Conversation lifecycle
How a Zild conversation moves from ingestion to AI handling, human handoff, evaluation, reporting, and follow-up.
Lifecycle overview
A conversation starts when an external event enters Zild or when an authorized integration creates one proactively. From there, the configured agent, channel, queue, template, and checklist determine how the interaction is handled.
| Stage | What happens | API evidence to provide |
|---|---|---|
| Ingest | Zild receives email, audio, Teams transcript, proactive WhatsApp request, or phone-call request. | Channel, integration type, source identifier, participant, timestamp, and payload content. |
| Context | Zild links the event to the tenant, bot/agent, queue, checklist, user, and conversation type. | botId, channelId, queueId, checklistId, conversationType, and variables when needed. |
| Action | The configured AI agent can respond, start a workflow, create a record, or leave the item for human review. | Approved templates and trusted metadata instead of free-form operational decisions. |
| Handoff | If automation should not continue, the conversation remains visible to a queue or user. | liveAgent, queue assignment, user ownership, and conversation status. |
| Evaluation | Zild Insight or Coach can score, summarize, classify, and flag errors or low-confidence outcomes. | Checklist, transcript/audio/message content, call status, duration, sentiment, and evaluation status. |
| Follow-up | The platform can drive tasks, reports, CRM activity, or further contact depending on configuration. | Zild conversation ID, external IDs, tags, and source-system references. |
Records to store
Store the returned conversation.id whenever an endpoint returns a conversation object. For voice bridges, also store callSid. For imports, store the source file or transcript identifier in your own system so support teams can reconcile payloads later.
What not to send
Do not send API keys in payload bodies. Do not include secrets in contentVariables. Avoid sending data the configured agent does not need. When sensitive data is required, prefer server-to-server calls and keep logs redacted.