Write the source before choosing the software
Create one approved policy source with clear eligibility, deadlines, exceptions, refund timing, and contact routes. Link related policies instead of copying the same rule into several pages. Add an effective date and an owner. If the policy is ambiguous for staff, a bot will make the ambiguity faster.
What the tool should do
A useful knowledge base supports search, version history, clear article ownership, analytics on failed searches, and a way to flag outdated content. If customers ask through chat, the assistant should show the policy source and recognize when it lacks enough details to decide eligibility. A knowledge base and an order system are different: the article explains the rule; account data confirms the customer’s case.
Compare realistic options
Zendesk’s Help Center fits teams already operating in its service workspace. Intercom supports help center content used by its customer service tools. eDesk offers a knowledge base for ecommerce support. ProxyAI’s Bot Knowledge can ground chatbot answers in approved documents and site pages. Each option has different publishing, access, and analytics features; test ownership and update workflows before importing everything.
Build an exception path
The customer-facing answer should state the general rule, identify what information is missing, and give a clear next step. It should not promise a refund just because a page mentions one. Route damaged items, late deliveries, chargebacks, and policy exceptions to an agent with the conversation context.
A return-policy article that helps without overpromising
A useful article answers the general rule first, then names the details that change the result. For example: “Unworn items may be returned within the stated window. Final-sale products and damaged items follow separate rules. Have your order number ready so support can check eligibility.” Replace the example with your real policy; never copy it as a legal rule.
Do not let an assistant infer eligibility from a general article alone. It needs an authorized order lookup and explicit decision rules to evaluate a specific purchase. Otherwise, it should explain the policy, ask for the required information, and route the case to a person. Review failed searches after each policy change.
ProxyAI’s Bot Knowledge is designed to retrieve answers from approved content, with exact and meaning-based search, spreadsheet support, scheduled refresh, and OCR for scanned PDFs documented in its feature comparison. That can make existing return-policy material easier to use in chat; it does not decide whether an exception should be approved. Keep the policy source current and require human review for edge cases.