2 weeks out). For now it ships as-is with existing data, but within 2 weeks the report must include the additional parameter from the chat team.
  • Wed: Dimas followed up; Burhanudin said it had not been worked on yet and would pick it up. Lingga re-confirmed the type list above as final ("fix itu ya" / "Iyes mas").
  • Latest (Wed +): Dimas followed up again with Burhanudin — status still TBD.

Decisions / changes​

  • Assignment-source categories are fixed to Lingga's list above (confirmed by Burhanudin and Lingga, Wed).
  • Tracking granularity is per participant, not per room (per Puji, Jul 16).
  • Burhanudin Hakim picks up the chat-side implementation (Wed). No ticket/date recorded — TBD.

Follow-ups​

  • Implement per-participant assignment-source tracking in chat — Burhanudin Hakim. (Chat-squad engineering work; not actionable in this repo.)
  • Confirm deadline against the BI report checkpoint (~2 weeks from Jul 16) — Dimas Fauzi Hidayat. Tracked as OQ #5 in the Phase 1 PRD (deadline TBD until Dimas + Burhanudin agree a date).
  • Record this as a cross-squad dependency (chat ⇄ BI/data) — added to chatbot/delivery/roadmap.md ## External & cross-squad dependencies (2026-07-27). No timeline.md exists for this initiative, so the roadmap is the only mirror.
  • Add the unintended-room metric's dependency on assignment-source data to the impact-report PRD — Phase 1 PRD §14 dependency row + dependency-graph edge, PRD v1.4 (2026-07-27). The RFC was left untouched: it scopes the Phase 1 report only and does not build the unintended-room metric.