Group Chats Are Not DMs: Routing Rules for WeChat Customer Service Auto-Reply

E-commerce teams on the messaging platform WeChat navigate two distinct interaction channels: one-to-one direct messages, and group chats consisting of buyer groups, VIP groups, and store announcement groups. A common issue arises when teams attempt to use a single auto-reply bot for both scenarios, leading to problems such as the bot becoming inoperative in group chats or inundating them with responses.

This is due to the fundamental difference between a group chat and a private inbox; the latter requires its own specific routing rules.

A naive bot that replies to all inbound messages operates correctly in one-to-one conversations but presents three major issues in group chats: noise amplification, answer misattribution, and risk exposure. In a group of 200 buyers, ten members discussing sizing could generate ten bot responses, causing members to mute the group and lose their direct communication channel with the store.

Additionally, the bot's response to member A's question in a thread where member B initially posed a different inquiry can lead to mistrust in the bot's answers. Furthermore, aggressive bot posting within groups triggers WeChat's risk detection systems, potentially leading to account restrictions.

To address these issues, WeChat employs a routing table that treats Direct Messages and group chats as two separate rule sets, despite sharing a common knowledge base. DMs follow a rule where every inbound message enters the reply pipeline, aiming for a first reply within a minute. In contrast, group chats require explicit triggers like an @-mention or specific keywords such as "order" or "refund" to initiate a response.

Untriggered messages are read-only in groups. The bot's contact filter in groups whitelists only staff and verified buyers, preventing it from addressing every member. A per-hour posting cap for bots within groups, separate from the DM cap, ensures the bot holds messages for human review once the limit is reached. The design of triggers is more critical than the quality of answers; a bot that correctly responds to one @-mention per hour fosters greater trust than one that answers forty messages.

Session isolation is crucial to prevent context bleed, where personal transaction details from a DM can be unintentionally shared in a group chat. Each DM and group chat are treated as individual sessions, ensuring that contexts do not cross between them. This practice also maintains the accuracy of quality checks, as each session has its raw transcript material.

According to Weike AI Assistant, the success of this routing system can be measured by three key metrics: the proportion of group messages that necessitate human intervention (which should decrease), the answer-to-mention ratio in groups (ideally near 1:1 to avoid over-triggering), and the first reply time in DMs (which should not decline after implementing the new routing).

Using a shared reply library for both DM and group routing is recommended, as maintaining separate libraries for each channel can lead to inconsistencies and increased maintenance efforts.