The Supply Chain concept is widely used by leading companies in the field of logistics. In this context, the interfaces between the company, suppliers and customers are observed, both in terms of product, information and financial flows.
The conceptual model of the integrated Supply Chain decision process was proposed by Bowersox & Closs and is schematized in Figure 1.
In this model, the structuring of the logistics system starts with 4 decisions taken with marketing professionals:
 |
- Product: What products will be offered for sale? What is the mix of products to be placed in each sales region?
- Promotion: What promotion policies and discounts will be given for each product in each region?
- Price: What price level will be charged for each product in each sales region?
- Place/Customer Service: How will the product be distributed? Which locations will be served? What is the minimum requirement level for service? What is the service frequency? What is the service level in terms of product availability that will be practiced for each region/type of customer?
Thus, based on the definitions taken above, all logistics functions will be planned and structured to meet the different levels of proposed service. In addition, our goal is to meet the different levels of service at the lowest possible total logistical cost. A lower total cost means a simultaneous reduction in inventory, transportation, warehousing, purchasing, and technology use costs without the service level depreciation. A change in a certain logistical cost directly influences another. For example, minimizing the cost of purchasing by purchasing raw materials in greater volume can increase the cost of warehousing.
We note then that the first phase, that of “Knowing the consumer”, is fundamental for a good structuring and good performance of the logistic system. However, designing a consumer-based system is becoming increasingly difficult. Mainly for those professionals in the Logistics area who work in the retail sector. With the increase in competition and the emergence of a new form of commerce via the Internet, consumers have become more numerous and difficult to interpret.
How to respond to Product, Price, Promotion and Place questions so that the logistics organization is properly and efficiently structured?
Professionals in the logistics area have two concepts that are currently widely used in the retail and service banking sectors to help with these issues: Data Warehousing and Data Mining.
- WHY USE SUCH TOOLS?
We have noticed a reduction in the prices of Point of Sale (POS) terminals in recent years. Such equipment, in addition to speeding up the payment operation at check-outs, can be used to collect sales information. Coupled with the reduction in the price of computers and the reduction in the cost of storing data, what we notice is that most retail companies today have the possibility of accumulating sales and consumer information at an affordable cost. On the other hand, massive data collection alone does not contribute to leverage the company's marketing strategy. What currently happens for professionals in the area is access to a large volume of data but the difficulty in extracting information for decision making. Such information can be excavated through Data Mining tools.
Before going into detail on the concepts of Data Warehouse and Data Mining, it is interesting to show some success stories with the use of these two concepts:
- Wal*Mart is one of the largest retail chains in the United States. It is known for its policy of low inventory levels and constant resupply of products (low lots and high frequency) in addition to its aggressive policy with regional competitors. Using Data Mining tools that help forecast each item for each store in the company, it modified its automatic product resupply systems. In addition, it identified consumption patterns in each store, in order to choose the mix of products to be placed.
- ShopKo, an American retail chain, used Data Mining tools to determine which products are sold through the indirect sale of other products. As a result, it withstood competition from Wal*Mart in 90% of markets and increased its sales.
- Banco Itaú used to send more than 1 million direct mails to account holders, with a response rate of 2%. With a database containing the transactions of its 3 million customers in the last 18 months, using Data Mining tools, it reduced its postage bill by a fifth and increased its response rate to 30%.
- US phone companies saw a 45% reduction in service fees with new customers using personalized direct mail with Data Mining.
- DATA WAREHOUSE AND DATA MINING
But how did these companies in the retail sector achieve such gains? How did you structure your supply and distribution networks to meet different market demands? Through the creation of a Data Warehouse and the use of Data Mining techniques to discover information in its large data set.
A Data Warehouse is an integrated repository of information, available for analysis and for building search filters (queries). Such information is collected from heterogeneous sources of operational data and gathered in a database, centralizing information located in different sources and allowing them to be shared throughout the company. In addition, external sources can be included in a Data Warehouse, in addition to the company's operational data, such as consumer demographic information and personal information for each customer.
Data Mining, on the other hand, is a methodology that seeks a logical or mathematical description, possibly of a complex nature, of patterns and associations in a data set. In the context of data mining, learning is a set of techniques that perform two main tasks: (1) generalize rules across a set of known examples and (2) detail a structure of their conclusions. Among the tools or technologies used to implement a Data Mining project, we highlight: Neural Networks, Decision Trees, Time Series Analysis, Genetic Algorithms, Hybrid Approximations, Fuzzy Logic and Conventional Statistical Tools.
In order for Data Mining techniques to be used, it is generally necessary for the company to have a Data Warehouse, that is, a database that gathers past information from its operational activities. This is necessary because Data Mining techniques generalize patterns through known past results. We can say then that a Data Warehouse is a requirement for the implementation of a Data Mining project. Figure 2 summarizes the relationship between Data Warehouse and Data Mining.
 |
- EXAMPLES OF DATA MINING APPLICATIONS IN THE RETAIL SECTOR
The retail sector is one of the sectors where Data Mining techniques are most used. This is because it is an industry where there are several complexities inherent to sales. The sale of each item is sensitive to market factors (advertising and price) and external factors (fashion trends, income, competition). In addition to these factors, some products indirectly influence the sale of other products. The result of the interaction of all these factors and their relationships are complex to be translated through common analyses.
Thus, Data Mining techniques emerge as an alternative for dealing with these types of complexities. Furthermore, the basis for its application, data collection, is already accessible to many companies in the sector. Tables 1 and 2 below show the cutting-edge use of Data Mining in Retail and the companies representing its use in the US retail sector.
 |
Data Mining applications in retail can be grouped into two main categories: Sales Forecasting and Leveraging Marketing Strategies.
4.1. Sales forecast
The application in sales forecasting consists of determining a time series using past data. This topic has already been widely addressed by statisticians, mainly with the use of linear methods. However, when phenomena are very complex and non-linear, commonly used methods are not sufficient. In these cases, adaptive methods such as neural networks and induction trees have been used.
Regarding induction trees, they are algorithms that explore past sales data to extract prediction rules for the future. The principle of induction is relatively simple, but its application to sales forecasting can become complex. A neural network, on the other hand, is a function approximator and produces as output a function that minimizes a given parameter. Through known historical values, a network of functions is created representing the relationships between variables. Both for the optimization of induction trees and neural networks, genetic search algorithms are used.
Through the use of these tools, the logistics professional can have greater accuracy in forecasting time series of various products. In this way, it can anticipate decisions such as: the creation of stocks of products in specific regions, continuous shipment of products, creation of points and resupply of products or changes in pricing and promotion policies. With preventive rather than reactive decisions, the logistics professional can seek to minimize the total logistics cost by practicing an acceptable service level.
4.2. Optimization of marketing strategies
On the other hand, the application of Data Mining in the optimization of marketing strategies is a more complex conceptual problem, based on the ability to make predictions under different conditions and on the generation of business models. We have 5 main points of application:
- Direct mailing: Direct mail campaigns are expensive, it is important to narrow the range of all consumers into subsets of potential customers. This has been optimized using induction trees and neural networks.
- Market segmentation: is a broader problem of recognizing specific patterns of marketing segments that respond to a given characteristic. In this way, market segments with the same characteristics are identified. By identifying these segments, different pricing, marketing and service policies can be implemented. For the logistics professional, this information is extremely important. Different service levels can be determined for each market segment, in addition to the minimum acceptable service level for each region.
- Determination of consumer profiles: consumption patterns are identified. Such profiles can be used both for targeting sales promotions and for planning the layout of products at points of sale (which products to place next to each other). Analogously to the previous item, demands by service level are identified for each type of customer. Some customers are not willing to pay more for a better level of service, being satisfied with the level of service practiced.
- Sensitivity analysis: consists of evaluating the impact of changing one variable on another variable, such as the price elasticity of demand. If a large amount of historical data is available, calculating sensitivities is a problem of generalizing a function of a number of known values. In this way the impact of a price change on demand can be predicted and various analyzes can be performed.
- Product Category Management: consists of determining how products are sold in each store, how sales occur in relation to shelf placement, how promotions can be optimized, and which items should be placed side by side in layout combinations. This is a complex analysis of the product mix and its layout in the store.
- CONCLUSION
With the reduction in the price of data collectors for Points of Sale in the retail sector, the explosion and cheaper use of computers and the reduction in the cost of data storage, it is currently possible to implement a database (Data Warehouse) with the movements and characteristics of its consumers. Through the use of Data Mining techniques it is possible to obtain information from this large set of data, discovering relationships between variables and patterns in the huge database. Professionals in the field of logistics can use this information to direct marketing strategies. Through these directions, the entire logistics organization and its operational requirements will be defined, aiming to meet the different levels of service at the lowest possible cost.
In an increasingly competitive sector, knowing the consumer is still the greatest formula for success.
- BIBLIOGRAPHY
http://www.datawarehousing.com
http://www.data-warehouse.com
http://www.datamining.org
