Let's first look at an image from the first 'Power of Evolution · Liu Run Annual Speech' on October 30, 2021, about digitalization. When it comes to digitalization, it is not unfamiliar to practitioners in the FMCG industry. Many companies have started 'top leader projects' and invested heavily in digital transformation research. But what exactly is digitalization? What is digital transformation? Why do we need it? The concepts have always been vague. Liu Run used the analogy of oil extraction to distill all the answers about digitalization from abstract things. Digital transformation is the process of extracting data from the physical world, roughly refining it into information, refining it into knowledge, and aggregating it into wisdom. This process starts with data extraction and ends with aggregated wisdom. In reality, many companies and experts first study the knowledge level in digital transformation, but ignore the most basic data collection. If there are problems in data collection, the information cannot be accurate, the knowledge extracted becomes worthless, and the final wisdom is a splendid farce. Today, let's talk about the digital collection segment. What are the digital oil things for FMCG? Let's look at another image: What are the core pieces of information in digital collection? 1. Store sign information: The type of outlet, location, channel attributes, area, internal product structure, etc. The accuracy of this information provides reference suggestions for future channel cost investment and selection of visual materials like store signs. 2. Item & facing & share information: Market share, facing share, best facing positions, competitor information, etc. The accuracy of this information is to study the market competition landscape and reflect on one's own issues. 3. Display & spatial logic construction information: Foot traffic at the display location, visual merchandising status, competitor displays, etc. The accuracy of this information is to check the efficiency of frontline staff market operations, ROI, and competitive dynamics. In addition to the above, theoretically, all key factors in the store that affect sales are key information, such as cut cases, hangers/nets, price tags, pillar wraps, etc. Summary: Seeing the above information, I believe everyone will realize that data itself has no meaning; what is truly valuable is the content of the data. Like in a picture, seeing how many of our products and competitors' products there are is not significant in itself. Calculating the facing share percentage between our products and competitors' is meaningful; that is the value. Extracting valuable information from a pile of images is the core of collection work in digital transformation. What lies behind the above information? The FMCG industry differs from others in being labor-intensive. There are hundreds of channels to lay out, millions of outlets to cover, and 1.4 billion consumers to reach. No other consumer goods industry has such scale, and all of this requires personnel. 1. Huge amount of information For example, Coca-Cola serves over 3 million outlets, Nongfu Spring over 2 million, and a frontline worker can serve at most about 210 (35 per day, 6 days a week). That means hundreds of thousands of frontline workers are needed for these outlets. With different people and different situations, each outlet needs at least 5 images collected, so the daily information volume in the backend is unimaginably large. As Jinmailang Chairman Fan Xianguo said long ago: 'A million outlets entering the office should be managed like a space center.' 2. Behind truth is precision We have many means to ensure the truthfulness of big data, but precise extraction is not easy. There is a gap between truth and precision. For example, when collecting the shelf share of our products, manual statistics will definitely have large errors. We can ensure the authenticity of frontline staff's shelf images (the images are real), but we cannot precisely calculate the share (personnel calculation errors, and manual counting errors). Summary: The lower the human factor in data collection, the more it aligns with human nature, the more it frees frontline workers to focus on sales communication, and the more valuable the 'rough refining' of data becomes. A key to precise data collection: image recognition Times have changed. With the support of internet technology, managing a million outlets has undergone tremendous changes. Baidu PaddlePaddle EasyDL Retail's intelligent AI image recognition system is a sharp tool to achieve this. Let's briefly study a few scenarios. 1. Shelf scenario: the most common data collection information We can use multi-dimensional image recognition to ensure the 'rough refining' of data. For example: basic product information recognition, including product name, brand, specification, etc.; shelf layer recognition, including the number of shelf layers and the position of our products; scene recognition, including shelves, end caps, freezers, floor displays, cut cases, etc.; facing share statistics recognition, including facing share, unrecognized product information, empty spaces, and shelf utilization. Finally, we can accurately obtain relevant information about our products and competitors. These data collections can be easily handled by Baidu PaddlePaddle EasyDL Retail's image recognition system. 2. Freezer scenario: the most important data collection information in the beverage peak season Freezers are a key battleground for beverages. In the peak season, it can be said that those who get the freezer get the world. The most headache-inducing issue is determining whether self-invested freezers meet standards and whether purchased placements are qualified. This involves two data points: purity and fullness. These two issues often involve costs and later disputes. With image recognition and Baidu PaddlePaddle EasyDL Retail's intelligent image recognition system analysis, all possible disputes cease to exist. 3. Display scenario: The most important data collection information for whether the cost is worth it Displays are important for FMCG product placement and brand visibility, serving both product sales and brand image. Generally, displays are paid placements. The reasonableness of the cost is measured by three factors: visual merchandising status, total number of boxes in the display, and the footprint area. Baidu PaddlePaddle EasyDL Retail's intelligent image recognition system can easily judge these. 4. SKU profiling scenario in images From the source code above, we can clearly see: shelf layers, SKU names, shelf sequence numbers, confidence levels, SKU positions in the image, the order of SKUs from left to right on that layer, and the ranking of that position. Baidu AI Cloud can achieve these points. It can accurately extract valid information from an image, which is the best interpretation of the 'rough refining' of data in Liu Run's digitalization talk. Summary: Whether image recognition technology can be effectively applied to FMCG digital collection is one of the core aspects of 'rough refining' in enterprise digital transformation. It can elevate data collection from the truth level to the precision level, reduce 'human errors', and safeguard the next steps of 'information', 'knowledge', and 'wisdom'. Final thoughts: Over the past decade, the FMCG industry has undergone earth-shaking changes. Technology has penetrated all aspects of FMCG, and we must admit that its impact and changes on the industry are significant and profound. All brand owners should embrace these new technologies. Of course, market demand will inevitably bring a mix of technology companies. Take AI image recognition technology: we must have the ability to discern which are truly intelligent and which are fake. We should respect technology, free frontline staff, and use technology to make the complex FMCG world simpler. Are you 'watching' me?