Background and Challenges
A convenience-store brand with roughly 800 stores runs a membership and points program covering millions of users. Questions about membership rules, points redemption and returns kept the 20-person support team queueing at peak hours, with first responses often minutes behind; overnight inquiries sat unanswered until the next morning, wearing down member experience. The questions were highly repetitive, yet piling on headcount was no longer affordable.
Our Approach
We built a support agent and connected it to two entry points, WeChat Work and the mini-program’s customer service. Membership and points queries call the membership system’s API and return balances and redemption records in real time; returns-policy questions are answered from a curated knowledge base, with eligible cases opening an after-sales ticket directly; store lookup and campaign questions return the nearest store and current promotions based on the user’s location. Questions the knowledge base cannot answer transfer to a human agent, and those sessions flow back to improve the knowledge base weekly.
Before launch we built the knowledge base from historical conversations and reviewed it entry by entry. During the gray release the agent served alongside human agents, with the hand-off threshold loosened week by week.
Results
- About 70% of routine questions are handled by the agent, freeing human agents for complex cases (sample data, to be verified);
- First response dropped from minutes to seconds, and peak-hour queueing has basically disappeared (sample data, to be verified);
- Overnight inquiries are covered around the clock instead of piling up until the next day (sample data, to be verified);
- The human-hand-off rate keeps falling as the knowledge base iterates weekly on unanswered questions.