KR100713205B1 - Online customer scoring system based on dynamic customized model - Google Patents

Abstract

The present invention relates to a customer scoring technology, specifically, in order to provide basic functions (customer segmentation, customer index calculation, etc.) that the customer scoring technology should have, which can be applied to an actual industry based on a dynamic customer index specialized for the customer segment. An economic scoring system. The scoring system according to the present invention comprises: (1) means for a user to set an item necessary for calculating a customer index and its weight, and (2) an operating system for calculating a customer index by the data manager based on the customer index item and the weight. Means for extracting data, (3) means for calculating customer detailed index data using extracted operating system data and items and weights selected by a business user, (4) means for calculating customer indexes based on extracted customer index data; (5) It includes a means for utilizing the calculated customer index.

Customer scoring, customer index, dynamic index

Description

Online customer scoring system based on dynamic customized model

1 is an overall configuration diagram of a system according to the present invention.

FIG. 2 is a flowchart for explaining the structure of FIG. 1 functionally in more detail. FIG.

3 is a graph showing an acceleration change trend.

4 is a screen diagram illustrating application of a conversion probability.

Figure 5 is a trend analysis graph for showing the conversion probability application in a tabular form.

6 is an exemplary view of a UI screen for querying and using a customer index.

7 is an exemplary view of a UI screen for mileage allocation according to the customer index.

The present invention relates to a customer scoring technology, specifically, in order to provide basic functions (customer segmentation, customer index calculation, etc.) that the customer scoring technology should have, which can be applied to an actual industry based on a dynamic customer index specialized for the customer segment. An economic scoring system.

Customer Scoring is a credit management tool that enables you to measure risks and respond appropriately to identified risks. Technology has evolved around credit scoring, which assesses customer credit. The current state of technology development in is as follows.

-Purpose: mainly for the purpose of loan suitability assessment, bankruptcy prediction, credit assessment, insurance fraud detection, and departure analysis, mainly in the financial and telecommunications industries such as banking, insurance, and securities

-Process: Based on the accumulated characteristic data of the customers, classify the customers into good or default groups, and compare the characteristics by each group based on the scorecard that embodies the credit risk. Follow the process of scoring and evaluating current customers

-System: Credit Scoring System (CSS) and data mining solutions of financial sector are in use, especially mining solution is focused on elaboration and prediction of evaluation model, mainly foreign companies such as SAS, SPSS, IBM, etc.

Currently, in Korea and overseas, products that evaluate customers in terms of various indices, rather than customer evaluation in terms of credit management, are inadequate.

The present invention is a technology developed to provide basic functions (customer segmentation, customer index calculation, etc.) that customer scoring technology should have as an economical product applicable to real industry mainly on dynamic customer index specialized for customer segment. Provided as a product.

In addition, the present invention, in addition to the static index in the calculation of the customer index, applying the dynamic index (conversion acceleration, conversion probability, etc.) to predict the conversion possibility, it is possible to segment customers based on this. Furthermore, we presented the data to develop a marketing program for each situation.

The overall configuration of the system according to the present invention is composed of four functions as follows: 1) Setting items and weights necessary for calculating the customer index 2) Extracting operating system data 3) Extracting customer detail index data 4) Calculating the customer index.

1 is a block diagram showing an overall overview of a system according to the present invention. Each component is described as follows.

(1) a means for a business user to set items necessary for calculating a customer index and weights thereof (100)

The components of the detailed items (detailed indices) required for the calculation of the customer index, measurement method, and score setting (scoring) for each index are summarized in the table below. Table 1 shows the detailed index and measurement method of the registered index item, Table 2 shows the detailed index and measurement method of the visit index item, Table 3 shows the detailed index and measurement method of the response index, and Table 4 shows the purchase index. Detailed index and measurement method are shown.

No. Detail Index How to measure scoring S1 Basic Important Information One of social security number, name, password, authentication password, or email 0 or 1, 0-5 points distribution S2 Basic personal information One of Home Phone Number, Work Phone Number, Mobile Phone Number, Home Address, or Work Address 0 or 1, 0-5 points distribution S3 Additional personal information Date of birth, lunar-solar calendar, vehicle ownership, vehicle number, wedding 0 or 1, 0-5 points distribution S4 Job Additional Information Occupation code, position code, duties code, industry code 0 or 1, 0-4 points distribution S5 Family info -Number of registered family members (0.6 points for one person, 1.2 points for two or more persons),-1 item among family resident registration number, relationship code, family name, householder status, householder relationship * 0.6-Family email score of 0.8 0.6 or 1.2, 0 to 3, 0 or 0.8, 0 to 5 point distribution S6 Receive Email Information 1 item among product information receiving setting, newsletter receiving setting, service mail receiving setting, event mail receiving setting * 1.25 0 or 1, 0-5 points distribution S7 Interest information One of travel interest, car interest, health interest, women's interest, or technical interest 0 or 1, 0-5 points distribution S8 Anniversary Information Anniversary Days (e.g. 1 point, 2 points 2, ...) 0-5 points distribution S9 Additional Information Measured as (current-last modified date) (10 points for less than 2 months, 9 points for 2-4 months, ..., 0 points for more than 20 months) 0-10 points distribution Sum 49 points

No. Detail Index How to measure scoring S1 Weekly goal visits Number of visits to the target site for one week (0 for 0 times, 1 for 20% after shooting, 2 for 20-40%, 5 for 80 to 100%) 1 to 5 points distribution S2 Weekly target stay time Length of stay at the target site for 1 week (0 hours is 0 points, 20% after shooting 1 point, 20-40% 2 points, ..., 80-100% 5 points) 1 to 5 points distribution S3 Weekly induction visits Number of visits to induced sites in one week (0 for 0 times, 1 for 20% after shooting, 2 for 20-40%, 5 for 80-100%) 1 to 5 points distribution S4 Daytime induction residence time Length of stay at Judo site for 1 week (0 hour is 0 points, 20% after shooting 1 point, 20-40% 2 points, ..., 80-100% 5 points) 1 to 5 points distribution S5 Weekly service visits Number of visits to the service site for 1 week (0 for 0 times, 1 for 20% after shooting, 2 for 20-40%, 5 for 80-100%) 1 to 5 points distribution S6 Weekly service residence time Length of stay at the service site for 1 week (0 hours is 0 points, 20% after shooting 1 point, 20-40% 2 points, ..., 80-100% 5 points) 1 to 5 points distribution S7 Inquiries Number of inquiries in the last six months (one point for one time, ..., five points for five or more times) 1 to 5 points distribution S8 Additional score 10 points for less than 1 week (current-recent visit date), 9 for 1 to 2 weeks, 0 for more than 10 weeks 0 ~ 10 points distribution Sum 45 points

No. Detail Index How to measure scoring S1 Join event The number of event participations in the last 6 months (0 times 0 points, 1 point for less than 20% after shooting, ..., 5 points for 80-100%) 1 to 5 points distribution   S2   Campaign email Number of e-mails viewed (0 times 0, read rate less than 20% 1 after shooting, ..., 80-100% 5 points)   Total / 3 0-5 points distribution E-mail reading time (0 times 0, average reading time less than 20% after shooting 1 point, ..., 80-100% 5 points) Number of email link clicks (0 for 0, average link clicks for less than 20% after shooting, 1 for ..., 80 to 100% for 5) S3 Auto Send Email Measure same as campaign email Total / 3 0-5 points distribution S4 Integrated access banner Number of connections via banner (0 points on 0 times, 1 point on 1 times, ..., 5 points on 5 times) 0-5 points distribution S5 Click front ads Interstitial clicks (0 on 0 times, 1 on 1 time, ..., 5 on 5 times) 0-5 points distribution S6 Extra points If (current-last response date) is less than 1 week, 10 weeks, 9 points for 1 ~ 2 weeks, ..., 0 points for more than 10 weeks 0-10 points distribution Sum 35 points

No. Detail Index How to measure scoring S1 Visit the mall -Average monthly visits to product malls (0 points for 0 times, 1 point for less than 20% after shooting, ..., 5 points for 80 to 100%)-Average visit time for product malls (0 points for 0 hours, 20 points after shooting) Less than% is 1 point, ..., 80-100% is 5 points) Total / 2 0-5 points distribution S2 Product design Number of visits to product design screen (as above) Average visit time of design screen (as above) Total / 2 0-5 points distribution S3 Loan Plaza Visit Loan Plaza Screen Visits (Same as above) Design Screen Average Visit Time (Same as above) Total / 2 0-5 points distribution    S4    Campaign response -Number of emails read (0 times 0, read rate less than 20% 1 point, ..., 80-100% 5 points)-E-mail reading time (0 times 0 points, average reading time Less than 20% after shooting 1 point, ..., 80 to 100% 5 points)-Number of email link clicks (0 times 0, average link clicks less than 20% after shooting 1 point, ... , 5 points for 80-100%)    Total / 3 0-5 points distribution  S5  Purchase history The number of online insurance purchases (0 for 0 times, 1 for 1 time, 2 for more than 2 times)-Number of online loans (0 for 0 times, 1 for 1 time, 2 for more than 2 times) -If either of the above two times, 0 points or 0 points  Total / 3 0-5 points distribution S6 Consultation History Number of consultation questions (0 for 0, ..., 5 for 5 or more) 1 to 5 points distribution Sum 35 points

(2) a means for extracting operational system data for the data manager to calculate the customer index based on the customer index item and the weight (200)

In the present invention, since the main target is on-line customers, the operational data is regarded as weblog data. Weblog data includes the date of last visit, the number of visits, the average time of access, the average page view, the number of inquiry e-mails, the banner advertisement, the number of access, the average banner advertisement, the access time and the number of event participation. The above data may vary depending on the type of work. That is, the items that can be extracted may vary according to the work user.

(3) Customer detail index data extraction means 300-The customer detailed index data is calculated using the extracted operating system data and items and weights selected by the business user.

Customer detailed index data is a general index that is set up to make it easy to evaluate individual customers on all kinds of websites. Each index data is calculated as follows. Each customer detail index data is measured by a metric variable, and the measured values are sorted in ascending order, and the score is divided by 20% according to the ranking. Since each detailed index data does not have many actual occurrences, the detailed index data is measured and directly mapped to a score.

(4) a means for calculating a customer index based on the extracted customer index data (400)

1. Calculation method of customer index

You can measure the value of each customer index using the measurements of the subcomponents of the individual customer indexes. Once the detailed components of each index have been measured, it is possible to derive the value of each index. Derivation of each index value is carried out as follows. If the value of each subindex item is N1, ..., Nk, and each weight is W1, ..., Wk, the weight is expressed as in Equation (1). The weight of the base index registered in the system is assigned using the AHP method.

Ni = Si measured value, Wi = weight of i-th exponent, with i> 0.

The final index is measured by multiplying the value of each index to 100 using the weight measured in Equation 1. If this is expressed as an equation, it is expressed as Equation 2. Equations 3 and 4 are examples of the registration index and the purchase index measurement formula, respectively.

Zi = i-index value, Ni = Si measured value, Wi = i-th exponent weight, σ = constant (where i> 0)

In the present invention, the factor analysis is proposed as a statistical tool for measuring the customer index. Using the fact that several variables, one of the characteristics of factor analysis, describe one factor, the weights of the detailed components are determined by factor analysis of the detailed components of individual indices using the component score coefficient matrix of the factors. Set. Therefore, finally, the customer index calculation method is determined by the following equation.

Wi = weight set by factor analysis, Si = measured value of subindex i, k = number of subindex (where i> 0)

Meanwhile, the present invention is based on not only static indexes but also dynamic indexes (conversion acceleration, conversion probability, etc.) in which a customer can predict conversion behavior in a website, and predicts the possibility of conversion. This was used as an index of customer segmentation.

2. Acceleration of Conversion

Conversion acceleration is a concept that shows the increase or decrease of the conversion speed (speed at which a customer moves to the next page group at a specific time). From the customer's point of view, the conversion speed can be understood as the learning rate experienced on the website, and from the company's point of view, the conversion speed can be understood as the marketing effect that the company provides to the customer. Therefore, the increase or decrease in the speed of customer conversion on the website means the increase or decrease in the visit effect that the website provides to the visitor, and the increase or decrease in the learning rate of the customer. The conversion acceleration of the customer is defined as follows.

3. Conversion Probability

As mentioned earlier, conversion acceleration is the effect of positive or negative visits that customers receive when they visit your website. Since the likelihood of a customer transitioning to a higher level of customization is affected by the effect of the visit, the probability of the customer's conversion is determined by how much of the total visits is positive. Therefore, when a specific customer terminates navigation within the existing customization stage, the probability of switching to the next customization stage is defined as follows.

4. Example of transition acceleration

The data used in the analysis is C's two-month weblog from August to September 2004. The customer segmentation is classified as shown in Table 5 (Table 5: number of customers and composition ratio by customer stage), and FIG. 3 shows the change in conversion acceleration rate of the top 100 children based on the conversion probability.

step Number of customers Composition ratio Conversions Conversion rate visitor 74,211 73.31% 101,234 - Information searcher 24,967 24.66% 27,023 27% customer 1,920 1.90% 2,056 8% Loyalty 136 0.13% 136 7%

In Tables 5 and 3, the top 100 users with high conversion probability showed a rapid shift from the second page group to the third page group. The top customers of the information searcher stage are expected to move quickly from the top category related page group to the information retrieval related page group, so that they can easily move on to the next step because they enjoy the high positive effects provided by the information searcher level.

5. Application examples of transition acceleration

As shown in FIG. 4, when the segment is inquired, the conversion rate can be inquired, and the foundation for setting items and weights for each index was prepared. 4 is a diagram illustrating a screen showing an example of application of the conversion probability, and FIG. 5 is a diagram for trend analysis by showing an example of application of the conversion probability in a tabular format.

(5) Means of utilizing the calculated customer index (customer segmentation, marketing materials, etc.) (500)

An example of the UI screen for customer index inquiry / use is shown in FIG. 6, and an example of a UI for inquiring a customer index and allocating web hard mileage according to the corresponding customer index is shown in FIG. 7.

Problems and Future Prospects

Currently, the market for technology development targets the development of a system for elaboration and prediction of models through CSS (Credit Rating System) or data mining, and the online customer scoring system based on the dynamic customization model according to the present invention is a conventional method. Compared to the following.

Existing technology Online customer scoring purpose Credit, delinquency scoring, bounce prediction and management Customer segmentation and loyalty scoring Main field Finance, communication Manufacture, distribution, service, online Analysis data Offline Information Online and offline information Requirements Existing data interworking function Statistical function Modeling design function Estimation and prediction function Web data linkage function Metadata management Statistics function Index calculation Rule registration result search and analysis Template

 Unlike general scoring, online customer scoring requires differentiated techniques for functions such as web data linkage, metadata management, and logical organization of index calculation. In addition, online customer scoring is a situation where customers do not pay much for the product itself in terms of development practices by existing manual coding and the size of a relatively small system. Therefore, in order to commercialize the present invention, there is an economic problem that it must be an inexpensive product implementing only essential functions and functions specifically required for online customer scoring.

(1) Ripple effect of the present invention

Technical aspects-IT technicalization of consulting knowledge, increased utilization of offline data and web data

Economic / Industrial Aspects-Reduced the cost of adopting customer scoring in the enterprise, creating the customer scoring market that was not activated due to complex data mining techniques and the excessive cost of introduction, expanding the use of manufacturing and distribution industries, and replacing the market share of foreign products with domestic products. Possible import substitution effect

   (2) Application

   Online customer scoring system based on dynamic customization model is used in the following areas.

   -Categorize your online customers and find your best customers

   -Statistical analysis of the scored variables and aggregated data

   -Logical Construction and Equation of Scoring Rules

   -View and analyze customer evaluation results

   -Effect analysis on various promotions and sales activities

Claims (10)

A data manager means for extracting operational system data for calculating the customer index based on the items necessary for calculating the customer index set from the work user and its weight; Means for calculating customer detailed index data using the extracted operating system data and items and weights set by the business user; Means for calculating a customer index based on the customer detailed index data calculated by the means; Include means for leveraging the calculated customer indices, The customer detailed index data calculating means measures the customer detail index data by the metric variable, sorts the measured values in ascending order, divides the score by 20% according to the ranking, and gives a score. The customer index calculation means predicts the conversion possibility based on the dynamic index (conversion acceleration, conversion probability) that the customer can predict the conversion behavior on the website, and based on this, it is used as an index of customer segmentation. , Online customer scoring system based on dynamic customization model. According to claim 1, wherein the items required for calculating the customer index Registration index items related to user registration, Visit index items related to the target site, guide site, service site visit, stay, etc., Response index items related to reactions such as event participation, e-mail related activities, and advertisement access, On-line customer scoring system based on a dynamic customization model, characterized in that it comprises a purchase index items related to the purchase behavior, such as product visits, product purchases. The system of claim 1, wherein the system data extracted from the data manager means includes a recent visit date, a visit count, an average access time, an average page view number, an inquiry mail reception number, a banner advertisement, an access number, an average banner advertisement, an access time, an event, and the like. On-line customer scoring system based on the dynamic customization model, characterized in that the web log data including the number of participation. delete The method of claim 1, wherein the customer index calculation means 1) means for calculating a weight using the following equation;

Ni = measured value of Si, Wi = weight of the i-th exponent (where i> 0). 2) means for calculating a final index by the following equation by multiplying the value of each index by 100 using the weights;

Zi = i-index value, Ni = Si measured value, Wi = i-th exponent weight, σ = constant (where i> 0) (3) An online customer scoring system based on a dynamic customization model, including means for finally calculating a customer index by the following equation.

Wi = weight set by factor analysis, Si = measured value of subindex i, k = number of subindex (where i> 0) The method of claim 5, wherein the customer index calculation means An online customer scoring system based on a dynamic customization model, further comprising means for calculating a registration index and a purchase index according to the following formula.

delete The online customer scoring system based on the dynamic customization model of claim 1, wherein the customer conversion acceleration in the dynamic index is defined as follows.

The online customer scoring system based on the dynamic customization model of claim 1, wherein the conversion probability in the dynamic index is defined as follows.

delete

Images