Verifying the authenticity of outlets is part of the digital transformation of offline distribution channels for FMCG companies. However, traditional 'database collision' verification methods have shortcomings in ensuring authenticity. To address this, Baidu has developed a 'Four-in-One' solution to improve verification accuracy and reliability. In this context, Liu Xiaobin, head of FMCG at Baidu ACG AI Platform Department, introduced the deficiencies of traditional verification methods, the specifics of the 'Four-in-One' solution, and its positive impact on refined operations. The following is Liu Xiaobin's speech (partially abridged):
'Database Collision' Verification Has Vulnerabilities In the past year or two, while serving clients in outlet cleaning, deduplication, and authenticity verification projects, we discovered vulnerabilities in the 'database collision' verification method. The reason is easy to understand: the era when China's consumer goods market relied on demographic dividends for natural growth is gone, but the demand for growth will not disappear. Future growth will definitely rely on refined operations. The core of refined operations is customer assets, which are the 'outlets' of consumer goods companies. This is also an irreversible trend. Returning to the topic, to achieve the normalization of outlet authenticity verification, we first need to explain 'normalization.' The encyclopedia defines normalization as a state tending toward normalcy. In the FMCG industry, where offline stores change frequently and are complex, normalization should also imply 'becoming a daily routine' in addition to 'tending toward a normal state.' Since we aim to normalize outlet authenticity verification, but the current reality is 'non-normalized,' let's first understand the current industry solutions for outlet authenticity verification. Simply put, the current solution involves clients inputting existing outlet information, which is then processed and matched against an outlet database. If a match is found, we consider the outlet to be still real and valid. Because this solution involves operations between two databases, we commonly call it the 'database collision' verification method. Since we aim to normalize outlet authenticity verification, it means the current 'database collision' solution is not optimal. The reasons are mainly threefold:
First, the 'database collision' verification method is not based on the objective reality and truth. It is essentially algorithm-based matching of outlet names and addresses as text. In simple terms, we are verifying the 'description of the outlet' rather than the 'outlet itself.' Even if the confidence of matching information between two outlets is extremely high, can we be sure that the client's outlet still exists? In many cases, we cannot. The outlet may no longer exist in the objective world, but it still exists in the client's database and our own outlet database. Even if a match is found, the outlet has disappeared in the objective world, so from a business perspective, the match is definitely a failure. This is the first and most fundamental issue.
Second, for 'database collision' matching, clients need to input outlet information to match against our database. We know that most outlet information in client databases is manually entered by salespeople or scraped from systems. If the quality of the input information is low, the matching effect is even less guaranteed.
Finally, 'database collision' is a one-time or point-in-time action. After verifying outlet authenticity through 'database collision,' what does it represent? It only represents that the outlet information was valid at that point in time and before. Whether it remains valid afterward, and for how long, is unknown. Therefore, the 'database collision' solution is not a periodic or real-time feedback solution.
Based on the above three points, we believe that 'database collision' is not the best solution for outlet authenticity verification for FMCG companies. FMCG companies need a solution that is based on objective reality, does not rely on manual entry by salespeople, and can achieve low-cost, high-frequency, periodic verification. Based on this need, we developed a solution called 'Four-in-One,' which effectively addresses the business requirements for outlet authenticity verification.
The 'Four-in-One' Solution How to Verify Outlet Authenticity? We will introduce the solution from three aspects: design principles, model, and specific implementation.
First, the principle. As mentioned, the most suitable verification solution must be conducted in the real world. Therefore, the entire verification process should be front-loaded, not after data entry. It should achieve 'verification upon entry.' Shifting from 'post-verification' to 'pre-verification' is the core design principle of the Four-in-One solution.
Because verification is based on the real world, and in the real world, people must participate and implement it, we designed a business model that meets this principle. In the model, we reference the classic 'Internet human-machine verification' solution. We aim to turn every salesperson into a verifier of outlet authenticity, and extend every visit to include the benefit of outlet verification. The value of this model lies in its 'leverage' effect.
Based on the model, let's look at the specific implementation. As mentioned, our solution is called 'Four-in-One.' Why Four-in-One? Any outlet consists of four elements: store sign, store name, GPS, and address. Moreover, each outlet has one and only one set of 'associations' among these four elements. The core of the Four-in-One solution is to help clients generate an association relationship for each outlet and input it into their own database.
For a single outlet, Four-in-One is divided into two stages: 'unique verification upon first entry' and 'periodic verification during subsequent visits.' In both stages, the participants, i.e., the salespeople of consumer goods companies, only need to take a photo of the store sign. They do not need to do anything else; the remaining steps are handled by cloud technology. Therefore, the cost and threshold for implementation and deployment are extremely low.
Let's first look at the first stage, which is the unique verification upon first entry. Four-in-One uses an 'end-to-end' solution to replace the fragmented state of the previous four-element entry. Additionally, the entire solution must have certain risk resistance capabilities, such as checking whether the store sign photo actually contains a store sign, and whether the photo was taken in real time or imported from the phone's album. By preventing these potential loopholes, the Four-in-One solution can provide clients with a more authentic verification solution.
After generating the association relationship of the four elements through Four-in-One and entering it into the database, the first stage is complete. Since every outlet requires periodic visits, during the next visit, the salesperson again takes a photo of the store sign. We help clients extract the GPS information at the time of the photo. Based on a certain range around this GPS, clients can retrieve all stored store signs in the vicinity and compare them with the current photo. If a match is found, we consider the outlet to be still valid, and this process can loop indefinitely. Until one day no match is found, Four-in-One will issue a real-time alert, notifying the client that the outlet was not successfully retrieved and needs investigation.
The above is the complete process of the Four-in-One outlet authenticity verification solution and its control and monitoring of each outlet's lifecycle. The Four-in-One solution truly achieves low-cost, high-frequency, periodic freshness and verification of outlets. At the same time, because it is a technical solution, it can be well integrated with clients' SFA systems, truly achieving a one-time, permanent service effect. It truly realizes the 'normalization' of outlet authenticity verification.
Liu Xiaobin, head of FMCG at Baidu ACG AI Platform Department, joined Baidu in 2013 and has served as operations lead for Baidu's Sales Management Department, Marketing Center, and AI Open Platform. Since 2019, he has focused on leveraging Baidu's AI technology to drive the digital transformation and management upgrade of offline distribution channels for FMCG companies.