Running an online store across five countries with one person handling customer service is a math problem that does not work. The inbox fills up in five languages. "Where is my order?" arrives in Italian, Dutch, French, Spanish, and German, often on the same morning.

The operator was spending two to three hours a day copy-pasting into Google Translate, writing a reply, translating it back, and sending it. For the same three questions, every single day.

Italian Dutch French Spanish German

What we built

We integrated an AI agent directly into their Wix store. When a customer opens a chat, the agent detects the language in the first message, pulls the order status from the Wix API using the customer's email, and responds.

If the order is in transit: it gives the tracking link, the expected delivery window, and the carrier name, all in the customer's language. If it is delivered and the customer says they did not receive it: the agent escalates immediately with a summary of the conversation already attached for the operator.

No menu. No "type 1 for order status." Just a response in the customer's language within 30 seconds of them opening the chat.

What changed

70%
Tickets handled without human involvement
<5 min
Response time, down from 4 business hours
40 min
Daily customer service time, down from 2.5 hours

The operator now handles the 30% that actually needs a human: complaints about damaged products, return requests, unusual situations. The agent flags and summarises those before passing them over, so there is no reading back through a conversation to understand what happened.

The faster response time had an indirect effect that was not part of the original brief: fewer follow-up messages. When customers get an answer in 30 seconds, they do not send three more messages asking why nobody responded. The inbox volume dropped alongside the handling time.

What the agent does not do

Returns and refunds stay with a human. Return policies vary by country and the legal requirements differ across the EU. The agent detects return requests and hands them off with a draft response, but a person reviews and approves before anything goes out. That is intentional and it should stay that way.

The agent also does not have access to payment systems. It can tell a customer their order shipped. It cannot issue a refund or modify an order. Those actions stay behind a human approval step.

Where this works best

The bigger the language gap between your team and your customer base, the more value this produces immediately. If you are a German company selling into Spain and France, your customer service team is translating every conversation. That is pure overhead and it compounds with scale.

This setup works on Wix, Shopify, WooCommerce, or any platform with an order API. The agent logic is the same. Only the data source changes.


If you are selling across more than two countries and your customer service inbox is a translation exercise every morning, we can fix that. Usually in less time than you would guess.