What is AI and why do I care? Practical Applications of Artificial Intelligence for your business

Transcription

1 What is AI and why do I care? Practical Applications of Artificial Intelligence for your business

2 What is AI and why do I care? Practical Applications of Artificial Intelligence for your business ICBA LIVE 2018 Jonathan Glowacki and Madeline Johnson, CMB MARCH 13, 2018

3 Discussion Introductions What is AI? How can I feasibly use AI at my company? Technology requirements for leveraging data and performing AI Integrating data and AI The rise of chat bots and natural language processing Industry Examples of AI Case Study Q & A 3

4 Madeline Johnson, CMB, AMP Executive Financial Consultant, Milliman Madeline is an Executive Financial Consultant with the Milwaukee office of Milliman. She is dedicated to analyzing the financial risks associated with issuing and servicing mortgages, mortgage guaranty insurance, and credit enhancement structures. She specializes in secondary and capital markets, strategic planning, and business development in the mortgage finance industry. Her extensive experience includes executive consultant and senior management positions at Newbold Advisors, United Guaranty, Triad Guaranty, Mortgage Dynamics, Inc. and First Advantage Mortgage Corporation. She has been active in the MBA Metro DC serving as president, chairman, and board member. She received her Certified Mortgage Banker (CMB) designation in 1998 and the MBA CMB Society Fellow Award in She has been MBA/MW CMB committee chair since 2004.

5 Jonathan B. Glowacki, FSA, CERA, MAAA Principal and Consulting Actuary, Milliman Jonathan holds a Bachelor of Science degree in Mathematics and is a Fellow of the Society of Actuaries, a Chartered Enterprise Risk Analyst through the Society of Actuaries, and a Member of the American Academy of Actuaries. He has provided consulting services, including predictive analysis and econometric modeling, for mortgage servicers and investors, financial guaranty insurers, mortgage insurers, and government agencies. He has extensive experience in analyzing mortgage risk and mortgage-backed securities including evaluating loan repurchase risk, designing quality control processes, and estimating loan loss reserves. Jonathan has published articles and presented for organizations such as the MBA, PRMIA, and the Society of Actuaries. Jonathan has been involved in mortgage reform discussions with the Department of Treasury, FHA, FHFA, USMI, and others.

6 AI Experience Design and manage databases for developing AI tools Evaluate structured deals for clients Data Optimize Disposition Strategies Financial Models AI and Machine Learning Process Optimization Prepay, Default, Severity, Repurchase, and CECL Models Text Mining Natural Language Processing for Insurance Claims Analytics and Expenses Process optimization, borrower behavior and trends, market share analytics, and others 6

7 What should I know about artificial intelligence---- that my competitors are already talking about, and using 7

8 Artificial Intelligence Information driving decisioning The capability of a machine to imitate intelligent human behavior; A branch of computer science dealing with the simulation of intelligent behavior in computers; and Artificial intelligence creates business intelligence. Source: Websters Dictionary 8

9 Why now? Why the importance? Advances in Computing Massive amounts of data & IOT Better Algorithms and Deep Learning AI Algorithms/techniques Deep learning, Reinforcement Learning, etc Compute power CPUs, GPUs, TUs; Hyperscale compute / cloud, etc. 9

10 The primary challenge is and will always be the data. Data is the lifeblood of AI. An AI system needs to learn from data in order to be able to fulfill its function. Unfortunately, organizations struggle to integrate data from multiple sources to create a single source of truth on their customers. AI will not solve these data issues - it will only make them more pronounced. Forrester 10

11 Examples of AI in FinTech

12 Uses of AI in Mortgage Lending and other industries Chat Bots: Alexa Get me a Mortgage! Industries are using language processing to help with customer service: InsurTech; Automotive Insurance rating; Legal/ data mining analysis; and Predictive analytics. Applicability to banking industry: Quicken / Rocket Mortgage and Alexa have teamed to offer an online interactive mortgage process; Digital application processing and loan closings; and Servicing retention in portfolio. 12

13 What are FinTech Companies Doing? Most FinTech start-ups focus on non-risk bearing aspects of banking and insurance: Front End Policy Services: Lower storage costs, increase capacity and computing power; Back End Claims Services and Servicing Processes: Natural language processing (NLP); Customer Experience: Chatbots- using FAQs; Business Intelligence: New sources of relevant data; Unique tools to analyze data; and Using data to mitigate risks. 13

14 Examples of FinTech Entities Tyche Uses analytics to help clients estimate value of legal claims Insurify Virtual insurance agent consumers text a photo of their license plate to get insurance quotes Trov Sells item-by-item insurance for possessions for duration of insureds choosing, via an app. Launches in US in 2017 with Munich Re as U/W. Mnubo Provides analytics that generate insights from sensor-based data and external data sources Snapsheet Developing virtual automobile claims technology and services 14

15 Examples of FinTech Entities Carpe Data Next generation data solutions provides data prefill, data validation and commercial risk scores based on social media data. CoverHound Aggregator which allows customers to compare prices from many different carriers Hippo Insurance Provides on-line Homeowners insurance quotes in 60 seconds based on 3 simple questions. Partnering with TOPA and Swiss Re. Haven Life MassMutual direct sales to customers Google Compare (RIP) 15

16 How can I feasibly use AI at my company?

17 Start by Collecting Data Store data using object storage (i.e. data lake) Customer Performance External Data Lake Chat bots, machine learning, and other types of AI are becoming increasingly open and available as through a variety of analytics packages. It is likely these will become easier to use as technology develops. Therefore, one way to prepare for easier deployment of these tools is through collecting the required data today for future use.

18 Identify low hanging fruit What can I do with this data? Chat Bots: Customer service, payment and balance inquiries, sentiment analysis, call center check lists Machine Learning: Predict consumer behavior, identify trends, marketing opportunities Predictive Models: Reserve estimates, pricing, capital models

19 Develop an implementation plan Put the pieces together Objective: Have I articulated an objective for the project? Measure of Success: Is there a clear measure of success? Data: Do I have the data that I need? Resources: Do I have the resources (e.g. staff, technology, etc.) needed for development, implementation, and maintenance?

20 Case Example

21 Case Study: Finding Opportunities in a Purchase Market Using AI for Actionable Decisioning Desired Information: Risk characteristics in a new market area for purchase loans Borrower, Credit and Property An understanding of the market using data Data Used: Fannie/Freddie delivery Data Macro-economic data Lender for comparison Publically available data 21

22 Case Study: Finding Opportunities in a Purchase Market Using AI for Actionable Decisioning Objective: Use market data to identify our typical borrower and importantly identify what type of borrowers we are not serving for a given State Measure of Success: Identify a population of low-risk borrowers we are currently not serving Data: Freddie Mac, Fannie Mae, and Ginnie Mae acquisition data from 2016 through 2017 Resources: 1 employee, data currently exists and is processed 22

23 Data Drill Down for Mortgage Loan Opportunities Freddie 2016 Acquisitions for a Given Bank Understanding risk characteristics to define your market opportunities Data All Sellers Lender Difference $Amount $152 Brillion $7 Billion $144 Billion Int. Rate FICO LTV DTI % Purchase 37% 49% 12% % Retail 53% 37% (17%) Default Rate (est.) 1.44% 1.53% 0.10% 23

24 Data Drill Down for Mortgage Loan Opportunities GSE Data for Purchase loans Originated over the last three years Distribution by FICO Score 25% Industry Client Industry Default Risk Client Default Risk 20% 15% 10% 5% 0%

25 Data Drill Down for Mortgage Loan Opportunities GSE Data for Purchase loans Originated over the last three years Distribution by Loan-to-Value Ratio 35% Industry Client Industry Default Risk Client Default Risk 30% 25% 20% 15% 10% 5% 0%

26 Data Drill Down for Mortgage Loan Opportunities GSE Data for Purchase loans Originated over the last three years Distribution by Debt-to-Income 30% Industry Client Industry Default Risk Client Default Risk 25% 20% 15% 10% 5% 0%

27 Data Drill Down for Mortgage Loan Opportunities Create Machine Learning Model Key Variables State LTV Channel Interest Rate Purpose 27

28 Data Drill Down for Mortgage Loan Opportunities Model Predictions: Who is your customer? Data Lender Predicted as Not Client Loan Predicted as Client Loan Original Loan Amount $7 Billion $13 Billion $145 Billion Int. Rate FICO LTV DTI % Purchase 49% 50% 36% % Retail 37% 50% 53% Default Rate (est.) 1.53% 1.61% 1.41% 28

29 Data Drill Down for Mortgage Loan Opportunities Model Predictions: What loans are we not doing? Data Lender Predicted as Not Client Loan with Default Risk <= 2% Predicted as Not Client Loan with Default Risk > 2% Original Loan Amount $7 Billion $10 Billion $3 Billion Int. Rate FICO LTV DTI % Purchase 49% 47% 57% % Retail 37% 50% 47% Default Rate (est.) 1.53% 0.90% 3.74% 29

30 Data Drill Down for Mortgage Loan Opportunities Model Predictions: What loans are we not doing? Default Rate Number of Loans Lender Avg. FICO Avg. Int Rate Number of Loans Industry Avg. FICO Avg. Int Rate Lender Share 0.6% 2, , % 0.7% 2, , % 0.8% 1, , % 0.9% 1, , % 1.0% 1, , % 1.1% 1, , % 1.2% 1, , % 30

31 Cool data to integrate into your analysis Government and Census data Education Number and age of children Mean travel time to work Occupation and Industry Employment status Class of worker: civilian vs. Government Income and Benefits Salary vs. retirement incomes Non Family households Year moved into home Marital status Housing stock information American Community Survey (ACS) (obtained from reflects data compiled nationally as well as regionally. American Factfinder II, Department of Commerce Housing Affordability Data System (HADS) American Housing Survey (AHS) This data can be combined with other databases to utilize predictive analytics to find new market business opportunities 31

32 Linking Data to Publically Available Information Demographics in Florida State Demographics With related children under 5 years only With related children under 18 years Family Type Family Types in Florida Age Ranges in Florida 1,600,000 1,400,000 1,200,000 1,000, , , , ,000 0 Household Income in Florida 0% 5% 10%15%20%25%30%35%40%45% Single Families Married Families All Families Age<25 age age Age>59 32

33 2010Q4 2011Q3 2012Q2 2013Q1 2013Q4 2014Q3 2015Q2 2016Q1 2016Q4 2017Q3 2018Q2 2019Q1 2019Q4 2020Q3 2021Q2 2022Q1 2022Q4 2023Q3 2024Q2 2025Q1 2025Q4 Units of Houses in Florida Units of houses in MSA Home Price Index Linking Data to Publically Available Information Drill down by MSA or County to Find primary drivers of business Property values House Price Trends 1,000, , , , , , , , , ,000 0 Value of Owner Occ Homes State vs. MSA 35,000 30,000 25,000 20,000 15,000 10,000 5, House Price Indices Selected MSAs Quarter Florida Values Cape Coral-Fort Myers, FL Metro Area Bakersfield, CA Metropolitan Statistical Area Cape Coral-Fort Myers, FL Metropolitan Statistical Area El Paso, TX Metropolitan Statistical Area Las Vegas-Henderson-Paradise, NV Metropolitan Statistical Area 33

34 Units of Homes in Florida Units of Homes in MSA Units of Housing in Florida Units of housing in MSA Time and Age Information Comparing State to MSA Years in Home Ages of Homes Year Moved In State vs. MSA Year Built State vs. MSA 3,000, ,000 2,500, ,000 2,500,000 2,000,000 1,500, ,000 80,000 60,000 2,000,000 1,500, , ,000 80,000 1,000, , or later 2010 to to to 1999 Year moved in 1980 to and earlier 40,000 20, ,000, , ,000 40,000 20,000 0 Florida Cape Coral-Fort Myers, FL Metro Area Year Built Florida Cape Coral-Fort Myers, FL Metro Area 34

35 Using Predictive Analytics Lenders can view existing data differently by: Merging data sets to evaluate different ways to look at existing data Modeling predictive variables to find patterns Applying knowledge to new market areas 35

36 36 It is impossible to overstate the importance of artificial intelligence. Machine learning drives our algorithms for demand forecasting, product search ranking, product and deals recommendations, merchandising placements, fraud detection, translations, and much more. Though less visible, much of the impact of machine learning will be of this type quietly but meaningfully improving core operations. Amazon.com Inc. CEO Jeff Bezos

37 Questions?

38 Thank you! Jonathan Glowacki Madeline Johnson, CMB