Non-Traditional Data Sources for Analytic Models Gautam Gupta Senior Manager—Risk Analytics Emirates NBD © 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation’s express consent. Traditional risk models can be improved by using Non-Traditional data like CASA, debit card transactions, corporate data, paper trails etc. especially in geographies where credit bureaus don't exist or for segments with no credit history. Data Enrichment Campaigns and Customer Level Aggregation can improve and in most cases add to the use of Non-Traditional data in credit models. Models developed on Non-Traditional data can open credit access to under served customer segments at affordable prices. 2 © 2014 Fair Isaac Corporation. Confidential. Agenda ►Deficiencies in Traditional Risk Modeling ►Non-Traditional ►Data ►The Data Sources Enrichment Campaigns Need for Customer Level Aggregation ►Regulatory Issues with Non-Traditional Data ►Use of Non-Traditional Data Sources in Emirates NBD 3 © 2014 Fair Isaac Corporation. Confidential. Deficiencies in Traditional Risk Modeling ► In this section we discuss the deficiencies associated with the traditional data/processes in risk modeling 4 © 2014 Fair Isaac Corporation. Confidential. Risk Modeling Lifecycle Lifecycle Stage Data Depth Risk Models On boarding Relationship Loyalty Application + Bureau Product Behavior + Bureau Customer Behavior + Bureau Application Scores Account Behavior scores Customer Risk scores Models become more predictive as data depth on each customer improves—best at loyalty stage 5 © 2014 Fair Isaac Corporation. Confidential. Traditional Data Sources Internal Data Credit Application Internal Debt Repayment External Data Credit Card Transactions Credit Bureau Report Bureau provides current debt performance but not future customer potential; biased by prevailing economic conditions 6 © 2014 Fair Isaac Corporation. Confidential. Credit Modeling in Banks: Challenges 1. Credit Bureau No Credit Bureau Developing Countries Expatriate Population Underserved Segments 2. Missing Customer Dimensions Career Growth Industry/Employer Risk Life Stage Burden Lifestyle Risk 3. Regulatory Overload Credit Scorecard PD Model Most of the innovation in credit modeling now comes from NBFC’s, a result of their need for automation and lower regulatory burden 7 © 2014 Fair Isaac Corporation. Confidential. Non-Traditional Data Sources ► In this section we discuss some of the non-traditional data sources that can be used for risk models along with examples of modeling variables 8 © 2014 Fair Isaac Corporation. Confidential. The 360o Customer Profile Employment/Business E.g. Income, Designation Safety Net E.g. Spousal Support Debt Repayment E.g. Payment on due debts Investment Profile E.g. Deposits vs. Shares Lifestyle E.g. Frugal vs. Extravagant Circumstances E.g. Health issues Customer Financials E.g. Liquidity, Savings, Cash flow Customer Life Stage E.g. Had a baby Bank Relationship E.g. Product holding, Age on Books Traditional data sources cover Employment and Debts in most cases, other dimensions may be covered to a varying degree. In next few slides we describe a few non-traditional data sources that cover these other dimensions. 9 © 2014 Fair Isaac Corporation. Confidential. Non-Traditional Data Sources Checking/Saving/Deposit Account Data CASA CREDITS Salary Investment returns DEBITS Inward Remittances Debt Payments Rent Outward Remittances Bills POS Analyzing Customer CASA very similar to analysing Company Financials Liquidity ► Low Balances ► Deferred Salary Debt Service ► Service Ratios ► Interest Payments Cash Flow ► Credits vs. Debits ► Rate of Debits Profitability ► Savings Rate ► Interest Income Activity ► Transaction Volumes ► Volatility CASA data is extremely useful for Income Indexation exercises— both for salaried and self employed customers 10 © 2014 Fair Isaac Corporation. Confidential. Profile ► CASA Vintage ► ATM vs. POS Non-Traditional Data Sources Transaction Data ► All transactions including CASA, Debit card, Credit card, Loan , Prepaid card etc. Dimension What to look for What can it help identify Lifestyle Summary merchant analysis Can help describe if customer is frugal or extravagant—e.g. heavy purchase at luxury stores vs. discount stores Life stage Merchant Analysis Can help identify life stage like having a baby, buying a car etc Cash flow Variation in transaction amount with specific merchant categories Can help identify cash flow issues e.g. getting gasoline for smaller amounts as compared to previous months when larger amounts would be spend at one go Payment Habits Bill payments Late bill payments may indicate bad payment habits or cash flow issues Safety Net Insurance payments E.g. Payments for earthquake insurance while residing in a earthquake prone area indicates a risk averse profile An Asian Bank was able to use transaction information from Remittance Prepaid cards to offer small ticket personal loans 11 © 2014 Fair Isaac Corporation. Confidential. Non-Traditional Data Sources Corporate Data ► Currently employment variables limited to income, designation etc., obtained from application form ► Bank can use corporate data to profile customer employer Company Profile Corporate Financials Payroll Data Some Asian banks identify fluctuations in payroll (salary decrease, deferments etc.) and use this to project layoffs 12 © 2014 Fair Isaac Corporation. Confidential. Non-Traditional Data Sources Offline Data ► This includes all data that is not data-entered into systems and is available as a paper or pdf/image document ► Most of these collected as salary or address proofs Other Bank Statements Tax Returns Utility Bills A Bank in South America already using payment and usage characteristics from postpaid mobile bills to extend credit 13 © 2014 Fair Isaac Corporation. Confidential. Non-Traditional Data Sources Other Data Sources ► There are other data sources that can be used although for most of them their use is limited by the fact that they will need to be procured from third party sources Call Center Investment Profiles Point Of Sale Loyalty Programs Insurance Companies Healthcare Services Census Data Social Network Government Services Telecommunication Companies In developing countries state lenders already have access to limited amounts of census/municipality data for welfare payments 14 © 2014 Fair Isaac Corporation. Confidential. Data Enrichment Campaigns ► In this section we discuss the need for data enrichment campaigns and an example of one such campaign 15 © 2014 Fair Isaac Corporation. Confidential. Data Enrichment Campaigns Current Framework New Scorecard Request Improved Framework Gather Available Data Develop Scorecard Identify Data Gaps Primary Data Enrichment Secondary Customer Touch Points Third party sources At each interaction ask a question/confirm a belief Search for customer information in third party databases Data enrichment campaigns are a cost effective way of creating new or enriching existing non-traditional data sources 16 © 2014 Fair Isaac Corporation. Confidential. Data Enrichment Campaign—An Example 50$ Profit Optional Marketing information block on application Launch 3 months later 10% FILL RATE In 6 months Next Month 40% FILL RATE 60% FILL RATE 70$ Profit 75$ Profit In 6 months Next Month Reward Increase Reward 20$ 40$ Enhanced Segmentation should decrease the ‘Cost of Risk’ which in theory should be greater than ‘Cost of Data Enrichment’ 17 © 2014 Fair Isaac Corporation. Confidential. 6 months later The Need for Customer Level Aggregation ► In this section we discuss the necessity of aggregating non-traditional data at customer level and examples of complex variables that can result from such aggregation 18 © 2014 Fair Isaac Corporation. Confidential. Customer Level Aggregation Benefits Special Case: Family Level Aggregation 360o Profiling ►What? ► Aggregate at Consistency of Information Customer Data Aggregation Family Unit ►Why? Complex Variables ► Safety Net ► Aggregate Financial Behavior ►How? Improved Predictive Models 19 © 2014 Fair Isaac Corporation. Confidential. ► Common Address ► Family Surname Product specific models result in inaccurate segmentation— which in turn leads to suboptimal decision management Complex Variables Situations Where Customer Risk Low from Individual Viewpoint— Together May Show Other Wise Ex. 1 Tax declaration shows customer doesn’t have health insurance Transaction data shows heavy spends in baby stores Ex. 2 Liability data shows customer broke fixed deposit Transaction data shows non-essential spends have doubled Create single repository of all customer transactions (debit, credit, branch) to profile customer (e.g. Lifestyle using wholesome merchant analysis) 20 © 2014 Fair Isaac Corporation. Confidential. Regulatory Issues with Non-Traditional Data ► In this section we discuss regulatory issues with using non-traditional data and possible solutions for such issues 21 © 2014 Fair Isaac Corporation. Confidential. Regulatory Issues Customer Consent 22 © 2014 Fair Isaac Corporation. Confidential. Fair Practices Data Integrity Regulatory Issues—Suggested Solutions Incentivize Customer Data Sharing Agreements Government Partnerships Check Proxy Discrimination* Data Validation Fraud Monitoring *Implementing Anti-Discrimination Policies in Statistical Profiling Models, By Devin G. Pope and Justin R. Sydnor, American Economic Journal: Economic Policy 3 (August 2011): 206–231 23 © 2014 Fair Isaac Corporation. Confidential. Use of Non-Traditional Data Sources in Emirates NBD 24 © 2014 Fair Isaac Corporation. Confidential. Use of Non-Traditional Data in Emirates NBD UAE Credit Market ► No Credit Bureau ► Over 30 banks ► ~2 million Credit eligible population ► ~80% Expatriates ► Central Bank mandated customer lending limits ► Majority credit loss from layoffs Credit Modeling in Emirates NBD Transactions Paper Statements CASA Credit Models 25 © 2014 Fair Isaac Corporation. Confidential. Employer And the Benefits Improve predictive power of credit scorecards Scorecard GINI’s have improved by over ~20% PREDICT PRICE 26 Automate underwriting process to reduce costs Automated decisions for majority consumer lending AUTOMATE LAUNCH Risk-reward models to generate optimal pricing Launch new products to cater to a wider market Risk based pricing has benefited product take up New products have increased eligible base by ~50% © 2014 Fair Isaac Corporation. Confidential. Thank You! Gautam Gupta pebblestreet@gmail.com © 2014 Fair Isaac Corporation. Confidential. This presentation is provided for the recipient only and cannot be reproduced or shared without Fair Isaac Corporation’s express consent. Please rate this session online! Gautam Gupta pebblestreet@gmail.com 28 © 2014 Fair Isaac Corporation. Confidential.