Refer complaints to regulated entities and monitor redress

1. Create the trigger

Purpose: Set up a new workflow to capture citizen complaint referral requests.

• Navigate to the Actions tab in the AI agent settings.
• Click + New workflow and click Message received as the trigger type.
• Set Name to: Refer Complaint
• Set Description to: Triggers when a citizen wants to file a complaint about a financial institution and have it referred for resolution.
• Add training phrases: “File a complaint”, “My bank won’t help me”, “I want to report my bank”, “Complain about a financial institution”, “Refer my case to the regulator”, “I need help with a dispute”

2. Send an opening message

Purpose: Reassure the citizen and invite them to describe what happened before asking structured questions.

• Add a Send message action under the trigger.
• Set the message text to:

Hi! I can help you file a complaint and refer it to the appropriate institution for resolution. Just tell me what happened – include the name of the institution and as much detail as you can – and I’ll take it from there.

3. Extract complaint details

Purpose: Use the LLM action to extract the institution, complaint category, and description from the citizen’s message, looping conversationally until all required fields are confirmed.

Capture the message

• Add a Set chat variable action after the Send message.
• Set Variable name to complaint_input and Value to _user_input.

Configure the LLM action

• Add an LLM action after the Set chat variable.
• Set Model to GPT-4o and Temperature to 1.
• Set Input text to {complaint_input}.
• Set Output variable to complaint_details.
• Paste the following system prompt into the System prompt field:

[Insert Code block here – paste the complaint details system prompt from the Notion source.]

Inside the Success branch

• Add a Set chat variable as the first action – Variable: complaint_details, Chat script: _.json_parse(complaint_details).
• Add a Branch action with the following condition:

[Insert Code block here – paste the branch condition from the Notion source.]

If true – follow-up needed:

• Add a Send message: {complaint_details["follow_up_question"]}.
• Add an Ask survey action – Variable: more_complaint, Type: Text, Skip: On.
• Add a Set chat variable – Variable: complaint_input, Value: {complaint_input} {more_complaint}.
• Add a Jump to action back to the LLM action to re-run extraction.

Else – all complaint details collected:

• Add Set chat variable actions to unpack each field:
  • complaint_description = complaint_details["complaint_description"]
  • institution = complaint_details["institution"]
  • complaint_category = complaint_details["complaint_category"]
  • prior_contact = complaint_details["prior_contact"]

Inside the Error and Timeout branches

• Add an Ask survey action – Variable: complaint_input, Type: Text, Required: On, Skip: Off.
• Question: I didn't quite catch that. Could you describe the complaint again?
• Add a Jump to action back to the LLM action.

4. Collect citizen contact details

Purpose: Gather the citizen’s personal details so the referral can be filed and a follow-up sent.

• Add a Send message after the complaint extraction Else branch: Thank you. So we can file this on your behalf, please share your full name, national ID number, email address, and phone number.
• Add a Set chat variable – Variable: citizen_input, Value: _user_input.
• Add a second LLM action – Input: {citizen_input}, Output variable: citizen_details.
• Paste the following system prompt:

[Insert Code block here – paste the personal details system prompt from the Notion source.]

• Inside the Success branch: parse with _.json_parse(citizen_details), check the follow_up_question branch condition, and loop with more_citizen appended to citizen_input – same pattern as Step 3.
• In the Else branch, unpack: full_name, id_number, email, phone.

5. Collect supporting documents

Purpose: Give the citizen an opportunity to upload evidence before the referral is submitted.

• Add a Send message: Do you have any supporting documents – statements, screenshots, or letters – you'd like to include?
• Add an Ask survey action – Variable: evidence, Type: Attachment, Skip: On.

6. Confirm and submit the referral

Purpose: Let the citizen review everything captured, then create a ticket and push the referral to the institution via API.

• Add a Send message after the Ask survey:

Here’s what I have – Name: {full_name} | ID: {id_number} | Institution: {institution} | Category: {complaint_category} | Details: {complaint_description}. Shall I refer this now?

• Add Quick reply options: “Yes, refer my complaint” | “No, let me correct something”
• Add a Set chat variable immediately after – Variable: referral_choice, Value: _user_input.
• Add a Branch action on referral_choice:
  • If _.lower(referral_choice) == "no, let me correct something": capture correction with Ask survey into complaint_input and Jump to action back to the complaint LLM action.
  • If _.lower(referral_choice) == "yes, refer my complaint":
    • Add a Create ticket action – Subject: Complaint referral: {full_name} – {institution} | Description: Category: {complaint_category} | Institution: {institution} | ID: {id_number} | Prior contact: {prior_contact} | Details: {complaint_description} | Tags: complaint-referral, {complaint_category}
    • Add a Send API request to push the referral to the institution’s complaints endpoint. Pass {full_name}, {id_number}, {email}, {phone}, {institution}, {complaint_category}, {complaint_description}. Set Output variable to referral_response.
    • Add a Set chat variable – Variable: referral_response, Chat script: _.json_parse(referral_response).
    • Add a Branch: if referral_response["id"] != none, send confirmation:

Your complaint has been referred to {institution} with reference {referral_response[“id”]}. Our team will follow up and let you know as soon as there is an update. Thank you for reporting this, {full_name}.

• Otherwise, send an error message and route to Human handoff.

7. Test the flow

Purpose: Validate the full complaint intake and referral sequence using the AI agent tester before going live.

• Open the AI agent tester panel on the right side of the workflow builder.
• Type “I want to file a complaint” – confirm the opening message appears and the LLM asks about the institution.
• Provide a complaint description without naming the institution – confirm the LLM asks a conversational follow-up.
• Provide all complaint details – confirm the citizen contact section triggers.
• Provide all contact details – confirm the evidence upload step appears.
• Choose “Yes, refer my complaint” – confirm the ticket is created and the API fires with the correct reference.
• Choose “No, let me correct something” – confirm the flow loops back to the complaint extraction correctly.
• Simulate a failed API response – confirm the flow routes to Human handoff with an empathetic message.