First-level support agent for a Shopify shop
An AI agent that answers customers on WhatsApp and email in their own language, follows the shop’s support playbooks, looks up live orders through a Shopify MCP integration, and runs in autopilot or assisted mode per conversation.
The problem
A specialist online shop handled first-level support by hand across WhatsApp and email: order status, returns, product questions, in several languages. The same questions came back every day, response times slipped at peak, and the quality of an answer depended on who was on shift.
What we built
We built an agent that reads each incoming message, loads the applicable standard operating procedure from a knowledge base built from the shop’s own past conversations, looks up order and product data live through a Shopify MCP integration, drafts a reply, and has two independent LLM reviewers check it before anything goes out. Every conversation runs in autopilot (the agent answers itself once the customer stops typing) or assisted mode (it drafts, a colleague approves from the inbox). Staff steer it with notes inside the messaging platform and from an operator UI, and an evaluation harness scores every change against real conversations before release.
What changed
Routine questions are answered within minutes, in the customer’s language, with the same wording every time. The support team reviews drafts instead of writing from scratch and keeps full control over which conversations the agent handles alone. Every reply is traceable to the playbook and the order data it used.
Outcomes
- Answers in the customer’s language on WhatsApp and email
- Two reviewers check every reply before it is sent
- Autopilot or assisted mode, chosen per conversation