Agent Workflows & Submissions
Understand post-completion chains, interactive form steps, Conversation Review & Extraction, and where chatbot records are stored.
Agent Workflows & Submissions
An agent workflow connects a completed customer-facing agent to the next action. Configure it in Agents → Chain Builder for the agent that should start the workflow. A chain runs its steps in order and can contain up to eight steps.
What can a chain do?
Depending on the selected step and your chatbot's configuration, a chain can:
- Show a smart form and collect more information
- Send an email or SMS notification
- Hand the conversation to another conversational agent
- Run a private Conversation Review & Extraction step
- Show an end-of-chat feedback form
- Send a webhook, add a follow-up message, wait briefly, or reset the conversation
Some steps need additional setup, such as a notification destination or webhook URL. Check the step's configuration and test the workflow before relying on external notifications.
Forms pause a chain
When a chain reaches a smart-form step, it pauses so the customer can respond. After a successful submission, the workflow resumes at the next step. On supported messaging channels, the form is collected through a guided conversation rather than an embedded form card. If the customer does not complete the form, later steps do not run.
The form response is saved separately from the preceding agent's record. A chain does not combine a lead, booking, support ticket, survey response, or custom form into one submission.
Conversation Review & Extraction
This is an internal, chain-only agent: it does not take over the customer-facing conversation or send the customer a reply. Add it after an agent in that agent's chain to analyze the saved conversation and write a private structured review.
The review can include a short summary, intent, sentiment, resolution, urgency, follow-up, configured custom fields, and evidence quotes with confidence values. Set the review focus and fields in the review agent's configuration. It only has the conversation and completed-form context available to it; it cannot recover details the customer never shared. For reliable extraction, the customer-facing agent should ask for the information you want captured, and the reviewer should be configured to avoid unsupported assumptions.
If a review fails, its record is marked failed in Agent Outputs. You can retry that review using the saved conversation and the reviewer's current settings. A retry reruns the review; it does not replay the preceding customer-facing agent or other chain steps.
Where records appear
Open Submissions for the chatbot to review and export the relevant record type:
| Submissions section | What it contains |
|---|---|
| Leads | Lead-capture submissions |
| Bookings | Appointment requests and booking records |
| Support Tickets | Support escalations and ticket details |
| Form Responses | Smart-form and custom-form answers |
| Feedback | Survey and end-of-chat feedback responses |
| Agent Outputs | Private structured Conversation Review & Extraction results |
Agent Outputs can be searched, filtered by status, expanded to inspect extracted fields and evidence, opened alongside the source chat, and exported as CSV. Other records stay in their own section so teams can use the right workflow for each kind of customer data.
Before you retry a workflow:
Check the relevant Submissions section and any external destination first. A completed form or external notification is not undone if a later step fails. Retry a failed Conversation Review from Agent Outputs; do not resubmit the customer's form just to rerun the review.