Prompt

How do I set up a support agent for refund and exchange handling with knowledge base connected workflows?

Artificial Intelligence · AI Agents / Ai agents1 observationLast seen Jul 19, 2026

Latest observation

Jul 19, 2026 · OpenAI APIWeb search: off

Here’s a practical way to set up a support agent for refunds and exchanges with knowledge base–connected workflows.

1) Define the agent’s scope

Start by deciding exactly what the agent should handle:

  • Refund eligibility checks
  • Exchange requests
  • Order status lookup
  • Return window validation
  • Policy questions from the knowledge base
  • Escalations to a human agent

Keep the scope narrow at first so the workflow is reliable.

2) Build the knowledge base

Create or connect articles covering:

  • Refund policy
  • Exchange policy
  • Return shipping rules
  • Product condition requirements
  • Timelines for processing
  • Exceptions and non-returnable items
  • Regional variations, if relevant

Make the articles:

  • Clear
  • Structured with headings
  • Easy to search
  • Updated regularly

If your platform supports it, tag articles by topic such as:

  • refunds
  • exchanges
  • returns
  • shipping
  • policy-exceptions

3) Connect the knowledge base to the agent

Configure the agent so it can:

  • Search KB articles before responding
  • Cite the policy source when relevant
  • Use KB content to answer policy questions
  • Fall back to escalation if the answer isn’t found

A good rule is:

  1. Check the KB
  2. Check order/customer data
  3. Run the workflow
  4. Escalate if needed

4) Create refund and exchange workflows

Set up separate workflows for each case.

Refund workflow

Typical steps:

  1. Confirm order number
  2. Verify purchase date and refund eligibility
  3. Check return window
  4. Confirm item condition and exclusions
  5. Determine refund method
  6. Initiate return label or approval
  7. Send confirmation to the customer

Exchange workflow

Typical steps:

  1. Confirm original order
  2. Check stock availability for replacement item
  3. Verify eligibility under exchange policy
  4. Determine price difference, if any
  5. Generate return and replacement steps
  6. Notify customer of next actions

5) Add decision rules

Use rules to automate common outcomes.

Examples:

  • If item is outside return window, deny refund and suggest escalation or store credit if allowed
  • If item is defective, route to exception handling
  • If replacement is out of stock, offer refund or waitlist
  • If customer has no order number, ask for email and date of purchase

6) Connect external systems

To make the agent actually do the work, connect it to:

  • Order management system
  • CRM/customer profile
  • Inventory system
  • Shipping/returns platform
  • Ticketing/helpdesk system

This lets the agent verify orders, create return requests, and log cases automatically.

7) Set guardrails

Add rules for what the agent must not do automatically:

  • Large-value refunds
  • Fraud-sensitive cases
  • Policy exceptions
  • Angry/high-risk customers
  • Legal/chargeback threats

Route those to a human supervisor or specialist team.

8) Write prompts and response templates

Provide the agent with:

  • Tone guidelines: polite, concise, empathetic
  • Required questions: order number, item, reason, condition
  • Escalation language
  • Standard refund/exchange messages

Example template:

I can help with that. Please share your order number and let me know whether you’d like a refund or an exchange. I’ll check your eligibility and next steps.

9) Test common scenarios

Run tests for:

  • Eligible refund
  • Ineligible refund
  • Exchange in stock
  • Exchange out of stock
  • Missing order number
  • Defective item
  • Policy edge cases

Verify the agent:

  • Uses KB content correctly
  • Follows workflow logic
  • Escalates when appropriate

10) Monitor and improve

Track:

  • Resolution rate
  • Escalation rate
  • Accuracy of eligibility decisions
  • Customer satisfaction
  • Most common refund/exchange reasons
  • KB gaps

Use those insights to update policies, workflows, and articles.


Recommended setup pattern

A strong architecture usually looks like this:

Customer message
Knowledge base search
Intent detection
Workflow execution
System checks (orders, inventory, CRM)
Decision or escalation
Response to customer


Example workflow logic

If customer asks for refund:

  • Search KB for refund policy
  • Ask for order ID if missing
  • Validate order date and item eligibility
  • If eligible, create refund case
  • If not eligible, explain reason and offer escalation

If customer asks for exchange:

  • Search KB for exchange policy
  • Ask for order ID and desired replacement
  • Check inventory
  • If available, create exchange request
  • If unavailable, offer alternatives

If you want, I can also help you with:

  1. a sample workflow diagram,
  2. a prompt template for the support agent, or
  3. a step-by-step setup for a specific platform.

Observation history

Unlock observation history

Create a free Obsurfable account to unlock historical responses, model comparisons, and deeper brand intelligence.

How did Obsurfable measure this prompt?

Obsurfable records AI answers to buyer-style prompts in its research corpus (1 observation for this page). Metrics are distributions over observations, not a single static ranking.

Which AI systems does Obsurfable collect answers from?

OpenAI, ChatGPT, Google, Gemini, Google AI Mode, Anthropic, Claude, Perplexity, Grok, DeepSeek, Mistral, Copilot, and Meta AI.