李宁把AI智能体“编入”组织：50个Agent，等效220名员工

Summary 1: Core data and goals for Li Ning's AI deployment.

1. About 50 Agents are live, delivering 460,000 equivalent work hours over one year, equivalent to about 220 employees; the next step is to expand to 200 Agents within a year and increase equivalent work hours by an order of magnitude to 4.6 million.

2. About two-thirds of the 460,000 hours come from digital store managers, with the rest from sales-associate coaching, knowledge Q&A, customer service, business analysis, and logistics and supply chain; IT department coding time is not yet included.

Summary 2: Reusable operational methods.

1. Start with scenario screening: dozens of workshops early in the year reviewed more than 200 candidates and ultimately selected more than 20 priorities, using reusability and efficiency multipliers as criteria; if existing digital systems can be enhanced to solve the problem, do not force AI.

2. Digital store managers handle morning and evening reports, business diagnosis, and task follow-up, and assist with merchandise adjustments and scheduling; customer relationship maintenance remains human. Outstanding sales-associate interactions enter a content pool; those that lead to transactions earn points redeemable for benefits.

3. On security, first review WorkBuddy's data storage, invocation records, and legal provisions; apply dual policy and technical controls; outbound sensitive data is restricted.

Summary 1: AI enters brand R&D, design, and merchandise management.

1. In design and development, AI helps designers turn design concepts into 3D modeling and printing outputs and also participates in process and cost control.

2. In merchandise, a combination of large and small models supports allocation, inventory, and pricing management; in supply chain, veteran workers' experience, supplier process requirements, and quality standards are codified into AI knowledge for production-line workers to query.

Summary 2: AI enters store channels and sales-associate marketing.

1. At the front end, digital store managers handle morning and evening reports, business diagnosis, and task follow-up, and assist with merchandise adjustment and scheduling; customer relationship maintenance still requires people. Li Ning plans to expand from about 50 Agents to 200 within a year.

2. On the sales-associate side, AI learns how outstanding sales associates interact with consumers and the content they use; effective materials generated by AI enter an internal content pool for other sales associates, and original contributors that drive transactions can earn points redeemable for benefits.

3. In brand organization, Agents are assigned to positions and a silicon-based employee platform is built to clarify responsibilities, equivalent work hours, and process tasks, while exploring future performance evaluation.

Summary 1: Learnable AI operating tactics and opportunities.

1. Li Ning uses about 50 Agents to deliver 460,000 equivalent work hours, about 220 employees; the target is 200 Agents and 4.6 million hours within a year. Among key scenarios, six are related to open source; outstanding sales-associate content can enter an internal content pool, and contributors who drive transactions can earn points and benefits.

2. Scenario screening looks at reusability and efficiency multipliers; dozens of workshops early in the year selected more than 20 priorities from over 200 candidates. If existing digital systems can be enhanced to solve the problem, AI is not forced.

Summary 2: Risk notes and collaboration approach.

1. Security is a precondition for scaling. Companies need to confirm tool security commitments, study data storage, invocation records, and legal provisions, and use data compliance and trade secret committees to update policies; deploy security gateways and leak prevention to control outbound sensitive data.

2. Do not build models in-house; treat official Hub model capabilities as utilities and focus energy on business scenarios. Business departments can submit requirements, participate in development, and even use AI Coding to create tools distributed to subsidiaries; IT handles engineering, organizational context, and business systems.

Summary 1: AI needs in production and design.

1. In design and development, AI helps designers turn design concepts into 3D modeling and printing outputs, and participates in process and cost control.

2. In supply chain, veteran workers' experience, supplier process requirements, and quality standards are codified into AI knowledge so production-line workers can query it at any time; logistics and supply chain also have Agent applications.

Summary 2: Implications for digitalization and e-commerce.

1. Li Ning built AI Studio and incorporated AI planning and applications into five areas: design and development, supply chain, merchandise, retail terminals, and functional departments. In merchandise, a combination of large and small models supports allocation, inventory, and pricing management.

2. Scenario selection first looks at reusability and efficiency multipliers, distilling more than 20 priorities from over 200 candidates; do not blindly chase technology; if existing system enhancements can solve the problem, do not adopt AI.

3. Organizationally, set up a silicon-based employee platform and clarify Agent responsibilities, equivalent work hours, and process tasks; IT acts as an enabler, business participates in development, and IT handles engineering and connects business context.

Summary 1: Industry trends and customer pain points.

1. The trend is for Agents to move from tool-level efficiency toward entering business processes and organizational capabilities. Li Ning has launched about 50 Agents, equivalent to 460,000 hours a year and about 220 employees, and plans to expand to 200 Agents and 4.6 million hours within a year while building a silicon-based employee platform.

2. Customer pain points include precision in scaled operations, scenario screening, security and trust, data openness boundaries, and ROI calculation. Li Ning previously hesitated to scale because security commitments were unclear.

Summary 2: Possible solution approaches.

1. Use Workshops to build awareness, scan opportunities, and remove doubts; align understanding before refining requirements, and match resources by business value. Scenario screening looks at reusability and efficiency multipliers.

2. On security, use security commitments, policy updates, security gateways, and leak prevention. On data, do not send all data to AI; use AI tools to analyze data while tools continue to run in internal business systems or decision-support systems.

3. Technically, adopt combinations of large and small models, enterprise knowledge codification, AI Studio direction management, and official Hub model capabilities; treat models as utilities and focus on business scenarios.

Summary 1: Enterprise demand for AI platforms and related issues.

1. The demand is to bring dispersed Agents into the organization, clarify job responsibilities, equivalent work hours, and process tasks, and explore performance evaluation. Li Ning is building a silicon-based employee platform and aims to expand from about 50 Agents to 200 within a year.

2. Issues include security and trust, data openness boundaries, and ROI calculation. Previously, the company hesitated to scale because of concerns that information could be retrained or questions could be stored.

Summary 2: Latest platform practices and operations management.

1. Enterprise edition WorkBuddy removes irrelevant experts and Skills and builds its own Ning Expert, covering store operations, store daily reports, financial analysis, HR recruitment, etc. AI Studio incorporates applications into five directions.

2. In operations, executives require every person and team to use AI well; dozens of Workshops were held, and one optional training attracted more than 1,000 attendees. Business departments can use AI Coding to create tools distributed to subsidiaries, while IT handles engineering and connects organizational context and business systems.

3. For risk mitigation, update information security policies, communicate security red lines, deploy security gateways and leak prevention, and strictly control outbound sensitive data and files. In the short term, do not build models in-house and treat official Hub capabilities as utilities.

Summary 1: New industry developments and questions.

1. The development is that sportswear brands are incorporating Agents into the organization: Li Ning's about 50 Agents represent 460,000 equivalent work hours a year, about 220 employees; the target is 200 Agents and 4.6 million hours within a year, plus a silicon-based employee platform.

2. New questions include how to screen high-reusability, high-multiplier scenarios; how to assign positions and performance to Agents; how to calculate ROI from individual efficiency to organizational efficiency; and security and data openness boundaries.

Summary 2: Policy and regulatory implications and business model.

1. Policy and regulatory implications: based on data compliance and trade secret committees, update information security policies for AI use scenarios and communicate security red lines; study tool data storage, invocation records, and legal provisions; deploy security gateways and leak prevention.

2. In business model, explore a silicon-based employee platform, measure Agent value by equivalent work hours, and incentivize sales-associate contributions through a content pool and points-based benefits. Li Fei's view is that AI value goes beyond single-tool efficiency; it is not an efficiency problem but doing new things, and organizations rely on workflow collaboration.