How Large-Model AI Transforms Smart Customer Service – Alibaba Cloud Insights

In this speech from the 2024 Cloud Expo AI Innovation Forum, the speaker reviews the development of intelligent customer service across three roughly ten‑year cycles. From 2000‑2010, services relied on simple rule‑based expert systems with limited effectiveness. From 2010 onward, multi‑channel, multimodal chatbots emerged using deep‑learning models, yet single‑task models struggled with generalization.

The release of ChatGPT at the end of 2022 dramatically raised the ceiling for conversational AI, introducing generative large‑model capabilities that improve interaction quality, reasoning, and generality, ushering in a new era of generative AI‑driven smart services.

The speaker then details Alibaba Cloud’s practical experience since early 2023, covering four scenarios: enhancing customer experience with the Tongyi large model, boosting internal service efficiency via a full‑process Copilot assistant, converting expert knowledge into platform‑wide workflow capabilities, and deepening service insight through multi‑modal analytics.

Three concrete cases illustrate these advances:

Customer Service Assistant: a dialogue bot that clarifies vague queries, maintains multi‑turn context, and delivers precise, source‑traceable answers enriched with images, tables, and self‑service tools.

Internal Copilot for sales and support staff: a real‑time, workflow‑embedded assistant that recommends solutions during pre‑sale interactions and summarizes issues during post‑sale ticket handling, accelerating resolution and closing processes.

Intelligent Service Workflow: a low‑threshold, scalable method for converting scattered expert knowledge into executable business flows, enabling experts to define processes in natural language which the system translates into interactive tools for shared use.

From these cases, three guiding viewpoints emerge: (1) build domain data advantage early and continuously; (2) embed intelligent services into actual business flows through diverse interaction forms; (3) evaluate and optimize large‑model performance using domain data and expert feedback.

The underlying technical stack consists of three layers:

Domain data layer: online production of service knowledge, transforming raw data into knowledge bases, graphs, and workflows.

Model layer: a combination of a domain‑specific large model (based on Tongyi Qianwen) and specialized small models for tasks such as scene recognition and search augmentation.

Application layer: modular pipelines and an Agent framework supporting external customer‑facing chatbots, internal Copilot assistants, and global service insight tools.

Model optimization follows a cost‑to‑benefit hierarchy: prompt engineering, retrieval‑augmented generation (RAG), domain‑specific fine‑tuning with large high‑quality datasets, and finally end‑to‑end optimization integrating all techniques.

For small‑ and medium‑sized enterprises wishing to build their own smart services, a four‑step roadmap is proposed: (1) establish domain data advantage; (2) develop domain understanding via prompt engineering and RAG; (3) enhance business‑process efficiency with agents and Copilot‑style assistance; (4) expand into new intelligent service scenarios such as marketing, decision support, and multi‑scene coverage.

Implementation can leverage Alibaba Cloud’s Tongyi foundation model and Tongyi Baillian, or adopt open‑source frameworks for custom development.