Architecture and Routing Rules of an Online Customer Service System

This article outlines the end‑to‑end workflow of a customer service platform from the user’s perspective: robot → human agent → rating. The core of the system is the IM‑based human conversation.

The overall architecture consists of a visitor side (C‑end) offering robot chat, human chat, and satisfaction rating, and a service side with several consoles: Octopus one‑stop console, robot management backend, ticket console, and online service console.

1. Service Time Settings – Regular service hours are 08:00 to next‑day 00:00 on weekdays and weekends. Special periods (e.g., Double‑11, 618) have extended hours to handle traffic spikes.

2. Routing Rules – Incoming requests are classified by channel (App, mini‑program, etc.), source (product page), routing dimension, and allocation method to assign users to the appropriate agent group. Different product categories (shoes, cosmetics, 3C, etc.) require specialized agents.

3. Queue Overflow Rules – When the number of waiting users exceeds agent capacity, overflow handling includes push notifications, queue monitoring, and load‑balancing. Real‑time dashboards display retained queue counts and waiting times.

4. Online Customer Service System – The first‑generation IM system provides chat, configuration, and user management. The newer Octopus one‑stop console offers a three‑column layout: left (online/offline/leave users), center (chat area with rich media), right (user info, order lookup, ticket creation, product recommendation).

Data flow: client messages → backend → gateway → Octopus console via persistent long‑link. ACK three‑retry mechanism ensures reliable delivery; failures after three attempts are reported as send failures without persisting to the database.

4.2 SOP Assistant – A knowledge‑base with 900+ FAQs is matched to chat content, allowing agents to quickly send relevant answers through a unified assistant tool.

5. Session Termination – Sessions close after a timeout without user reply, after the agent confirms issue resolution, or when the user ends the chat. A single satisfaction survey is sent before closure.

6. Summary – The platform has evolved through multiple iterations, integrating routing, queue management, real‑time monitoring, and SOP assistance to handle high conversation volumes reliably.