Recently, China's leading industrial digital research and consulting organization iAnalysis released the "2022 iAnalysis·Artificial Intelligence Application Practice Report". The business intelligence decision-making platform built by Weizhi Technology for Meiyijia was selected as a typical case in the report. This case is also a representative of high-maturity AI solutions in the intelligent transformation of enterprises.
Ai Analysis pointed out that the implementation of the artificial intelligence industry generally faces the lack of standardized methods and automation capabilities in data collection, processing and management for AI development, and the lack of methods to locate high-value application scenarios.
Weizhi Technology's full-stack spatiotemporal AI technology system first solves the problem of insufficient "depth" and "breadth" of data through data spatiotemporalization, updates the data architecture, improves data quality, and creates a data foundation to support intelligent transformation. Secondly, a large-scale knowledge network centered on the industrial chain and enterprise map is constructed using a spatio-temporal map, and large-scale, real-time and dynamic correlation analysis is realized around the key indicators of enterprise ecology and enterprise business development. Finally, we deeply explore business scenarios and use graph intelligence to solve the closed loop of business decision-making in high-value application scenarios. Among them, the Meiyijia business intelligence decision-making platform provides effective services in terms of door rating, sales forecast, attribution analysis, operation optimization and marketing optimization. business decision-making guidance.
Case sharing:
Case: Meiyijia builds a business intelligence decision-making management platform to realize intelligent operations of offline retail
Meiyijia Holdings Co., Ltd. is the second largest convenience store chain group in China. Since its establishment, Meiyijia has taken Guangdong as the center and gradually deployed its business nationwide. At present, the group has more than 20,000 Meiyijia brand stores, with a total daily customer flow of more than 25 million.
The core logic of offline retail is to connect "people" and "goods" with "field" as the center, and the respective characteristic elements of "people and goods field" and the relationship between them affect store operation strategies and results to varying degrees. . For Meiyijia, which is accelerating the expansion of stores in East China, North China and other regions, after early informatization construction, it has built ERP, PIM, BI and other business systems, and has accumulated rich internal "people and goods field" data. At this stage, Meiyijia hopes to use intelligent methods to analyze and optimize the characteristic elements and relationships of people, goods, and places to support the steady expansion and refined operations of stores.
Combining factors such as business needs and the feasibility of intelligence, Meiyijia decided to introduce intelligent solutions in three important scenarios: store location selection, store operations, and marketing optimization. In terms of store location selection, Meiyijia needs to use intelligent methods to determine the appropriate store expansion area, consumer habits in the area, and its own positioning in the city; in terms of store operations, it specifically includes store evaluation, product-store matching, Application scenarios such as sales forecast and competitive product analysis help Meiyijia better formulate business strategies and improve store performance; in terms of marketing optimization, it needs to be combined with crowd targeting to optimize advertising strategies.
In order to achieve intelligent decision-making in the above scenarios, Meiyijia decided to build a business intelligence decision-making management platform. However, Meiyijia has deficiencies in the completeness of relevant data and AI algorithm capabilities. The details are as follows:
1) Data level: Meiyijia only masters in-store operating data and back-end data of its own supply chain system, but lacks external Data related to geographical location, surrounding crowds, and surrounding competing products. At the same time, when exploring solutions with external vendors, Meiyijia needs to ensure the privacy and security of its data when sharing internal data.
2) AI algorithm level: Intelligent analysis of location selection, product selection, consumer profiling, and marketing requires the support of a large number of AI models. Meiyijia needs external manufacturers to provide relevant AI models and use the models to solve business problems. In addition, the data distribution in different regions will be quite different. In many cases, general models cannot be directly applied, and the model needs to be adjusted and migrated based on factors such as region and environment.
Based on spatiotemporal data and supported by AI models, Meiyijia provides intelligent decision-making for all aspects of store operations.
After evaluating the manufacturer's capabilities in data, AI algorithms, application solutions, etc., Meiyijia chose to cooperate with Weizhi Cooperate in science and technology to build a business intelligence decision-making management platform. Weizhi Technology is a spatio-temporal artificial intelligence platform provider that focuses on the digitization of offline scenes and the spatio-temporal integration of online and offline. It uses spatio-temporal AI technology to create digital twins to provide services to cities, transportation, finance, real estate, retail and brands. Refined scene services and intelligent solutions.
Based on the Phy-gital spatio-temporal intelligence platform, Meiyijia has built a business intelligence decision-making management platform. Its architecture is divided into three layers: the bottom layer is a joint data warehouse, which includes the spatio-temporal data provided by Phy-gital Technology and the data provided by Meiyijia. Store data; the middle is the technical service layer, including AI models, spatiotemporal knowledge graphs, business calculation logic, etc.; the upper layer is various analysis applications, including store location ratings, store performance ratings, attribution analysis, crowd portraits, sales forecasts, prices Forecasting, competitive product analysis, etc.
Figure 4: Meiyijia Business Intelligence Decision Management Platform Architecture
In response to data-level issues, the platform integrates Weizhi Technology’s basic marketing intelligence data and Meiyijia’s store data through a joint data warehouse. Weizhi Technology's marketing intelligence basic data includes data on passenger flow, portraits, crowd flow preferences, surrounding ecology, traffic conditions, business environment, competition and cooperation relationships, etc. Specifically, it includes road network, transportation, AOI (block), POI ( Points) and other static data, as well as surveying and mapping-related dynamic data such as people flow, portraits, scenes, enterprises, and the economy. On this basis, a spatio-temporal knowledge graph is formed by sorting out the feature management among "people, goods, and places". In order to protect Meiyijia's data privacy, Weizhi Technology combines its spatiotemporal data and model services with Meiyijia's internal data, and deploys it in the form of an all-in-one machine in Meiyijia's restricted environment. Model construction, training, Management and publishing are all completed on an all-in-one machine to ensure that Meiyijia’s data does not leave the database.
Figure 5: Business intelligence decision management platform pre-training model warehouse logic
For issues at the AI algorithm level, the platform is based on technical models such as classification, sorting, clustering, anomaly discovery, regression prediction, and abductive reasoning analysis, combined with services such as feature engineering and automatic feature calculation, to form portraits, site selection, product selection, and marketing Predictive analysis models in other aspects. At the same time, for the problem that general models need to be adapted to changes in regional data characteristics, Wisdom Technology performs model migration by matching adaptable use cases or features, and solves the problem of small data volume and sample problems through data enhancement or small sample learning. Sparse problem.
The value and effect of the implementation of the business intelligence decision-making management platform
Meiyijia implemented the business intelligence decision-making management platform, using data intelligence to empower store development and operations, and achieved the following business values.
1) Store location selection: The platform provides data support and intelligent decision-making basis for Meiyijia to select locations and expand stores in Guangdong and East China. Specifically, by analyzing the habits of users in the current area, the habits of permanent and mobile people, the characteristics of the site itself (surroundings are hospitals, communities, commercial centers, etc.), the surrounding competition situation (supply and demand saturation), transportation convenience, etc. Use data to determine whether a certain location is suitable for opening a store, and provide a score for the store in that location and the factors that affect the score. This greatly improves the efficiency and accuracy of decision-making compared to the traditional method of relying on research and expert experience, and can support the decision-making of opening and closing stores of Meiyijia's 3,000 stores every year.
2) Store operations: The platform effectively helps Meiyijia brand stores and franchise stores totaling 40,000+ stores to conduct digital and intelligent operations. In terms of store ratings, sales forecasts, attribution analysis, operations optimization and marketing optimization, offline big data and intelligent prediction results are provided to assist decision-making and effectively guide store operation optimization.
Summary of the project experience of the business intelligence decision management platform.
First, the offline retail industry involves a large amount of data, many dimensions, and relatively frequent dynamic updates. These data need to be relied upon in site selection, product selection, supply chain distribution, and marketing. Decisions are made in relatively real time, so the complexity and difficulty are great. However, the core of retail data still revolves around people, goods and places and their relationships. When faced with complex data scenarios, you can consider using knowledge graph technology to sort out data in each dimension by people, goods and places, and then form places and places, people and places, and goods. Relevance to the field; at the same time, consider using AI models to describe complex change patterns, analysis ideas, and decision-making experiences.
Second, only in-store data cannot achieve intelligence in offline retail scenarios. Therefore, it is also necessary to introduce spatio-temporal data outside the store and combine external data such as geographical location, surrounding crowds, and surrounding competing products with internal data to achieve 1+1> 2 effect. Moreover, when combining internal and external data, it is necessary to consider ensuring the privacy and security of the data through corresponding technologies or product design methods.