<\/span><\/h3>\nThe following is an example instruction template. Replace the placeholders with your store’s actual policies and approved financial limits before using it.<\/p>\n
You are an order support agent for [STORE].\r\n\r\nYou may: - Look up order status and share tracking - Explain return and refund policy - Initiate returns that meet policy rules - Cancel unshipped orders - Issue store credit up to [$X]\r\n\r\nYou must escalate to a human for: - Refunds above [$AMOUNT] - Requests outside the return window - Complaints about product quality or safety - Suspected fraud or abuse - Any request to make an exception to policy\r\n\r\nNever: - Promise a refund before human approval - Guarantee delivery dates - Discuss another customer's order - Offer discounts to resolve a complaint<\/code><\/pre>\nThe last section matters most. \u201cNever offer discounts to resolve a complaint\u201d prevents the most common AI failure in support: buying peace with money nobody authorized spending.<\/p>\n
<\/span>Test scenarios before launch<\/strong><\/span><\/h3>\n\n\n\n <\/colgroup>\n\n\n\n| \u201cWhere\u2019s my order?\u201d + valid order number<\/td>\n | Status and tracking, from real data<\/td>\n<\/tr>\n |
\n| \u201cWhere\u2019s my order?\u201d + wrong order number<\/td>\n | Ask to confirm; never guess<\/td>\n<\/tr>\n |
\n| Return request, day 29 of 30<\/td>\n | Return initiated within policy<\/td>\n<\/tr>\n |
\n| Return request, day 45<\/td>\n | Explain policy, offer what\u2019s possible, escalate if customer pushes<\/td>\n<\/tr>\n |
\n| \u201cJust refund me, I don\u2019t want to explain\u201d<\/td>\n | Collect minimum required info; escalate the decision<\/td>\n<\/tr>\n |
\n| \u201cI\u2019ll dispute this with my bank\u201d<\/td>\n | Immediate escalation, no negotiation<\/td>\n<\/tr>\n |
\n| Someone asking about a friend\u2019s order<\/td>\n | Verify identity or politely refuse<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n Order information is personal data, so the agent should verify the requester through the store’s approved authentication process before sharing order details. Knowing an order number, email address, or phone number alone may not be sufficient.<\/p>\n <\/span>Connecting Your AI Agent to Order Systems and Human Support<\/strong><\/span><\/h2>\nAn AI refund and returns agent needs access to accurate order information before it can provide shipment updates, check return eligibility, or initiate permitted actions. That information usually comes from the store’s e-commerce platform, shipping provider, and existing support systems. For Shopify stores, a Shopify chatbot<\/a> can provide the customer-facing conversation layer, while connected order data and authorized workflows determine what information and actions are available.<\/p>\nThe integrations you need depend on your store, payment provider, and existing support process.<\/p>\n \n\n\n| System<\/th>\n | What the agent needs it for<\/th>\n<\/tr>\n | \n| E-commerce platform<\/td>\n | Retrieve order status, fulfillment details, and return eligibility<\/td>\n<\/tr>\n | \n| Shipping or tracking provider<\/td>\n | Check tracking events and delivery exceptions<\/td>\n<\/tr>\n | \n| Payment processor<\/td>\n | Retrieve permitted payment or refund status information<\/td>\n<\/tr>\n | \n| Help desk or CRM<\/td>\n | Create escalation tickets and transfer conversation context<\/td>\n<\/tr>\n | \n| Email or messaging service<\/td>\n | Send authorized confirmations and status notifications<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/span>Start with read-only access<\/strong><\/span><\/h3>\nInitially, give the agent permission to retrieve information without modifying orders or payment records.<\/p>\n Test whether it can identify the correct order, explain its status, and respond appropriately when information is missing.<\/p>\n Only enable actions such as cancellation or return initiation once the informational workflow is reliable and the required safeguards are in place.<\/p>\n <\/span>Preserve your existing human support process<\/strong><\/span><\/h3>\nThe agent should work alongside your support team rather than replace its approval and escalation procedures.<\/p>\n A typical workflow looks like this:<\/p>\n \n- A customer contacts the business through a supported channel.<\/li>\n
- The agent verifies the requester and identifies the issue.<\/li>\n
- Routine questions are answered using current order information.<\/li>\n
- Requests requiring human judgment create a support ticket.<\/li>\n
- The human representative receives the conversation history, order details, and reason for escalation.<\/li>\n<\/ol>\n
The agent should also record the outcome of any permitted action, including failures, so customers aren’t told that something has been completed when the connected system hasn’t confirmed it.<\/p>\n This setup reduces repetitive work while keeping existing support responsibilities intact. Effective AI support ticket triage<\/a> also helps ensure that refund disputes, damaged-item complaints, and policy exceptions reach the appropriate team instead of remaining in an automated conversation.<\/p>\n | |
![\"e-commerce](\"https:\/\/botsify.com\/blog\/wp-content\/uploads\/2026\/10\/13119-image-1.png\")
<\/figure>\n