Introduction to CRM and Credit Risk
1 CRM
The concept of customer
life cycle (customer life cycle) comes from the practice of customer relationship management (CRM: Customer Relationship Management), which is used to describe the stages that customers go through when they accept different products or services. Stage, post-purchase behavior stage (this stage will introduce concepts such as customer persistence, loyalty, and advocacy)
The concept of customer lifetime value (CLV: Customer Lifetime Value) refers to the total value generated by customers in the entire customer life cycle in the future. CLV can be used as an indicator to measure the level of customer relationship.
Customers will generate different values at different stages. Before the transformation period, the company invests in marketing costs, and the value generated by customers is negative. Losing customers late will cause less losses to a company. Therefore, companies should focus on selecting appropriate customers, reducing customer churn rates, adopting customer retention strategies and cross-selling strategies. Similarly, in the retention/abandonment stage, companies should focus on customers Make selective reservations to maximize returns
Customer Relationship Management (CRM: Customer Relationship Management) refers to the management methods that companies can use to improve their success in the process of facing long-term customer relationships.
Customers are an important asset of the company, and the cost of developing a new customer is generally higher than the cost of retaining an old customer. Therefore, maintaining long-term and stable relationships with high-value customers is the key for enterprises to obtain sustainable competitive advantages.
The goal of customer relationship management is to assist enterprises in sales management, use marketing tools, provide customers with innovative and high-quality personalized services, and improve old customers. To increase the operating efficiency of the enterprise, and supplement the corresponding data mining technology and database marketing technology to coordinate the relationship and interaction between the enterprise and customers
2 CRM means and purpose (10C framework)
3 Customer Information
The collection of customer information is mainly used for customer analysis. The goal of customer analysis is to find a single accurate perspective to formulate strategies, so as to optimize the acquisition and retention of customers and define high-value customers.
Descriptive information: basic attribute information of customers, including demographic information such as gender, age, geographic location and income; also includes self-descriptive information, product preferences and evaluation information. From these data, it is possible to segment about customers Useful characteristics and classifications of, such as early adopters (the adoption of a new product during the introduction and growth phases of a product has a strong impact on later adopters), cost-effectiveness seekers, or specific customer personas. This information can come from buy-sell information, registration Records, surveys, return visits, and situational interviews. This kind of information is generally easy to collect, but the quality is difficult to ensure
. Behavioral information: customer behavioral information, that is, the general patterns that customers show when using products and services; including purchase behavior, registration, Browsing and using different devices, etc. For example, the survey found that customers of some specific product categories (consumer electronics, furniture) tend to use tablet computers to buy at night, and tend to use desktop computers during the day. The characteristic of behavioral information is real-time Collection, needs to be aggregated
Interaction information: The interaction information between the customer and the website, including the click information, navigation path and browsing behavior of the website or software. The main purpose is to test the practical performance of the website or software, such as obtaining the level corresponding to the click interval by simulating the real interaction. Approaches include: A/B testing, Google Analytics, laboratory collection and other
attitude information: customer preference information, such as preferences, choices, desires, brand recognition and feelings, etc.; through questionnaires, specific attention Group surveys and usability tests, etc. Some well-known questionnaire companies are often used to quantify the impact of behavioral and interactive information on attitude information. These attitudes may affect some self-descriptive information quantified in the descriptive information
4 Credit Risk
FICO Credit Score Considerations
Classification: 35% of repayment history, 30% of credit accounts, 15% of credit years, 10% of newly opened accounts, 10% of credit types
The classification of scores is based on the importance of each classification in the general individual, and the importance of each classification may vary for specific groups (such as people who are just starting to use credit cards)
5 Credit Score Cards
Functions: 1. Decision-making category: whether to lend, whether to agree to apply for a credit card; 2. Amount category: loan limit, credit card limit
Advantages: 1. It is convenient for business personnel to operate foolishly; 2. It is convenient for supervision by regulatory authorities (to prevent gender and race discrimination); 3. Easy to monitor and adjust
6 Business understanding
Define the target variable, that is, define good and bad customers: Delinquent loan repayment can be divided into: delinquency less than 30 days (1 month), 31-60 days (2 months), 61-90 days (3 months) ); delinquency for more than 180 days is basically considered to be bad debts
Expand market share: high tolerance, you can define the "bad" delinquency period longer, such as 3 months or more to
stably develop business and expand profits: reduce the occurrence of bad debts, "bad debts" "The delinquency period is defined as shorter (such as more than 1 month),
and the credit situation of customers who are in arrears for one month and two months is considered to be quite different. If the data allows (the proportion of customers in January-February is not large), The customer data from January to February can be deleted
to obtain data: 1. The application score obtained by filling in the information before processing (before the loan); 2. The behavior score generated by the behavioral characteristics after the processing (in the loan).
Multiple models: different Products, different groups, should correspond to specific credit scoring models
7 Modeling process and statistics
The basic process of establishing a scoring model
Binning of input variables
Modeling, generally using logistic regression to build a model
Specifying business parameters to convert the logistic regression coefficients into scoring
Model testing
WOE (Weight of Evidence): The weight of evidence changes in the same direction as the default ratio The importance of different bins
IV (information Value): information value, indicating the importance of the variable.
IV<0.02, almost no help for prediction; 0.02≤IV<0.1, some help
0.1≤IV<0.3, good for prediction Great help; IV≥0.3, great help
8 WOE and IV
WOE binning principle The
number of bins is moderate, not too much or too little The
number of records in each bin is reasonable
The binning should show obvious trend characteristics
The difference between adjacent bins should not be too large
9 Generating a credit scoring model
The score needs to be controlled within a certain range (for example, 0-1000).
For a specific score, there is a certain proportional relationship between good customers and bad customers, that is, odds, odds=xPctGoodxPctBad, for example, when the score is 800, the ratio is 50:1
and increases a certain amount When the score value is doubled, the odds ratio is doubled, for example, when the score is increased by 45 points, the odds are doubled (from 50:1 to 100:1)
Score=Offset+Factor×ln(odds) Score+pdf=Offset+Factor×ln(2× odds)
pdo:points to double the odds
Initialization: For example, Score=800 corresponds to odds=50, pdo=45.
The corresponding Offset and Factor can be calculated
10 Ratings for each category
Logit§=ln(odds) on the left side of the regression equation obtained by Logistic logistic regression, and substitute it into the credit score equation on the previous page:
