INTRODUCTION
Traditionally lender’s uses Credit Scoring as a major tool in its Credit Underwriting process. Traditionally lender verifies the Customer’s Credit Worthiness through the Credit Score of an individual. It is majorly based on the past records of the Individual i.e., Payment history, delinquency in payments, default in payment to the lenders, Time period lapsed since last default etc. A Credit score is the score which is generated through the credit rating companies to identify the ability of a person to repay the obligations. Score generally lies between 300 to 900. 300 being the lowest one and 900 being the highest one. Customers who do not have any credit history might have score like 0 or -1. Credit Information Companies which are present in India are CIBIL, Equifax, Experian, CRIF.
Alternative Data scoring system in lending is quite a newer concept which is introduced with the emerging Fintech Models and also introduced in easy of doing business with the help of AI and Machine learning Models.
Alternative Data Scoring system is a procedure where credit underwriting is done with the help of alternative data available with the lender’s for analysing the credit worthiness of the customers along with the Credit Score provided by the Credit Information Companies (CIC).
ALTERNATIVE DATA
Alternative Data is the data which collected from the non traditional sources and is not included in the credit databases of the National Credit Rating Companies/Credit Information Companies. Alternative Data is the data which is used to enhance consumer lending decisions. Alternative Data can be like the Utility bills, Fuel Expenses, Money spent on leisure, Own/rented house, Insurance premium paid, Location intelligence, age of Customer, Application present in the Customer’s Phone/device, Meta Data of device etc
Alternative Data is collected from the customers by the lending institution in order to a have a better understanding of the credit worthiness of the customer.
HOW THIS DATA IS COLLECTED
Generally this data is collected by the Lending Institutions/Fintech either by asking the customer’s certain information like Date of Birth, Location Intelligence, Own/Rented house etc or using the technology i.e., scrapping.
Scrapping is the technology used by Fintech where consumers provides certain sets of permission to the Fintech companies to fetch some data from the device the customer is using to apply for the loan/Credit. By using the permission, Fintech enters into the device of the consumer to fetch the transactional SMS (not personal SMS), Fetch the list of Applications installed by the customers, Meta Data of the Device etc.
Each Data Fetched by the Fintech is Segregated using into a useful form or structure so that a decision making can be done using the data fetched i.e., A score can be given using the available Alternative Data.
WHAT POSSIBLE DATA IS SCRAPPED
The possible data that is scrapped from your device can be :-
- Transaction SMS stored in your device (In order to know your delinquency, previous loan history, Monthly Utility Expense obligation, spent on leisure etc)
- Applications Installed in your device (In order to understand whether the customer has other fintech apps installed in device, Dating apps, betting apps which can be part of negative list of lender)
- Meta Data (In order to understand the information of your device like make and model of phone, operating system of the device, IMEI number etc)
HOW ALTERNATIVE DATA SCORING IS DONE
A model is created by the Fintech Companies where Data is trained to give an Alternative score for each data available. These model is trained using the past data/ experiences using a Machines Learning Tools. Such Machines Learning tools are then converted into a AI model to respond faster whenever a new customer provide the data to the Fintech Companies.
For this Data training and AI models, Data scientist play a vital role in creating a statistical model where data is trained based on data available and the result that data provides. Like the Customers who has defaulted in payment Vs the Customer who has not defaulted in payment.
For Example Data Says that people below the age of 21 usually defaults in payment as they do not have stagnant job or recurring income and people above age 21 is good in making payment. So data scientist will help to train the data in a manner that a customer below age of 21 will get the lowest score or get rejected in the Alternative Data Scoring System.
Based on the Data Training, an algorithm is made which will be used in scoring the Customers alternatively. Score can be based on the weights of the each independent variable which has an impact on the Dependent variable. (In above Example: Independent Variable is age while the Dependent Variable is Default/no Default).
HOW WEIGHT IS ASSIGNED TO THE VARIABLES
Weights are nothing but the correlation coefficient i.e., statistical measure of the strength of a linear relation between 2 variable. It is between -1 to 1. How strong the relationship is defined by these weights. If a Relationship is strong then it will be closer to 1 or -1 and if the relationship is not strong then it will be closer to the 0. -1 depicts the negative relation and 1 depicts the positive relationship. Negative relationship is also a relationship and is used in Machine learning model.
Based on the Data analysis, weights are assigned to the Structured Data so that a machine learning model is created to provide the scoring of the customers using these alternative data.
HOW THIS SCORING SYSTEM IS USED BY FINTECHS/LENDERS
Altogether using the weights of Scrapped data, Information provided by the customer, an overall credit policy is designed by the Fintech Companies/ Lender. This Overall Credit policy also includes the Traditional Credit Score provided by the Credit Information Companies.
Using all the data, a holistic policy is created and such data is passed through a business rule engine or risk engine (based on above generated algorithm) which gives an instant decision that whether to provide loan/credit to such a particular person or not.
CRITICAL ANALYSIS
- Digitization helps in financial inclusion and India is growing in this regard. So this model helps the Fintech Companies/ Lenders in more accurate decision making.
- This also leads to Easy of doing business to various lenders/ FI’s.
- These machine learning models helps in decreasing the NPA’s and also helps in increasing the passing rate of customers for lending.
- It also decreases the human interference in credit underwriting once a model is created. So it leads in decrease in cases of underwriting frauds.
- Generally FI’s / Lenders losses money due to 2 main reason i.e., Fraud and other is NPA (due to underwriting fraud/error). These models help in decreasing the underwriting error/fraud.
- These models can be majorly used in Personal Loan/ Product financing etc.
WORD OF CAUTION
Any individual providing permission for scrapping the data shall not provide permission for following: –
- Scrap the personal SMS.
- Read/Modify/Copy the Gallery.
- Read/Modify/Copy the Phone Book.
- Any other information which seems to be unusal for credit underwriting.
Author Bio
