AI conversations
Start an outbound AI call or chat from an automation, then wait for the outcome — or watch the conversation and react while it runs.
The AI conversation step starts one outbound conversation with a flow or agent: a phone call, or a chat over web chat, SMS or WhatsApp. The automation then continues with the outcome.
Waiting for the outcome
- Sync — the step waits for the conversation to finish, then the automation continues.
- Async — the run is suspended while the conversation takes place and resumes when it has finished. Suited to calls that may take minutes, or conversations that stretch over hours on WhatsApp.
Watching it live
A running conversation can be monitored from the automation, and its progress used in conditions:
| Field | Type | What it holds |
|---|---|---|
| Last customer message | text | The most recent thing the other side said |
| Last assistant message | text | The most recent thing the AI said |
| Transcript so far | text | Everything said up to now |
| Duration | number | Seconds since the call was answered |
| Message count | number | Messages exchanged so far, in total, by the customer and by the AI |
| Status | text | The call or chat status |
| Answered | yes/no | Whether the call was picked up |
| Human support requested | yes/no | Chat channels only |
| Structured outputs | object | Usually filled in only after the conversation ends |
Conditions compare these with operators such as contains, equals, matches regex, greater than or is set.
Uses
- Follow-ups — a missed-call automation that sends a WhatsApp message and waits for the reply.
- Testing — AI-to-AI conversations against a flow, graded on the outcome. See Evaluations and benchmarks.
- Escalation — watch for a phrase in the transcript and alert a team while the call is still going.
Note
Each conversation is billed like any other call or chat, in addition to the automation run.