DingTalk setzt erneut auf KI: Gründet Branchen-spezifische Modellteams, die dem CTO berichten | Exklusiv von Intelligence Emergence

Text | Deng Yongyi

Editor | Su Jianxun

Intelligence Emergence has exclusively learned that DingTalk recently established a new business line - industry-specific models, which exists as an independent team reporting to DingTalk's CTO, Zhu Hong.

This is also an important move in DingTalk's AI strategy implementation after the return of its founder, Wu Zhao.

A person familiar with the matter told Intelligence Emergence, "After the team was established, DingTalk has contacted several industry clients, and several industry/enterprise-specific models are currently in progress."

Since returning to DingTalk in April, Wu Zhao has prioritized product experience and AI innovation. Previously, Intelligence Emergence exclusively reported that since April, DingTalk has undergone a comprehensive rectification - covering a wide range, from product design and troubleshooting to rectification, and Wu Zhao has been deeply involved on the front line.

Since the popularity of ChatGPT, DingTalk has completed the integration of basic large model capabilities. In August 2023, DingTalk opened its intelligent foundation (AI PaaS) to ecological partners and clients, encouraging partners to use large models to reinvent their products; then in January 2024, it launched an AI assistant with perception, memory, planning, and action capabilities, capable of performing tasks across applications.

According to the figures previously disclosed by DingTalk, it currently has over 25 million enterprise organizations, among which more than 2.2 million enterprises are using AI on DingTalk, covering 20 primary industries such as manufacturing, healthcare, finance, and retail.

The establishment of the industry-specific model team is an indication of the continued implementation of large models in the enterprise sector after technological and product development.

The challenges of enterprise AI implementation are not small. On the one hand, although most enterprises, especially small and medium-sized enterprises, have a strong demand for AI, they generally lack professional technical teams and data processing capabilities; on the other hand, while general large models are powerful, they are difficult to meet the professional needs of vertical industries and require in-depth customization and optimization for specific scenarios.

Intelligence Emergence has learned that DingTalk's industry-specific model team mainly targets enterprise clients and third-party partners on the DingTalk platform. For example, for small and medium-sized enterprise clients without sufficient AI talent resources, DingTalk will provide full-process model training and data engineering services, including front-end data labeling, cleaning, and model tuning, all completed by the DingTalk team.

Compared with platforms such as Alibaba Cloud's Bailian, which are targeted at developers, DingTalk's industry-specific models are closer to business scenarios. "The industry-specific models are co-created by DingTalk and business personnel in enterprises who understand the business and the industry, precipitating industry know-how so that enterprise clients can use the models faster and better," a DingTalk insider told 36Kr.

Currently, DingTalk has achieved initial results in industry-specific models. The Doukou Gynecology Large Model launched in July is actually the first vertical-specific large model successfully implemented on the DingTalk platform. As a vertical model in the medical field, it can improve the diagnostic accuracy of six major gynecological symptoms from 77.1% to 90.2%.

While promoting industry-specific models, DingTalk is also accelerating the completion of the AI ecological closed-loop. Another new move is to revamp the application market.

Agent has become an "open secret" in the competition among major tech companies in 2025. Since 2024, major tech companies have launched Agent platforms, including Alibaba Cloud's Bailian, ByteDance's Coze, Baidu's Ernie Agent, and Tencent's Yuanqi.

DingTalk also launched the AI Agent Store in April 2024. Intelligence Emergence has learned that after Wu Zhao's return to DingTalk, an important task is to reconstruct the logic of the Agent market.

In addition to a significant change in the application recommendation method, DingTalk will make more efforts in the Agent market in the future, opening up capabilities to more ISVs and enterprises, helping enterprises build Agent applications, and achieving a commercial closed-loop through DingTalk, connecting the entire Agent application ecosystem.

Welcome to communicate

Welcome to follow

This article is from the WeChat official account "Intelligence Emergence", author: Deng Yongyi, published by 36Kr with authorization.