Shopify helpdesk automation

What the Shopify Admin API exposes

An AI agent connected to Shopify via OAuth can perform the following with the standard scopes:

• read_orders, look up order status, line items, fulfillment, shipping address.
• read_customers, fetch customer purchase history, last order date, lifetime value.
• read_products, product details, variants, current inventory, pricing.
• Write scopes (write_orders, write_customers), make refunds, modify orders, update customer notes. Granted on demand for agents that perform actions.

Conveya uses only the read scopes by default. Write actions require an explicit per-agent permission.

Common automation patterns

Order status answers

The most-requested automation. The agent receives a customer's email or order number, looks up the order in Shopify, and responds with the current status, expected delivery, and tracking link.

Refund and return handling

The agent gathers context (order, reason, customer history), proposes a resolution, and either executes the refund directly or routes to a human for approval, depending on the merchant's policy thresholds.

Product availability and recommendations

Inventory queries are answered live so customers never see stock that has already sold out. Recommendation prompts use the product catalog plus the customer's past orders.

Pre-purchase questions

Shipping times, return policy, sizing, answered from a knowledge base that also indexes the merchant's policy pages and FAQs.

Compliance and data handling

Shopify requires every published app to implement three GDPR webhooks: customers/data_request, customers/redact, and shop/redact. Apps that do not persist customer data can acknowledge these and return, no actual deletion is needed because nothing was stored. Apps that cache or persist data must delete on receipt.

Conveya does not persist customer or order data. Every Admin API call is fetched on demand to answer the question and discarded after the response.

Build vs buy

Building a Shopify helpdesk agent in-house is feasible, Shopify's API is well-documented and the LLM tool-calling protocols are stable. The effort is in everything around the agent: OAuth flow, token refresh, multi-tenant credential encryption, conversation memory, human handoff UI, GDPR compliance, multilingual support, observability, and ongoing prompt iteration. A well-built proprietary solution typically takes 6-9 months of one engineer's time before reaching feature parity with off-the-shelf tools.