Two-thirds of customer tickets resolved before a person touches them

The chain's fourteen stores shared a three-person customer service team and a queue that never emptied: order status, returns, product availability, and store-hours questions arriving nights and weekends when nobody was on. Median response time was measured in hours — and in days after a promotion.

The chain could not staff its way out. Hiring to peak volume was unaffordable at retail margins, and the generic chatbot it had trialed answered from a script that didn't know the chain's return policy from from its loyalty terms.

Scoping meant reading a month of tickets. Most volume clustered into a dozen intents whose answers already existed — in the commerce platform, the policy binder, and the promotions calendar — just not anywhere a customer or an associate could reach quickly.

The signed acceptance criteria set the escalation rules first: anything involving a complaint, a payment dispute, or an ambiguous request goes to a person, with the full conversation and order context attached. Automation earned the routine tickets only.

The agent answers in the retailer's voice from a maintained policy and promotions library, pulls live order and inventory status from the commerce platform, and handles return initiation end to end. It cites the policy it is applying, so customers see the reason, not just the answer.

Escalations arrive as briefings rather than transfers: the conversation, the order history, and what the agent already tried. The service team works exceptions instead of repetition.

67% of tickets now resolve without a person, at a 4-minute median response — around the clock, including the promotion peaks that used to bury the queue for days. Satisfaction scores on automated resolutions track those of the human team.

The three-person team hasn't grown and no longer needs to. Ticket analytics from the system now feed the merchandising calendar: recurring product questions have caught listing errors the old queue would have absorbed silently.