The agent classifies every reply (positive, neutral, negative, OOO, opt-out) and acts: books meetings, answers simple questions, schedules follow-ups, processes opt-outs, and escalates what needs a human.
Classification
Replies are classified within seconds with a confidence score. High-confidence classifications trigger automated handling; low-confidence ones escalate with the agent's best guess attached.
Automated actions
- Positive + scheduling intent → calendar link or direct booking.
- Simple question → agent answers from your approved knowledge.
- Neutral "not now" → polite close and re-engagement scheduling.
- Opt-out → immediate suppression.
What always escalates
Complaints, complex questions, negotiation-adjacent topics, named-account replies, and anything the agent reads below its confidence threshold. Escalations carry the thread plus a suggested response.
Tuning
Sequences → Reply Handling. Tighten (escalate everything) while building trust; loosen as the agent's suggested responses prove out. Most teams reach 70–80% autonomous reply handling within a month.
