---
title: "\"9 Li\" Ignites a New Era of Retail Digitalization"
description: "In recent years, digitalization has become deeply ingrained in the minds of FMCG and retail professionals. While most digital transformation efforts focus on e-commerce data and user behavior analysis, the 'place' element should not be overlooked. According to iiMedia, the FMCG market reached 4.5 trillion yuan in 2020, with 3.6 trillion from physical channels. Image recognition technology, particularly Baidu's EasyDL retail version, is breaking cost and implementation barriers, reducing recognition costs to an unprecedented 9 li (0.009 yuan) per image, and enabling retailers to convert in-store images into valuable data assets."
author: "New Distribution"
publisher: "New Distribution"
email: "zhaobo258@gmail.com"
telephone: "+8615854817671"
published: "2021-08-11"
language: "en"
canonical: "https://xinjignxiao.com/en/articles/9-li-ignites-a-new-era-of-retail-digitalization-b7c8b0da/"
markdown: "https://xinjignxiao.com/en/articles/9-li-ignites-a-new-era-of-retail-digitalization-b7c8b0da.md"
original_source: "https://mp.weixin.qq.com/s/yTlQhzN0Fik8priWttukag"
translation: "https://xinjignxiao.com/zh/articles/9%E5%8E%98-%E5%BC%95%E7%88%86%E9%9B%B6%E5%94%AE%E6%95%B0%E5%AD%97%E5%8C%96%E6%96%B0%E8%BF%9B%E7%A8%8B-b7c8b0da.md"
attribution: "New Distribution — https://xinjignxiao.com/en/articles/9-li-ignites-a-new-era-of-retail-digitalization-b7c8b0da/"
usage_policy: "https://xinjignxiao.com/ai-policy.txt"
---

# "9 Li" Ignites a New Era of Retail Digitalization

> In recent years, digitalization has become deeply ingrained in the minds of FMCG and retail professionals. While most digital transformation efforts focus on e-commerce data and user behavior analysis, the 'place' element should not be overlooked. According to iiMedia, the FMCG market reached 4.5 trillion yuan in 2020, with 3.6 trillion from physical channels. Image recognition technology, particularly Baidu's EasyDL retail version, is breaking cost and implementation barriers, reducing recognition costs to an unprecedented 9 li (0.009 yuan) per image, and enabling retailers to convert in-store images into valuable data assets.

In recent years, the term "digitalization" has been deeply imprinted in the minds of every FMCG and retail professional. A quick search for "retail + digital transformation" yields over 100 million results, and a brief browse reveals that most digital transformation efforts focus on e-commerce data and user behavior analysis.

Although FMCG retail has always revolved around the three elements of "people, goods, and place," with people being the primary element, "place" should not be overlooked in this wave of digital transformation.

According to iiMedia report data, the national FMCG industry scale reached approximately 4.5 trillion yuan in 2020, of which 3.6 trillion came from physical channels.

It is evident that although online channels are growing rapidly, offline channels are currently and will remain the foundation of sales composition and growth for consumer goods companies in the long term. The digital transformation of offline physical channels is also urgent for consumer goods companies.

**The First Step in Digitalization: Forming Data Assets**

The core of digital transformation is to form data assets, accumulate and utilize them, and ultimately formulate strategies and action guidelines for subsequent business operations. Millions of FMCG sales representatives travel to 8 million retail physical terminals nationwide every day, transmitting back massive amounts of image data.

How can these images be converted into corporate data assets, and then accumulated and utilized? Compared to traditional manual processing, image recognition is undoubtedly the most efficient and effective tool!

As the name suggests, image recognition simulates human cognitive processes through information processing and computer technology to identify objects in images and make meaningful judgments accordingly.

Image recognition has already been applied in many aspects of our lives: parking lot license plate recognition, security systems, smart logistics, production line quality inspection, and more.

In the FMCG and retail industry, image recognition enables rapid, accurate, and large-scale analysis and interpretation of in-store conditions such as distribution, display, and promotional information, all of which can be grasped by brand companies and retailers.

When these on-site images become specific data indicators, business decisions and management rules can be implemented with solid evidence.

**Bottlenecks in Image Recognition Application and Solutions**

Since image recognition is so effective for the FMCG retail industry, why hasn't its application been as deep and widespread as in other fields? Overall, there are two major obstacles facing brand owners and retailers:

**1. Recognition Cost**

When image recognition technology first entered the retail sector, it was still experimental, and technology companies quoted $1 per image. This sky-high price was only affordable for brands with global contracts.

Later, as domestic technology companies joined the development of image recognition, the cost dropped from $1 to 1 yuan, and now to about 0.1 yuan per image.

Although the reduction in recognition costs has encouraged some large brands to embrace new technology, even 0.1 yuan is still far from being affordable for large brands to use freely, let alone for many small and emerging brands.

On the other hand, among the current major domestic image recognition technology suppliers, a quote of 0.1 yuan is already considered very reasonable. At such times, a truly capable disruptor is needed to break the norm, and Baidu has taken on this role.

In 2017, Baidu launched EasyDL, a zero-barrier AI development platform based on the PaddlePaddle open-source deep learning platform, providing enterprise AI application developers with one-stop services such as intelligent annotation, model training, and service deployment.

Among them, EasyDL Retail Edition provides professional services for the FMCG retail industry, achieving low-cost, high-precision acquisition of product image recognition models, and completing intelligent in-store display and expense verification. With EasyDL Retail Edition, recognition costs are reduced to an unprecedented 9 li (0.009 yuan), directly reducing by an order of magnitude compared to market prices.

While offering competitive prices, Baidu's image recognition technology has consistently maintained its "engineering geek" practicality and excellence, providing technology products truly suitable for FMCG retail application scenarios: including a SKU recognition neural network specifically tailored for the consumer goods industry: SKU-Net, as well as recaptured image recognition, similar (duplicate) image detection, shelf stitching SDK, storefront text recognition, price tag recognition, shelf/freezer layer detection, display scene detection (main shelf, freezer, end cap, stack display, cut-case display, etc.). Among them, the stack display recognition model also supports statistical capabilities with spatial reasoning.

(Example of recognition results)

Supporting all this is Baidu's profound technical strength. In 2020 alone, Baidu ranked first in China with 9,364 AI patent applications, and its core R&D expenses accounted for 21.4% of revenue!

In fact, many brands have already used Baidu's image recognition technology in their operations. For example, through a box detection model with spatial reasoning capabilities, it supports a top domestic beer manufacturer in controlling the distribution progress and execution quality of contracted stores in its circulation channels, as well as obtaining accurate inventory information.

Combining storefront text recognition, SKU recognition, and recaptured image recognition, it supports a domestic baijiu manufacturer in the daily work of sales representatives, including routine store visits, new store additions, and in-store execution quality.

A top domestic dairy group has the group formulate unified marketing rules, and its subordinate branches each train their own SKU recognition models based on EasyDL Retail Edition, implementing the group headquarters' digitalization process with low cost and high efficiency.

**2. Implementation Difficulty**

Generally speaking, any technology requires an implementation step to truly function for the user.

For image recognition technology, the implementation content includes: in terms of implementation process, how to integrate into the collection end system (e.g., various SFA systems), how to complete image transmission, how to combine output recognition results with business management systems, data visualization presentation, and corresponding analysis and processing, etc.

In terms of implementation outcomes: SKU-level model training, model accuracy parameter tuning, model iteration, etc. All of this requires brand owners and retailers, as users, to have a professional technical team to complete the application implementation. However, this condition discourages many users who intend to use image recognition technology to improve efficiency and complete digital transformation.

Baidu EasyDL itself is a self-service open platform suitable for enterprises with certain technical capabilities to use independently. The platform shares pre-trained models and also includes a public library of some mainstream beverage and daily chemical SKU items, which can be used directly without SKU modeling training.

However, as anyone in FMCG retail knows, new product launches in the FMCG industry are continuous, massive, and rapid.

In addition, interface development with internal systems, API/SDK embedding with SFA systems, digital presentation and analysis of recognition results, and even some brands face the challenge of building and developing the front-end collection system from scratch. This is tantamount to requiring a brand owner or retailer to transform into half a technology company. For most customers, this is unrealistic.

Therefore, Baidu has partnered with ISV partners with considerable experience in various industries. Baidu provides strong core technology, while these ISV partners provide customized solution implementation for different customers. In the FMCG and retail field, Baidu's ISV partners include Xiaoling Technology.

At its inception, Xiaoling Technology leveraged its nationwide crowdsourcing network as its core capability, committed to providing consumer goods companies with extremely complete and accurate in-store information collection services. With its extremely deep and broad coverage, professional planning capabilities, and rigorous data verification capabilities, it has gained favor from a large number of consumer brands.

In the past two years, Xiaoling Technology has become a senior ISV partner of Baidu, joining forces with Baidu. Combining Baidu's technical capabilities with Xiaoling Technology's extensive experience in the consumer goods industry, they jointly provide customers with a full set of low-cost, high-precision image recognition services, from technology empowerment to solution implementation.

**What Else Can Retail Terminal Digitalization Do?**

After solving the cost and implementation bottlenecks, it is believed that the application scope of image recognition technology in the FMCG retail industry will expand rapidly. After solving the most basic digital conversion of real retail terminal images, let's take a step further and see what else we can do for the digitalization of the entire retail terminal channel.

As mentioned at the beginning of this article, there are approximately 8 million retail terminals nationwide, of which about 6.3 million are traditional small stores and small and medium-sized supermarkets. At the same time, the well-known large and medium-sized retailers have also been exploring the development of small formats in recent years. These traditional small stores are 75% concentrated in third-tier and below cities, contributing 40% of FMCG industry shipments.

For FMCG brands to maintain continuous incremental innovation in physical channels, making good use of retail channels in lower-tier cities has become a winning strategy.

However, where are the potential waist retail stores that can truly bring sales? How much potential do they have? What is the most efficient way to cover them? Baidu, in conjunction with Xiaoling Technology, has launched the "Smart Store Expansion" and "Smart Scheduling" products, which can precisely answer these questions.

Among them, "Smart Store Expansion" is based on massive data of retail outlet exterior and interior features, and through a sales potential scoring model, it provides targeted channel expansion strategies for consumer goods companies.

The "Smart Scheduling" product mainly uses map and traffic information big data, according to enterprise store inspection rules, and uses artificial intelligence models to calculate the optimal person-store ratio and optimize store visit routes, effectively improving the efficiency of frontline business teams.

*To learn more about Baidu's smart technology applications in the retail consumer goods field, feel free to scan the QR code or contact us by email:
Mail: ai-fmcg@baidu.com

**Are you "watching" me?**


---

## 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