Track Hyper | SEMIR: An AI Organization Grown on DingTalk

Author: Zhou Yuan / Wall Street News After this year's Spring Festival, "DeepSeek" ignited the entire society's attention to generative AI (GenAI). Semir also took the opportunity to "press the acceleration button": a letter from the chairman to all employees set the tone, and posters stating "AI is not an option, but a necessity" were posted in restaurants and at elevator entrances; the Human Resources Center collaborated with the Digital Center to integrate the "learning-practice-incentive-review-communication" organizational loop into the daily work of 3,000 employees. As Bai Yun, Senior Director of Semir's Human Resources Center, said, this is a systematic advancement from culture to mechanism, "We hope that all 3,000 colleagues can also engage in the use of AI." Unlike most companies that use AI technology to enhance individual efficiency, Semir's approach resembles an organizational-level reconstruction. At Semir, one can "see what an AI organization might look like": employees form cross-departmental teams from the bottom up to solve business problems together with IT; Human Resources has adapted the system from course training to certification, and then to promotion and compensation incentives, "After AI enhances efficiency, HR has the responsibility and obligation to help employees consider what other more valuable things they can do," Bai Yun said. These changes can be summarized in one sentence: This is an AI organization that has grown on the DingTalk platform. So, how did this clothing company "grow on the DingTalk platform"? How did it transform AI technology from a tool into an organizational capability? At the same time, how did it rely on a replicable mechanism to turn "individual efficiency" into "value chain reshaping"? Rising: Starting from Pointed Attempts Semir's AI journey began in June 2023. Lu Lina, Senior Director of Semir's Digital Center, recalled: "We directly established an AI project team, and from the very beginning, it was a company-wide project." In fact, Semir's entry that year was quite restrained—small cuts, scenario-based, high-value scenario pilots. At that time, the precision of text-to-image generation was not sufficient to support "marketing blockbusters," so the Semir team first implemented it in areas with lower precision requirements, such as ordering materials and e-commerce innovation pages; at the same time, they continuously trained to improve the model's performance in the clothing vertical scene, collaborating with colleagues who had strong business backgrounds and enthusiasm for the tools. The real turning point occurred after this year's Spring Festival. "Not only did the management pay more attention, but ordinary employees also experienced many impacts brought by AI," Bai Yun stated, emphasizing cultural mobilization first: the chairman's letter + posters + internal promotion formed a general awareness and atmosphere; mechanism incentives followed closely: a 100-day check-in for all employees, a collection of golden ideas, and a business-related challenge competition were launched simultaneously. In the first half of the year, three representative achievements were recognized in the challenge competition, among which the most notable was the "Qualcomm Million T" project. Wall Street News learned that this project, based on "Qualcomm fabric + Qualcomm pattern," promotes rolling fabric planning, combines intelligent AI to quickly generate images based on current fashion trends, enhances efficiency by 85% with new technologies like 3D, and meets personalized needs through rich fashion prints, achieving flexible quick response through reprocessing. In addition, production is determined by sales, and the speed of factory warehouse delivery is increased by 93%, creating a fast response system that connects the entire chain from research and development to delivery, achieving quick response for small orders and near-zero inventory operations In fact, this is not just a single-point efficiency improvement, but a value chain reshaping of "tools + processes in parallel." When this momentum takes root, it does not fall into the stereotype of an "age gap." Lu Lina's judgment is: it's not about age, but about mindset and curiosity. Semir organizes "pioneer teams" in each business unit, calling them "Digital Intelligence Pioneers" in 2024, evolving into various forms such as "Strike Force," "Club," and "Spark Ignition" in 2025, embracing AI from super individuals to super organizations. "These are not rigid requirements imposed by the company, but rather spontaneously formed by the organization." Semir's AI organization "Spark" nourishes each other with business scenarios, forming the first growth curve of AI diffusion at Semir. More importantly, unlike many companies that merely view AI technology as a tool for improving efficiency for individuals or single workflow segments, Semir elevates AI from a "productivity" tool to a means of reconstructing "production relationships." In other words, many companies only advocate "AI is powerful, we need to use AI," but do not adapt at the organizational level, such as incentive mechanisms, skill mastery grading certification for transforming workflows with AI technology, etc. Some enterprises turn AI into an "efficiency dividend" for a few, but with unchanged production relationships, outstanding individuals may "leave with the tools," becoming "super individuals" that the organization cannot retain. Semir's answer is systematic adaptation: learning resources, incentive policies, promotion pathways, and job profiles are all aligned with AI. To be specific, Semir has established a promotion and certification system for "AI super individuals." Bai Yun clarified, "From this year (2025) to next year's employee promotions, we will prioritize the AI certification indicator. Those with such capabilities, combined with performance achievements, will be prioritized for promotion recommendations by the business units; when it comes time for salary adjustments next year, those with certifications will be appropriately prioritized for salary increases." Since July 2025, Semir has launched digital talent certification, receiving a total of 508 applications from various first-level organizations of the shareholding company, with 56% of applicants passing the certification. Semir has also promoted updates to AI talent profiles and AI "super individual" talent level certification: incorporating digital capabilities into profiles as "driving factors"; HR and business jointly advance intermediate talent certification. According to Lu Lina, "Last year, when we promoted talent certification, there were also general managers from business units who actively applied and passed the certification." An internal mechanism for sharing AI practices has formed among Semir's management, with general managers of business units conducting closed-loop reviews of effective practices. In this mechanism, AI is no longer just a productivity tool but is designed into production relationships: avoiding the learning burden being "outsourced" to employees while establishing behavioral incentives through tangible promotions and compensation as "anchors." DingTalk in Hand: The Workspace as the Entry Point In the process of Semir's AI organizational practice, they have adopted DingTalk as the "main operating system" for the AI organization. "As long as DingTalk is in hand, these tasks can be accomplished," said Lin Jianxia, head of AI innovation at Semir, noting that in terms of implementation, Semir has constructed four interlocking pieces The first piece of the puzzle is the learning area. The online knowledge base has opened up the entire link scenario: planning insights, design and development, marketing promotion, terminal retail, and operations are all related to "general tools + specialized tools"; corresponding courses are launched through "Sen Academy," combining "general courses + job-oriented courses," with short chapters that support fragmented learning; compulsory courses will be pushed according to positions in a "thousand people, thousand faces" manner; learning earns points that can be exchanged in the cafeteria and convenience store. The second piece of the puzzle is the benchmark case library, which is relatively simple: it consolidates excellent scenarios from various business units, and the "AI Sen Doctor" IP is regularly promoted in the employee group, creating continuous exposure and reuse. The third is the AIGC small forum/community: derived from the 2023 "AI Design Competition," it has evolved into an internal community with "topic circles." Senma has over 3,000 employees, with more than 1,700 voluntarily scanning the code to join. Here, there are insights on "smart use of AI tools" and various "toolkits" compiled, becoming a hub for inspiration/reuse. The most important piece of the puzzle is the AI form (multi-dimensional form) work order system, which is the core pilot of Senma's AI organization: problems - diversion - response - tracking are all systematized. Image description: Senma's internal AI applications are flourishing everywhere. To put it simply: the backend uses AI to automatically tag "problems/expectations/suggestions/complaints" and automatically assign them to the corresponding staff; the system provides AI feedback as an "emotional buffer" — first giving users a receipt that "someone has heard" before waiting for manual follow-up explanations; then, for complex processes like judges scoring, re-evaluation, and personalized permissions, the same set of AI forms is used. This achieves a closed loop of "learning - practice - sedimentation - reuse - feedback - re-iteration" at a single entry point; more importantly, each step comes from the business site: the tool list in the learning area is mostly composed of "useful things tested by frontline employees in various different fields," which in turn can be reused by all employees. If the above is the mechanism (infrastructure) supporting the AI organization, then the infrastructure must be used for business practice; the most obvious feature of Senma's AI organization is the reconstruction of business production relationships: AI is not for show, but to achieve practical application. Observing the implementation of Senma's AI organization, we can analyze it from "Qualcomm's million T" to "AI training goods," which can serve as a perspective window for value chain strategies. The "Qualcomm million T" project is a typical example of Senma "moving from efficiency to value chain-driven." In this regard, Lu Lina has a precise summary: tools + processes, then parallel them to ultimately reshape the value chain. Looking at "AI training goods," this embeds AI into the capabilities of terminal sales guidance to build a link to revenue growth, with the approach being co-creating training goods products with DingTalk. According to Lin Jianxia, the 1.0 version is first launched in scenarios like training goods, matching, and customer retention, setting multiple rounds of weights (such as guidance, product comparison, experience stages), continuously optimizing the template based on data feedback; through the template, tasks are then activated, pushing training to designated store sales staff, who then submit videos on their mobile phones, with AI automatically scoring based on the established expert model and providing improvement suggestions The first round of co-creation is selected in the Bala franchise system. On one hand, this system has been implementing "super shopping guide quarterly training + video assignment check-in" practices offline for many years, and AI just happens to "stitch the gaps"; on the other hand, its evaluation system and business rules are well-developed, making it easy for template training and implementation. Lin Jianxia revealed that over 500 shopping guides are currently using it; after seeing the results, Mini Bala and Semir brands have successively come to expand the scenarios, "but AI products are not just for use; they must be adapted to their own business rules." The inspiration for "AI product training": using business rules as a anchor point to achieve "scene standardization - business templating - scalable replication," and allowing business units to conduct secondary training based on category differences. This is similar to "Qualcomm's million T," both embedding AI into key business backbones rather than "sticking it on the surface." Marketing and Operations: Finding True Customers Semir's practices on the marketing side show that finding the right internal customers is more important than the tools themselves. For example, in text-to-image generation: from "generating background images in the conceptual stage/replacing extreme shooting environments," to combining holiday hotspots with "new styles for children" to create a series of posters; the expression of functional points (such as heating, moisture resistance, wind resistance) is more concrete. Another example is the content middle platform + AI mixed editing fission, which means that content is centrally processed by the middle platform, allowing terminal shopping guides and consumers to help brands with secondary word-of-mouth promotion at a low threshold. Interestingly, Lin Jianxia said that when the brand marketing department applied this mixed editing tool, the feedback was "very poor"—the efficiency of AI mixed editing did not match the brand's quality requirements for videos; instead, the operational promotion's demand for multi-channel dissemination and video quantity transformation was more aligned with the tool's efficiency improvement points. Therefore, "the same tool has completely different customers in different business departments; it is essential to find the right people." Lin Jianxia said that embedding AI capabilities into business processes and finding the "right" people is more critical than the strength of the capabilities themselves. On the customer service side, Semir distinguishes between To B and e-commerce To C: in the logistics customer service scenario, "the focus is on problems, and the demand is to quickly find results," where adding AI customer service can significantly relieve pressure; however, e-commerce customer service must be responsible for the warmth of brand service and sales conversion, making the team "more cautious." These trade-offs, which vary by customer, demonstrate Semir's sensitivity to the balance between AI implementation KPIs and business: it is not about "using it if it can be used," but rather "it must bring verifiable business value to be implemented." In the AI era, the role of traditional IT departments or teams is also being rewritten. Lu Lina has split the team into two groups: one group focuses on upgrading digital infrastructure, building application and data integration capabilities around products, retail, and supply chains, while adding algorithms; the other group acts as "evangelists," promoting "everyone can use AI," spreading tools and methods to the front lines of business. "It used to be that the IT department (or digital center) created tools based on business needs; now the business system can create its own tools." Lin Jianxia said this means that the role of the digital center has changed to "evangelism and platform": providing platform tools (such as AI spreadsheets/multi-dimensional tables, AI assistants), co-creation templates (such as AI product training), methodologies and training (such as general courses, case libraries, certifications), allowing frontline businesses to use "existing bricks" to build "their own walls." The HR department is responsible for "creating a complete loop from culture, incentives, to training." The AI organization has more important value links, including: certification as governance, turning individual sparks into organizational assets. For example, Semir's AI talent certification is a key mechanism for achieving the transformation of "individual sparks into organizational assets." Currently, Semir's AI talent certification is divided into three levels: beginner, intermediate, and advanced, covering three major channels: AI, BI, and system applications (including but not limited to RPA and AI forms). Intermediate application talents are the focus identified by Semir. After meeting the standards for general courses and question banks, they must pass the "application case report review"; the case must clearly answer "what scenario, what problem, what tools were used, and what is the ROI"; the review emphasizes business value: it must be "used for work." Advanced talent certification requires that the certifier not only creates actionable value using digital tools but also has the ability to build workflows or systems to form systematic applications that enhance organizational efficiency, along with actual application outputs. In other words, it aims to identify business individuals who can "build an efficiency enhancement system using available tools." The scale and intensity of Semir's AI talent certification are significant, with over 300 applicants for AI-related positions. According to Bai Yun, this year, 150 AI talents were certified from the AI case submissions. To ensure fairness, the judges are organized into four groups based on "cross-departmental, blind selection," with each group consisting of five judges, each from at least four different departments of business or technical experts. "Each group is assigned 70 cases" — the workload for judges is evident: "With over 60 people, each reporting for 10 minutes totals 600 minutes." Judges assess both the technical rationality and the completeness and effectiveness of the entire case. More importantly, this certification is entirely conducted on AI forms: case submissions, attachment uploads, judge scoring, review circulation, permission isolation, and "personalized experiences for thousands" are all completed on "one form." This not only realizes a streamlined and transparent review process but also allows "excellent cases to immediately enter the knowledge base - be disseminated - and reused," forming a growth flywheel. Platform and Co-Creation: Why DingTalk Regarding the "platform selection" issue, Semir's standards are pragmatic: speed matching and ecological linkage. First, the foundation for growth — the AI platform — must have frequent updates. After the release of DingTalk 8.0, Lin Jianxia stated that the AI form (formerly "multi-dimensional form") "iterates very quickly and effectively meets our demands." DingTalk's meeting AI transcription supports "languages from over 100 countries, including dialects," facilitating communication with overseas departments or clients. The third aspect is ecological collaboration, where a single entry connects the learning area, AI/RPA/multi-dimensional form area, and benchmark case library, linking suppliers at the front end and agents at the back end, creating a collaborative ecosystem throughout the supply chain. Finally, there is a co-creation orientation, such as AI practice, which "focuses on core application scenarios, co-creating with leading clients, and refining vertically." Semir promotes this through a method of "running through a single scenario - replicating across departments - expanding coverage." These modules are very friendly to B-end developers, as the digital center collaborates with tools like "Tongyi Lingma" on the code side, allowing for direct internal integration of applications and subsequent large-scale reuse Therefore, Semir is not about "buying a bunch of tools," but rather "organizing business production on a unified platform." This is precisely the meaning of "growing on DingTalk": learning, spreading, practicing, feedback, reviewing, replicating, and further reusing and refining are all completed on the same network, where the nerve endings of the organization and the capillaries of business are fully connected through the DingTalk platform. If we extend the timeline, Semir's AI practice has roughly gone through three steps: first is point-based efficiency improvement (2023): piloting tools like "text-to-image" in ordering materials, e-commerce innovation pages, and other aspects; the second phase (2024) drives the point-based growth of AI sparks across various business links through strong scenario-based applications; next, organizational mobilization (first half of 2025): a letter from the chairman + posters + three mechanisms (check-in/golden ideas/challenges), forming a network with the learning system and "vanguard - elite camp." The third step, value chain reshaping (from 2025 to now): represented by "Qualcomm's million T" and "AI training goods," embedding AI into the core processes of product development, supply chain, and terminal training; using certification systems and AI forms as governance foundations, turning business improvements into reusable organizational assets. AI is not about replacement, but empowerment: Semir has not left "empowerment" as just a slogan, but has written it into mechanisms: promotions, salary increases, certifications, courses, communities, cases, work orders, and forms, all interlocking. Because of this, employees' creativity can smoothly transform into the structured capabilities of the organization, continuing to feed back into business growth. Looking back today at the years since the emergence of GenAI, the societal heat of "DeepSeek" is merely an external trigger. What truly makes Semir an "AI organization grown from DingTalk" is its organizational engineering: using the DingTalk workspace to support learning, collaboration, application, and governance; connecting rules, processes, and people's feedback with AI forms; activating bottom-up innovation through challenges/golden ideas/check-ins; transforming "individual sparks" into "organizational assets" through certification; and using the dual drive of "evangelists + infrastructure" to change the role of frontline business from making demands to creating their own tools. The essence of this engineering lies not in "how many tools were used," but in whether the organization has reformed its production relationships around AI. Semir's AI organizational practice provides the industry with a replicable path: first, create a solid atmosphere and incentives, then build a robust learning and case framework, followed by refining processes through the DingTalk platform, ultimately achieving AI as the "cells and flesh" of the value chain. When "organization online, business online, feedback online" is truly realized, AI will no longer be a passing trend, but will sink into the daily operations and cycles of the enterprise, ensuring its longevity