Gorgias
95% solved, then failing every case
Topped Toys, a Canadian Shopify store, had the Gorgias team set up its AI support agent. The setup took several weeks. The agent followed logic the vendor wrote, including editing and cancelling orders, and answered customers on its own. The owner writes that it worked well for a few months. Then the answers changed: the agent forgot what the store sells, recommended products it does not carry, stopped following the order logic, and promised customers things the store cannot do. Support replies at the second level took over 36 hours, and the team that built the agent offered no support. The owner disabled it.
How it wentPer the owner's review: the agent went from solving about 95% of issues to failing in 100% of cases, in his words. He says the working months convinced him of the value of this kind of agent, and he would pay close to a full-time salary for one that is supported. The failure he names is not the idea; it is the change with no support behind it (owner's review on the Shopify App Store, January 4, 2026).
Try it · based on the source's workflow
Audit the support agent for [store name]. Here are [number] recent conversations it handled: [paste]. Check each answer against our catalogue [paste or link] and policies [paste]. Flag any answer that was irrelevant, named a product we do not carry, or promised something we cannot do, such as [refunds, dates, order changes]. Then write: a short list of hard rules and policy limits for the agent, the topics that must always go to a human, and the support response time I should demand from the vendor before I rely on it again. Do not contact customers or change any setting.
A starting template based on the workflow the source describes. Review it before you use it; you are responsible for what it does.