---
title: "AI-Driven Perfect Market Closed Loop"
description: "Trax, one of the earliest companies in product image recognition, founded in 2010, now processes over 1.5 million stores in more than 90 countries daily, converting product displays, promotional materials, and sales activities into readable structured digital assets for real-time, transparent sales operations and decision support. Although online business is growing rapidly, offline remains the mainstay, with about 3.3 million stores in China's urban areas."
author: "Mike Wang"
publisher: "New Distribution"
email: "zhaobo258@gmail.com"
telephone: "+8615854817671"
published: "2020-09-04"
language: "en"
canonical: "https://xinjignxiao.com/en/articles/ai-driven-perfect-market-closed-loop-96c61369/"
markdown: "https://xinjignxiao.com/en/articles/ai-driven-perfect-market-closed-loop-96c61369.md"
original_source: "https://mp.weixin.qq.com/s/zai6B-2BHAkpCQBQR_vlXg"
translation: "https://xinjignxiao.com/zh/articles/ai%E9%A9%B1%E5%8A%A8%E5%AE%8C%E7%BE%8E%E5%B8%82%E5%9C%BA%E9%97%AD%E7%8E%AF-96c61369.md"
attribution: "New Distribution — https://xinjignxiao.com/en/articles/ai-driven-perfect-market-closed-loop-96c61369/"
usage_policy: "https://xinjignxiao.com/ai-policy.txt"
---

# AI-Driven Perfect Market Closed Loop

> Trax, one of the earliest companies in product image recognition, founded in 2010, now processes over 1.5 million stores in more than 90 countries daily, converting product displays, promotional materials, and sales activities into readable structured digital assets for real-time, transparent sales operations and decision support. Although online business is growing rapidly, offline remains the mainstay, with about 3.3 million stores in China's urban areas.

Trax is one of the earliest companies in the world to do product image recognition, founded in 2010. Now, every day in more than 90 countries, it uses image recognition technology to convert product displays, promotional materials, and sales activities in over 1.5 million stores into readable structured digital assets, enabling sales and management teams to conduct real-time, transparent sales operations management and decision support. Although online business is growing rapidly, offline business is still the mainstay. There are many stores worldwide; in China, there are about 3.3 million stores in cities and towns.
The country with more stores than China is India. In China, there are over 240,000 large, medium, and small supermarkets, and the remaining 3 million are mom-and-pop stores. Therefore, many technologies must be practically applicable in these small stores to help enterprises and distributors grow their business. **-01-**
We often talk about perfect store execution. In the current sales system, brand owners and distributors spend a lot of time every day supervising sales representatives' performance in stores. But today's theme is how to achieve incremental growth through digital technology. From this perspective, if we only focus on ourselves, are we missing potential growth opportunities? The 3.3 million stores mentioned above—who in the market covers all 3.3 million stores? Is this a growth opportunity? **There is a 28/80 rule in the market: 80% of sales are concentrated in 20% of stores, and the FMCG industry is no exception.** In Beijing's 6,500 supermarkets, 16% of stores contribute 80% of sales; in Shanghai, it's 26%; in Guangzhou, 23%.
These stores are definitely the golden stores we should prioritize for distribution. They may exist outside the brand's current distribution network, so we must break away from the relatively narrow competitive view of only looking at our own stores and use a broader perspective to observe whether there are more offline growth opportunities in the entire market. Because offline contributes over 75% of total sales, and in online consumer goods, what sells the most? Baby formula and diapers. But the sunflower seeds we eat, the milk we drink, and the laundry detergent we use at home—95% of those are still sold offline. So solving offline issues can drive the growth of the overall foundation.
To solve these problems, we need new innovative tools. If we still rely on our own sales reps, working hard every day to run and search, it's hard to achieve growth. Now there are many ways to capture these blank market opportunities, such as using mobile apps to quickly take a photo, or using store cameras and inspection robots to help us obtain market truth and data in a longer-tail, lower-cost way. In fact, there are many people in China, and everyone can become a data collection source. We can send an order telling them, "Go to the nearby community and check if a certain product is sold. If not, tell me the store addresses," so sales can quickly obtain this information and promptly distribute or replenish.
Let me share a real case: In 2016, we cooperated with a frozen food company. At that time, their executives discussed with us that all hypermarkets had been stocked, but growth was very sluggish. Where was the growth point? After a brainstorming session and a project test, they found that these frozen foods could also be placed in ice cream freezers. Where are these stores? They are small and medium-sized supermarkets. We used internet crowdsourcing to have ordinary consumers survey community stores to see if they had freezers. If a store had a freezer, it indicated a potential opportunity to place products there.
If a store had both a freezer and competitive brands, it meant the competitor's network had already entered, so we needed to accelerate our layout. If our products were already in the store, it meant our coverage was complete.
This formed a funnel network, **expanding our product distribution through store acquisition in short, medium, and long-term expansion paths. Through this method, we helped this company acquire about 30,000 to 50,000 new stores each quarter. This is a new growth opportunity achieved through marginal innovative technology.** **-02-**
Is there a more efficient way to acquire golden stores? We are in a technology era, and China's technology level is now very advanced. The following image shows the distribution of modern and traditional channels in Shanghai, as well as the distribution of dining outlets nationwide. We can directly extract this data, and the distribution opportunities are in these stores.
**These stores are classified by tags, and different important business districts are distinguished according to tag stores, such as which ones need investment first and where to deploy sales personnel first. We must first invest in the most important business districts, and then optimize the number of personnel in each district, rather than deploying the same number everywhere, which is inefficient. This solves the problem of where to sell all products.** Digitalization is most important for solving where to sell products. After humanized product design, through sales in sample stores, we build models, and then use characteristics such as the age structure and activity frequency of surrounding communities to complete an optimized ranking of all stores. Through algorithms, we identify the most important stores around communities.
Whether it's products or promotional expenses, we must first invest in these stores. We often see data in the market, such as two indicators: numeric distribution rate and weighted distribution rate. If you have 10,000 stores and you distribute to 5,000, the numeric distribution rate is 50%. But stores vary in size—large, medium, and small. If you distribute to large stores, your weighted distribution rate might be 80%. The best weighted distribution rate is the largest. This is the same message we are conveying: with a relatively limited number of stores, we must first control the most core and important golden stores. **Through golden stores, we use algorithms to plan resources based on people's information, life information, and preferences. What is the most important business district? What is the most important store? Where should we invest? This is the biggest role of golden stores: making resource planning very effective and orderly, rather than having sales reps go out and develop stores every day.**
Another issue: when there are many golden stores, we also need to ensure that in-store sales have good execution and recovery capabilities, and that SKUs with higher sales are in the store. If SKUs are not in the store or out of stock, is replenishment timely?
In exchanges with peers, we often see a figure: about 70% of product shelving and inventory out-of-stock issues frequently occur. Do you think your core SKUs are definitely in the store? Not necessarily. Two years ago, we cooperated with a daily chemical company. At that time, their baby products had been sold in China for 30 years. The client said all core SKUs were sold in major stores, but in reality, that wasn't the case.
Using the methods mentioned above, we analyzed the actual distribution of their 1,500 offline distributors. The final numbers were shocking: of the 14 core SKUs, only 7 were actually in stores. If consumers can't find the SKU they want in the store, they basically go to competitors.
Because China is vast and stores are widely distributed, product distribution is difficult. Additionally, for product re-shelving, 49% of merchants do not have good placement. Products placed in poor positions have a big difference in sales conversion. Another dynamic indicator, such as the location and quantity of secondary displays and promotional prices, shows that actual execution is not as high as imagined. This is the actual situation in stores.
By using fixed cameras, robots, and mobile apps, we can see the real situation in stores. Whether it's a large store or a small store, what happens in each store can be fed back in a timely manner. If there is a shortage or the product is not placed in a good position, the system will immediately send information to the regional sales manager and the head office, and improvements can be made.
**The benefit of new technology is that it allows for finer granularity in decision-making, from quarterly or monthly to weekly or even daily. When your decisions are faster, you gain growth opportunities faster.** Trax can not only identify products on the main shelf but also various secondary displays, such as floor stacks, box stacks, hanging strips, and POSM. These are the areas where input-output feedback is most needed, and this requires deep research into retail scenarios and continuous iteration of recognition capabilities.
Additionally, there are 3 million small stores in China, not just high-end hypermarkets. These small stores have many technical interference factors. For example, behind freezers there are often floor stacks. In the photos Trax collects, we use automated technology to segment scenes—identifying which are freezers and which are floor stack displays. After automatic segmentation, we perform automatic data statistics in these segmented scenes: what is the investment in main shelves, and what is the investment in floor stacks.
By using these technologies, we can discover market opportunities and execution differences. More importantly, through these innovative technologies and market insights, we can achieve refined management. **Previously, we only looked at static indicators, such as whether products are sold in stores. But now we are entering an era of refined management, capturing dynamic indicators with fine granularity.** For example, when buying milk, people look at the price and production date. In the store, if the date of Mengniu and Yili milk differs by one day, you might buy the fresher one. These changes in dynamic indicators determine the advantages and disadvantages in that store.
For FMCG, it's a satiety consumption need: you drink one box of milk a month; giving you two boxes is too much. So when monitoring static performance, feedback on dynamic indicator changes allows us to make decisions more quickly. If we find that competitors are doing special promotions in the store, we must quickly implement strategies to ensure our competitiveness is on par. This is what needs to be done in the "product" aspect.
The above mentioned seeing the truth of the shelf at the first time. Why do we need to see it at the first time? Trax previously did a dairy customer study. Consumers go to the store to buy things, with planned purchases and impulse purchases. The number of planned dairy purchases is 49%, and impulse purchases are 51%.
When planning to buy milk, do they necessarily want to buy a specific brand? For example, within the 49% planned purchases, the proportion planning to the category is 70%, and planning to the brand is 22%. This means consumers just want to buy milk; the specific brand is arbitrary. Only 22% specify a brand.
What can influence this action? Among different brands, it's the dynamic indicators mentioned earlier: in-store displays, promoters, and POSM price indicators. If we capture this information and feed it back to the brand, we can optimize actual execution, and ROI in a single store will improve. Whoever pays attention to these first will gain an advantage in terminal competition. **-03-**
Another aspect is product innovation. Many clients have created innovative products. China now has tens of thousands of new SKUs each year, but the success rate of new products is less than 6%.
That is, the actual success rate of new products is very low. Nielsen previously did a study: among brands with double-digit growth for two consecutive years, what were the growth drivers? For every 10 billion yuan in sales growth, 7% came from new products, 34% from media investment and brand communication, and 59% from excellent execution, in-store promotions and displays, leveraging new channel penetration, and deepening the broader market.
These contribute the majority of the market. **The vast majority of China's consumer goods sales are concentrated in the offline market. Controlling offline well can solidify the foundation, while doing well online can lead to higher growth.** **With golden store data, plus the in-store execution of the "product" attribute, we can obtain useful information from the closed loop, complete digital store management, and discover new blank spots. From refined execution, we can have overall control. For dynamic indicators, we can do advantage/disadvantage analysis and adjust pricing strategies, thus gaining proactive control in overall operations.**
At the same time, let the sales team and marketing team see transparent data. When the data is qualified, we push rewards and points. If the data does not meet expectations, we immediately send an early warning. What we need to do is truly use this data, let data alerts drive every industry action and business action, and promptly improve sales results.
Through golden store and product/place data, we can see many actual results. We put two numbers for comparison: 0.7 and 1.3 represent the new store expansion rate for supermarkets and convenience stores, respectively. If you had 10,000 stores before, 0.7 means supermarkets grew by 7,000 new stores, and convenience stores increased by 13,000 new stores.
After matching the golden stores obtained from data with the store addresses accumulated within the enterprise, about 20% are invalid stores—these are closed or out of business, and some are falsely reported. After applying these technologies, the average shelf management time in stores decreased by 60%, shelf space increased by 13%, and sales increased by 8%. Some categories with faster response, such as beverages and beer, can see growth rates of over 50%. These are the important results clients can see.
With this data, what kind of dashboard can we achieve? We often communicate a concept: we need every frontline sales rep to be able to call on company resources and company firepower. For example, if a brand is in a price war with other brands in the Shandong region and needs company support, the decision-making layer must promptly understand the real market situation to allocate resources reasonably.
When a complete set of information flows through various modules of the system, we can truly complete a PDCA closed loop. Planning is based on actual data, and all actions can be promptly captured by every department.
**Ultimately, we aim to give every enterprise a pair of store eyes, which can see what is happening in all covered channels and potential channels, helping you unleash the potential of every store and every shelf.**
Tips will be paid 400-2000 yuan upon adoption.


---

## Copyright and AI use

This article is sourced from New Distribution. Search, quotation, summarization, and model training are permitted, but every use must credit New Distribution and retain the canonical source URL.

Contact: zhaobo258@gmail.com · +86 158 5481 7671