Bot-vs-Human Traffic Split (A/B)
Let a Chatbot Admin route only a configured percentage of incoming chats to the chatbot
and auto-assign the remainder to human agents — so bot quality can be validated on real
customer traffic before committing to full automation. Built by reusing the existing
incoming-message routing decision (process_incoming_message_with_resolve) and the
existing human auto-assignment path (SendMessageAutoAssignAgentWorker), adding a
per-channel split percentage and per-conversation variant tagging for measurement.
QA Lane
Lane B — keeps a human QA gate. No Lane-B trigger is present (a bot-vs-human routing experiment), but no E2E test specs exist for this initiative yet, so the Lane-A entry bar (100% E2E, spec-mapped coverage) is unmet — Lane B by default. Classified 2026-06-29.
Phases
| Phase | Goal | Status | Epic |
|---|---|---|---|
| Phase 1 — Per-Conversation Split (Confluence) | Per-conversation random split (X% bot / rest human) on a single channel, self-serve config, variant tagging + measurement | Draft | TBD |
| Phase 2 — Sticky Per-Customer Split | Deterministic per-unique-customer bucketing (a returning customer always gets the same arm) | Planned | TBD |