JPH06119309A - Purchase prospect degree predicting method and customer management system - Google Patents

Abstract

PURPOSE:To calculate the purchase prospect degree even for a customer whose customer attribute is partly deficient by estimating a defective customer attribute from the customer attribute of a similar customer. CONSTITUTION:First and second data files are stored within a memory 13. In the first data file, the attribute values of the customers who were made business objects in the past are stored without deficiency. In the second data file, the attribute values of the customers who will be made business objects are stored. The contents of the first data file is delivered to a predictive model preparing means 14 to which the data of object mechandise is imparted and a purchase prospect predictive model is prepared here. In an estimating means 15, defective attributes are estimated from customer attribute designation models, regarding the customers in the second data file to be a selection object. In a data processing means 16, the purchase prospect degree for every customer is calculated from the predictive models and the estimated attribute values and it is outputted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a purchase prospect prediction method and a customer management system.

[0002]

2. Description of the Related Art Conventionally, information groups that contribute to the purchase of products such as age, annual income, industry, family structure, savings amount, loan balance, etc., that is, customer attributes for each customer, are stored in a file as data. , As shown in FIG. 13, one customer attribute I considered to contribute to the purchase of the target product I
(For example, annual income) is set on the horizontal axis, another customer attribute J (for example, age) is set on the vertical axis, and a prediction model is created by mapping the expected purchase appropriate range based on these two customer attributes from past experience and the like. A system is known in which the purchase prospect of each customer is estimated by applying the customer attributes stored in a file to this model, and the customer is managed based on this result.

[0003]

However, the above-mentioned conventional system has a problem that when a customer attribute is partially missing, it may not be applicable to the prediction model. In addition, in the conventional system, since the characteristics of customers who purchased products in the past are not realized in the prospective prediction model, there is another problem that the prediction accuracy is low, and the purchase prospective degree is determined for each extracted customer. There is another problem that the sales priority cannot be determined because it is not available. An object of the present invention is to provide an improved purchase prospect prediction method and a customer management system using the same, which can solve the above-mentioned problems in the conventional system.

[0004]

A feature of the method of the present invention for achieving the above-mentioned object is that the missing customer attribute is used as a similar customer when predicting the purchase likelihood of the customer who sells the product. Of the customer who purchases the target product from the customer attributes of the already-purchased customer group, and predicts the purchase probability by comparing the similarity between the customer characteristic and the attribute of the target customer. There is a point in doing so.

Another feature of the present invention is a customer management system for managing a customer based on customer attribute data in the data file, the data file storing data relating to the customer attribute. Data creation means for obtaining customer characteristic data for purchasing a given product by multivariate analysis using customer group data that purchased the same product, and by multivariate analysis based on the customer attribute data without loss in the data file. Means for obtaining estimated data about missing customer attribute data for a required customer, means for creating a purchase prospect prediction model for the target product based on the customer characteristic data, and customer attribute data including the estimated data Of the required customer to the target product by comparing And data processing means for obtaining data about lies in that a management processor in response performs customer management process for customer management to the data processing means.

[0006]

[Operation] In the case of a customer whose customer attribute is partially missing, the missing attribute is inferred, and the product purchase probability is estimated by the product purchase prediction model considering the customer characteristics of the customer group who purchased the same product in the past. It is calculated. In the customer management system, required customers are managed according to the product purchase prospects obtained as a result.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 shows an embodiment of a customer management system according to the present invention. In the customer management system 1 of FIG. 1,
2 is a computer, 3 is required data for the computer 2
Is an input device for inputting data into a CRT display device, 5 is a printer, and 6 is an address printing machine which operates in response to data obtained by a computer. Computer 2
Is a central processing unit (CPU) 7, a random access memory (RAM) 8, a read only memory (ROM) 9,
I connected to the communication control unit 10 and various external devices
The I / O interface device 11 is configured as a computer system having a known configuration in which the I / O interface device 11 is connected to each other via a bus 2A.

A data file in which a large number of customer attribute values are stored in the RAM 8 is created in the ROM 8.
A customer management program for processing the data in the data file and managing the customer is stored in the store 9. The contents of this data file can be added, updated, and corrected by the input device 3, and in the illustrated embodiment, the first data file in which the attribute values of customers who have been sales targets in the past are stored without any loss, It is divided into a second data file that stores the attribute values of customers who are to be sold in the future. The file may be created in the RAM 8 or may be created on a magnetic tape by providing a magnetic tape reader as an external memory device.

The customer management program stored in the ROM 9 is executed by the CPU 7, and the purchase prospect degree of the target product input from the input device 3 is determined by using the contents of the data file stored in the RAM 8. The method predicts each customer and processes the result for customer management.

FIG. 2 is a block diagram showing the execution contents of the above-mentioned customer management program as a functional diagram. A first data file and a second data file are stored in the memory 13. The content of the first data file is sent to the prediction model creating means 14 to which the data regarding the target product is given, and the purchase prospect prediction model is created here. On the other hand, the estimation unit 15 estimates the missing attribute from the customer attribute estimation model. The data processing means 16 performs data processing for calculating the purchase likelihood for each customer from the prediction model and the estimated attribute value, and outputs data indicating the purchase likelihood.
This output data is processed by the output control unit 17 for customer management, and an external device for customer management, for example,
The data is output to a required one of the CRT display device 4, the printer 5, the address printer 6, and the communication control unit 10 having an automatic dial function connected to the public line 12 and used for customer management.

Next, in order to realize the function shown in FIG.
The flowchart of the customer management program stored in M9 will be described with reference to FIGS.

In FIG. 3, initialization is performed in step 20 and then step 21 is entered in which required data such as data indicating a target product necessary for management is input and stored in the RAM 8. . Then step 22
Enter, where the estimation model is created.

FIG. 4 shows a detailed flowchart of step 22. First, in step 221, customer attribute items to be estimated are set. In the illustrated example, annual income is set as this item. In the next step 222, the annual income, which is the customer attribute item to be estimated, is set to 1 million yen, 2 million yen ,. ． ． Categorize as follows.
Then, the procedure goes to step 223, and the customer attribute items to be estimated are discriminated and analyzed using the customer attributes other than the annual income to be estimated as explanatory variables.

In this discriminant analysis, as shown in FIG. 8, by taking appropriate indexes A and B on each axis, each category (1 million yen range, 2 million yen range, ...) Is optimally separated. It is for doing. As these indicators A and B,
In the example shown, age and education are used (step 224). If the categories are optimally separated in this way, the center point a of the sample data of each category,
b, c ,. ． ． To decide. In this way, the index A,
When the estimation model (see FIG. 8) for estimating the annual income is created from B, the next step 23 is entered.

At step 23, a purchase prospect prediction model is created from past data on the purchase of the target product. This creation procedure will be described with reference to the detailed flowchart of FIG.
Then, all customer attribute items (age, annual income, ...) Are divided into the required categories based on the customer data in the first data file, and the customer attributes classified in the next step 232 are used as explanatory variables for the target product. Discriminate and analyze whether or not purchased.

Therefore, as shown in FIG. 9, indexes C and D for optimally separating the purchaser and the non-purchaser are appropriately determined (step 233), and the sample data of the purchaser and the non-purchaser are sampled. Determine the center points α and β. In this way, when a prediction model for predicting the degree of purchase or non-purchase, that is, purchase prospect from the indices C and D is created, Step 2
Enter 4.

At step 24, the value of the missing attribute is estimated for the customer in the second data file to be selected. In this case, the annual income is a missing attribute, and the missing attribute item is extracted in step 241 of FIG. In the next step 242, since the annual income of the target customer is estimated by applying it to the estimation model obtained in step 22, the index A (age) for this customer,
And each value of the index B (educational background) is calculated.

FIG. 10 shows how the estimated point X is obtained from the value Ax of the index A and the value Bx of the index B of the customer.

The calculation for this estimation is step 243.
Then, the confidence is obtained for each category of annual income which is an attribute to be estimated. That is, the estimated point X (see FIG. 10) and the center points a, b of each category,
c. ． ． And ax, bx, cx. ． ． If so, the confidence level Z100 of 1 million yen level is Z100 = (ax / k) × 100 (%) The confidence level Z200 of 2 million yen level is Z200 = (bx / k) × 100 (%) 3 million yen level Z300 = (cx / k) × 100 (%) ... However, K = ax + bx + cx + ...

These confidences can therefore be expressed, for example, as shown in FIG. After that, step 25 is entered, where the purchase prospect degree is calculated.

FIG. 7 shows the steps of calculating the purchase probability. In Step 251, Step 23
In order to calculate the purchase likelihood degree using the prediction model obtained in step 1, first, each value of the indexes C and D is calculated from the attribute value of the customer to be selected. Next step 252
In, using the calculated values of the indexes C and D,
First, the estimated point Y in the CD plane in which the indices C and D are taken on each axis is determined (see FIG. 12). Thereafter, the purchase prospect degree PRO is calculated as PRO = αY × (αY + βY) ⁻¹ × 100 (%). Where αY is the distance between α and Y, β
Y is the distance between β and Y.

In this way, the value of the purchase likelihood degree PRO is calculated for all the customers to be selected, and step 26 is entered. In step 26, the purchase probability P
A printer 5 that enables the CRT display device 4 to confirm the name, address, and telephone number of a customer whose RO value is a predetermined value or more.
Output processing for customer management such as output by, label printing of name and address by an address printing machine, automatic dialing of a customer's telephone number in sequence, etc. is executed. What kind of customer management processing may be performed may be predetermined in the program, or may be instructed by the input device 3 each time. In the case of the application to direct mail telemarketing described in the above embodiment, there is a special effect that the sales priority can be determined by narrowing down the customer even if the customer attribute is partially missing. There is.
The form of customer management according to the present invention is not limited to the above-mentioned example, and the present invention may be applied for other appropriate management.

Further, the present invention can be applied not only to the above-mentioned direct mail telemarketing but also to various other business activities. For example, application to product analysis such as analysis of top-selling products based on the number of expected customers in the sales area, application to store quality analysis such as analyzing the quality of the store by analyzing the best-selling products in the store. It is most suitable for application such as customer stratification analysis that performs customer stratification based on product prospects. Even in these applications, even if some of the customer attributes are missing, it can be estimated, so that the customer attribute missing data, which was previously difficult to use, can be effectively used, and various analyzes can be performed widely. Can be done.

[0025]

As described above, according to the present invention, the purchase prospect degree can be calculated even for a customer whose customer attribute is partially missing, so that the present invention can be widely applied to the customer. Since the past purchase status of is realized in the purchase forecast model, it is possible to make a more accurate purchase forecast, and by performing customer management based on the purchase forecast thus obtained, the sales Activities can be performed more efficiently.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram showing an embodiment of a customer management system according to the present invention.

FIG. 2 is a functional block diagram showing the contents of data processing executed by the computer shown in FIG.

FIG. 3 is a flowchart showing a customer management program executed by the computer shown in FIG.

FIG. 4 is a detailed flowchart of an estimation model creating step of FIG.

5 is a detailed flowchart of the prediction model creating step of FIG.

FIG. 6 is a detailed flowchart of a missing value estimation step in FIG.

FIG. 7 is a detailed flowchart of steps of calculating a purchase likelihood degree in FIG.

8 is an explanatory diagram for explaining an estimation model creating step in FIG. 3. FIG.

FIG. 9 is an explanatory diagram for explaining a prediction model creation step of FIG. 3.

10 is an explanatory diagram for explaining a missing value estimation step in FIG. 3. FIG.

11 is an explanatory diagram for explaining the certainty factor obtained in the missing value estimation step of FIG. 3. FIG.

FIG. 12 is an explanatory diagram for explaining calculation of a purchase likelihood degree in FIG. 3.

FIG. 13 is an explanatory diagram for explaining a conventional purchase prediction method.

[Explanation of symbols]

 1 Customer Management System 2 Computer 3 Input Device 4 CRT Display Device 5 Printer 6 Address Printing Machine 10 Communication Control Unit 13 Memory 14 Prediction Model Creating Means 15 Estimating Means 16 Data Processing Means 17 Output Control Unit

Claims (2)

[Claims] 1. A method of predicting a purchase likelihood of a customer who sells a product, in which a missing customer attribute is estimated from a customer attribute of a similar customer, and the target product is purchased from a customer attribute of a group of already purchased customers. A method of predicting a purchase probability, wherein the purchase probability is predicted by extracting the characteristics of a customer who makes a purchase and comparing the similarity between the customer characteristic and the attribute of the target customer. 2. A customer management system for managing a customer based on the customer attribute data in the data file, comprising a data file in which data relating to the customer's attribute is stored, and purchasing a given target product. Data creating means for obtaining customer characteristic data by multivariate analysis using customer group data that purchased the same product, and loss for a required customer by multivariate analysis based on the customer attribute data without loss in the data file Means for obtaining estimated data about customer attribute data, a means for creating a purchase prospect prediction model for the target product based on the customer characteristic data, and comparing customer attribute data including the inferred data with the prediction model. By doing so, data regarding the degree of purchase prospects of the required customer for the target product can be obtained. Customer management system, wherein the order of the data processing means, further comprising a management processor which in response performs customer management process for customer management to the data processing means.