·3 min read
Customer ops without a mystery answer
A support assistant can close the chat. The question is whether the refund, the SKU and the policy it used are still on the ticket.
Photo: Clark Street Mercantile on Unsplash
Retail support is a volume problem wearing a language problem. Most chats are the same five intents: where is it, can I return it, why was I charged, is this in stock, talk to a person.
A fluent assistant can close a lot of those. Klarna reported that its assistant took two-thirds of customer-service chats in the first month. The interesting number is not the deflection. It is whether the answer matched the policy on that day.
The failure is specific. The bot promises a refund the returns desk will not honour. It invents a delivery window. It cites a promotion that ended last Tuesday. The customer is not angry at “AI.” They are angry at the store.
So the product is not a personality. It is retrieval over the live catalogue, the order, the returns window and the exception list, plus a handoff that does not make the shopper repeat the ticket.
McKinsey’s State of AI keeps finding the same split: pilots are cheap; production needs owners, evaluation and a place the work actually lives.
Log the source on the ticket. If the assistant cannot find the policy, it should say so and pass the chat, not improvise goodwill the finance team will unwind later.
That is slower than a confident guess. It is also how you still have a brand after the first busy Saturday.
The TruFyre Way
A fluent refund the desk will not honour is a brand problem. TruFyre grounds the assistant in the live catalogue, the order and the returns policy, and logs the source on the ticket.