Hosted by New Distribution, the Breakthrough 2019 (5th) FMCG + Internet Conference was grandly held from March 15 to 18 at the Chengdu Longemont Hotel. The event attracted over a thousand industry professionals, including distributors, manufacturers, and internet companies from across the country, filling the venue to capacity. The following is a transcript of the speech delivered by Pinlan CEO Li Yifan at the Brand Channel Digitalization Forum on March 16, organized by New Distribution for our readers. Hello everyone, my name is Li Yifan from Pinlan. Pinlan is a technology service company that provides AI-driven marketing, logistics, and manufacturing solutions for brands and retailers. In the field of AI applications, our team has served clients such as Shinho, Vipshop, SAIC Motor, and Ziroom, and we are committed to becoming AI experts in item recognition. Today, I will share with you: What marketing value AI item recognition technology can bring to brands and retailers, and how to implement AI to achieve digital transformation in enterprise management. Pinlan is a startup with a strong technical background, deeply rooted in the industry, combining user needs with technology to solve enterprise pain points. This is my WeChat; if you are interested, we can discuss further after the session. Our initial entry into the FMCG/retail sector was driven by a major backdrop: Traditional retailers and brands conduct 80% of their business offline, making shelves the main battlefield for enterprises. According to statistics, hundreds of billions of RMB are invested annually in the shelf war, with high costs. Many processes in these high-investment channels rely heavily on human resources. For example, when expanding key KA business channels from first-tier cities to third- or even sixth-tier cities, managers face a large number of frontline sales staff who cannot be directly supervised. Thus, effectively managing these personnel, reducing management costs, and collecting reliable, high-quality information become pain points for enterprises. We believe that before solving this pain point, it is essential to understand the journey of digital transformation, which mainly involves three steps: The first step is informatization. That is, the recording of marketing data begins to use intelligent tools. The adoption of informatization has been very successful in the past two years; only a few small distributors and stores still rely on paper and pen for recording information. The second step is datafication. Enterprises have vast amounts of data in their manufacturing, warehousing, and distribution systems. Initially, these systems are often siloed across different departments, so we need to integrate the data. Once data is integrated, managers can gain an overall view of product performance and make effective decisions. According to observations, Alibaba Group has done the best in datafication, and the penetration rate of datafication has been gradually increasing. So, after datafication, how will enterprises develop further? The third step is intelligentization. This is where AI excels. I believe that in the FMCG industry, such as food and beverages, we are still some distance away from intelligentization. However, the cost of using big data and AI for digital transformation is currently high; even data collection through internet devices requires substantial material support. Our company's goal is to enable industry peers to use the best intelligent technology at the highest cost-performance ratio, implement AI technology, and effectively solve industry pain points. Taking the issue of data transparency as an example, informatization does not necessarily mean data transparency, and using intelligent tools does not guarantee that the data collected is more authentic and reliable. This is because during execution, your salespeople might not honestly record the information. Why? The main reason is that our current systems collect data based on forms, meaning data is still entered by humans. I believe that in the future, when data collection becomes unmanned, the data used for supply chain decisions will become more authentic and transparent. Based on these considerations, we believe effective enterprise decision-making consists of two parts. The first part is a collaborative intelligence collection system, helping brands and distributors effectively collect market product movement data. The second part is an operations management optimization system, enabling intelligent management and execution to optimize pricing, display, and distribution. The intelligent optimization system is very valuable, but to obtain it, you must lay a solid foundation in datafication and informatization. Today, I will use product data collection as an example to show you how to achieve collaborative intelligence collection. What I just described are the pain points and our AI solutions. Based on these solutions, Pinlan has built its team to provide technical and product support for these ideas. To achieve these business goals, we need to combine three technologies: First, AI technology, which helps enterprises analyze data, videos, voice, and text, ultimately aiding decision-making. Second, cloud computing technology, which is crucial for integrating data. Third, IoT technology, such as cameras in warehouses that can automatically record cargo information, completing dozens of times the work of a human and operating 24/7. Through continuous R&D in AI-based image recognition and integrating IoT devices, Pinlan launched PinShi, an item recognition platform. Our technical team consists of algorithm engineers from prestigious universities such as Carnegie Mellon University and Tsinghua University. Our COO has held key positions at Oracle and IBM, with 18 years of marketing experience. Combining technology and marketing, we provide customized solutions for our clients. Additionally, we collaborate academically with institutions like Tsinghua University to bring the latest technological support to enterprises. Over the past year, we have completed numerous implementation projects. On one hand, we partner with cloud service providers like Microsoft and AWS to ensure our products have hybrid cloud deployment capabilities and high availability. On the other hand, Intel is also a very important partner; combining our top-notch software with their powerful hardware support enables enterprises to use our AI products efficiently and conveniently. Through these partnerships, we have brought intelligent production, manufacturing, and marketing services to various types of clients. Currently, our PinShi item recognition platform is being implemented in different scenarios. Now let's look at the architecture of the PinShi platform. As shown in this architecture diagram, the top layer is the application microservices layer, where each client's microservices are unique. The middle layer is our core product—the PinShi platform. The PinShi technology platform provides a rich set of services. First, we collect and aggregate data; the core engine of PinShi is the AI model, which requires data for training. Second, we build AI models and deploy them as services; once data is aggregated, we use these two services to output AI models. Third, we offer important and cutting-edge capabilities in item recognition, such as 3D item recognition and fine-grained recognition. With these technologies, we can accurately identify items that are easy for humans to distinguish but difficult for machines, such as juice products that differ only in subtle packaging details. In the application microservices layer, we completed the Anji inventory counting project. Anji holds a 70% market share in the automotive logistics sector. We are currently developing a complete automated warehousing and logistics system, and the inventory counting part uses our technology. In the lower part of this image, there is a bright section, which is our vision system; the bright part is the light source, and the protruding part in the middle is the camera lens. With this equipment and software, we automatically complete the counting of all shelves in the warehouse and output the number of parts in the warehouse. Previously, Anji's warehouse relied on manual forklift operation and barcode scanning for counting, with one person counting about one pallet per minute. With our system, the forklift can run automatically and complete counting in one pass, improving efficiency nearly tenfold. Because it is fully automated, no additional warehouse staff are needed; just one machine is required. The second scenario I will discuss shortly. The third is our shelf inspection scenario, where we use our self-developed Xiaolan robot for inspection. The supermarket in this scenario is relatively small, and the products on the shelves are arranged with depth. Compared to the warehouse scenario above, automotive parts are mostly large boxes. Large boxes do not have depth relationships; only the outermost barcode needs to be recognized. In the supermarket inspection, we accomplished two scenarios. First, because of the depth arrangement, we used Intel's 3D camera equipment combined with our self-developed 3D product recognition algorithm to recognize products with depth. Second, we analyze the front-row product display to identify display compliance, such as misplaced displays and out-of-stock items, which our AI capabilities can also handle. On the far right is a production line transformation we did for a textile factory. In the food traceability scenario, food packaging requires barcode labeling, and we can collect data in high-speed motion environments. Overall, AI platforms like Alibaba and Baidu are also expanding into similar scenarios. For example, Baidu's face recognition is free for a limited number of calls. So what is the difference between our PinShi platform and theirs? First, we are a partner of Baidu AI. Baidu's AI platform provides general AI capabilities, but in specific scenarios and fields, they also need partners to provide targeted capabilities. In our interactions with clients, we encounter very detailed scenarios with many engineering challenges that require fine-grained engineering modifications. Second, the algorithm APIs released by large platforms tend to be conservative and may not incorporate the latest algorithm results from March 2019. In contrast, our technical team brings the most cutting-edge and effective algorithms into real-world applications. Third, we focus on product recognition and have accumulated over 5 million product data points. Clients only need to provide a list of products they want to analyze, and our platform can quickly output results. This is the inspection robot—Xiaolan, and this is the touchscreen interface on the Xiaolan robot. That concludes my sharing. Thank you.