Digital and Big Data Opportunities in Credit Risk Banking Congress Warsaw, October 2015
Six key trends are expected to change bank risk management Expanding breadth and depth of regulation De-biasing judgmental decision-making Technology and math as new risk muscle Changing customer expectations Growing cost discipline Emerging risk types 1 Qualitative Credit Assessment SOURCE: McKinsey; Risk Management of the Future McKinsey & Company 2
At the heart of the new risk muscle are Big Data and Advanced Analytics Big Data Very large internal and external data sets Structured, unstructured and a combination of different data types Fast and real-time data capture Advanced Analytics Use of advanced techniques (e.g., sophisticated algorithms, neural networks) to generate insights that would be impossible to gain or would not be achievable at the same speed, scale or accuracy with conventional methods Backward- and forward-looking perspectives that turn insights into real-time decisions and actions Standardization and codification of methodology in easy-touse tools/models McKinsey & Company 3
FinTechs globally recognize this and are exerting innovation pressure on banking 12,000+ FinTech innovations from Start-ups AND Incumbents... McKinsey & Company 4
Large banks have started to build a strategy around how to manage FinTech attackers ILLUSTRATIVE 1 2 3 Partnerships/ suppliers Incubate startups Invest via VC Acquisitions There is no single solution for banks how to leverage FinTech innovation Banks should develop an in-house capability to Monitor FinTech activity Cherry-pick the best solutions to be incorporated in their DNA sequence Banks should use a combination of approaches to leverage fintechs 4 Own innovation hub within the bank 5 SOURCE: McKinsey; FinTech McKinsey & Company 5
Banks are starting to use databases beyond the traditional data sources to improve underlying models Example factors Simple variables Big data variables Traditional data sources Length of customer relationship ATM usage history (location, time, amount) Structured data Non-traditional data sources Color of iphone Super-market purchase history (item, price, location, time) Unstructured data Customer sentiment data from social media and discussion boards McKinsey & Company 6
A universal bank in Central America entered into a JV with a local supermarket in order to tap the large unbanked population for growth Our approach involved building three models Using only supermarket transactions and age (as loyalty program captures date of birth): Risk model: used for pre-screening and selective pre-approval Income model: used to assign lines Need-based segmentation model: used to target customers for specific campaigns (e.g., credit card vs. personal loan for specific appliances on sale) leveraging highly granular purchase (SKU/bar code) level data (i.e., a true big data application) Process of the data aggregation to develop variables Customer database Transaction summary Transactions details ID links to bank data such as: Credit on-us performance Off-us performance Payroll data Date of birth Time of transaction Location of transaction (store ID) Payment method ID of item bought (SKU) Quantity of item Price of item bought Data set included: 1 million customers 700 million transactions 0.1 million different products Advanced variable transformations synthesize highly predictive "signals" while filtering out noise: Identification of thousands of "marker products" for high/low risk/income Assignment of weights to markers Measurement of intensity Balancing of off-setting and/or contradictory information CLIENT EXAMPLE McKinsey & Company 7
Machine learning surfaces insights within large, complex datasets, enabling more accurate risk models The actual phenomenon (real historical data) How Traditional stats see it How Machine Learning sees it v 1 Real life phenomenon come display complex, non linear patterns v 2 Example: Decision to provide a mortgage to a specific client: 25 years old, with masters degree Variable remuneration Fashion industry employee Rich family v 1 In simple regression techniques, the expert will pre-determine the range of transformations he/she sees as making business sense (linear, quadratic, etc.) v 2 Example: the client is wrongly identified as high-risk client through traditional stats No mortgage provided v 1 Our client High risk Medium risk Low risk ML algorithms, instead are able to tease out the right pattern, for the expert to sense check. Beware over-fitting and hypothesisconfirmation though! v 2 Example: the client is correctly identified as low-risk client through Machine Learning Mortgage provided These algorithms learn with every bit of additional information as they identify new hidden patterns SOURCE: McKinsey McKinsey & Company 8
Possible economic impact Machine learning is already widely used by various industries and is expected to be adopted by banks Machine learning in various industries E-mail spam recognition Weather forecasting Speech recognition Movie/video recommendation Face/language recognition Future application in banking Return on investment Churn Retail Churn SME Rating SME Collection - Retail Collection - SME Next product - Retail Next product SME EWS 1 Retail EWS 1 SME Rating Retail Stop high-risk employee behavior and insider threat Credit card fraud detection Applicability 1 Early warning system SOURCE: McKinsey McKinsey & Company 9
This big data and advanced analytics revolution is delivering the promise of digitalization: example credit process DISGUISED CLIENT EXAMPLE Self-managed product structuring incl. live-pricing/calculators View client s borrowing potential based on available data Modify and create new facilities Quickly execute renewals, term extensions, and limit increases Up to date data sourced from risk assessment engine List of product specifications that can be obtained/granted instantaneously/within 24h View overall lending Customer and RM can see overall business lending products in customer portfolio One click to see facility details Up-to-date data sourced directly from core banking system Approval and transactions status Notifications to clarify next steps Quickly identify approval status of requests including Approved, Conditionally Approved, Standard process Straight through processing and ability to fulfill self-served (e.g., instant lending button to close online) Customer view Transaction pricing Final approval Doc submission Online/mobile customer connectivity: Self-administered interactive data entry for document submission (e.g, credit support documents) with builtin plausibility checks SOURCE: McKinsey Credit Process Service Line McKinsey & Company 10
Significant impact for every component of the RORAC possible Gross income Operating costs Loan loss provisions Other 1,262 Net income EUR millions, Qx 201x 1,418 5,119 19,517 11,718 Capital 2 34,940 RORAC (annualized) Financials for ABC Bank 5.4% Higher interest income from loan business, e.g., Increase in loan volume through sales campaigns with lower turn-down rate due to better customer pre-selection & cross-selling Increase in margin/loan volume from stepwise introduction of risk-differentiated offers (e.g., packages or prices) and cross-selling of higher margin products Lower sales and operating costs, e.g., Targeted and effective origination process (e.g., risk-pre-screening, policy pre-filters) More efficient underwriting process (e.g., risk-based Ramji differentiated process across products) Significant reduction of relative risk costs, e.g., Better selection of risks, e.g., with combined risk scores, risk clustering of customer segments Improved risk monitoring/early warning across product categories Significantly improve of capital efficiency Better calibration of the models, leading to reduced RWAs Optimization of LGD and EAD models for better capital efficiency CLIENT SANITIZED DATA Typical impact 1 5-15% 5-10% 10-30% 10-15% 1 Impact not additive and to be verified on individual bank portfolio ; SOURCE: Q3 2012 report McKinsey & Company 11
QUESTIONS Further details: Raj Dash Snr Expert Risk Advanced Analytics London Office Raj_Dash@mckinsey.com Arkadiusz Gesicki Local Partner Warsaw Office Arkadiusz_Gesicki@mckinsey.com McKinsey & Company 12