prompts.chatprompts.chat
Login
All classifiers
Decision chainRefunds3 steps · 5 endpoints

Refund request resolution

Confirm a refund is requested, check the request is complete, then recommend a decision under your policy.

Scenario paths

Pick a predefined path to highlight it through the chain. No model is called; outcomes are illustrative.

Steps
  1. Refund request detection
  2. Refund information check
  3. Refund policy eligibility
Endpoints
  • Prepare refund for approval
  • Explain the policy
  • Ask for the missing details
  • Billing specialist review
  • Back to support triage
What each endpoint means

  • Prepare refund for approval · Automated

    Code checks amount limits and earlier refunds. Amounts above your limit wait for a person to approve.

  • Explain the policy · Stop

    Reply from a reviewed template, offer alternatives such as a plan change, and include how to appeal.

  • Ask for the missing details · Automated

    Send the template for what is missing. The flow runs again when the customer replies.

  • Billing specialist review · Human review

    A person decides, with the classifier answers attached as context.

  • Back to support triage · Handoff

    Not a refund request; handled as an ordinary support message.

    Continues in Support inbox triage

Steps

  1. Refund request detectionNoulEntry

    Confirm the customer is asking for money back.

    Question, inputs and notes

    Asks:Does the customer ask for money back, a refund, a reversed charge, or a credit for a payment?

    Receives

    • Inputmessage
  2. Refund information checkChoice

    Make sure the charge and the reason are known before deciding.

    Question, inputs and notes

    Asks:What, if anything, is missing from this refund request that `policy_requirements` needs before a decision can be made?

    Receives

    Branches

    • Complete: Code attaches the matched charge and computes days since purchase.
  3. Refund policy eligibilityChoice

    Recommend a decision under the written policy.

    Question, inputs and notes

    Asks:Under `policy`, how should this refund request be decided?

    Receives

About this chain

Goal and input

Ask customers only for what is missing, apply the policy consistently, and keep every money-moving action behind code checks and approval.

The message, the customer's charges and values computed in code, and the policy text.

message
The customer's message.
account_charges
Recent charges, looked up in code.
refund_policy
The refund policy, as written.
days_since_purchase
Computed in code for the matched charge.
How it runs
  • Charge lookup, day counts and amount limits are computed in code and passed in the state. The classifiers never decide amounts.
  • Every path that could move money ends at a code check or a person. None of them issues a refund on a model answer alone.
Sources (4)

Primary documentation this chain builds on.