CAS Big Data December 6, 2012

Transcription

1 CAS Big Data December 6, 2012

2 Big Data I. Definition II. Growth in Storage, Computing Power and Data III. Effect on the Organization -- Tasks and Roles IV. Big Data s Impact on Other Industries V. Data has always been crucial to insurance. How is Big Data different? VI. Important Elements in the Insurance Environment VII. Insurance Big Data Example Usage Based Insurance 2

3 Definition At what point does: Lots of become Big? 3

4 Definition At what point does: Lots of become Big? Typically definition is operational, i.e. working with Big Data requires the coordination of multiple computers 4

5 Definition At what point does: Lots of become Big? Typically definition is operational, i.e. working with Big Data requires the coordination of multiple computers New software (e.g. Hadoop) is needed to support distributed computing 5

6 McKinsey View on Big Data Effectiveness 6

7 Significant Improvement is Possible UBI Depends on Big Data for Insights 7

8 Trends Affecting Big Data Cost to process and store data is shrinking Process - Moore s Law Storage exabytes in disk storage 6 exabytes on PCs, notebooks and mobile devices Average company 200 terabytes Wal-Mart has 2.5 petabytes (12.5 times larger) 8

9 2010 World Data Storage Capabilities Storage in Exabytes (10 18 ) Other Disk 9

10 Database Size Name Power of Ten Example Kilobyte 3 Xmas card list is about 500 kb in Access Megabyte 6 This PowerPoint is 2.9 meg Gigabyte 9 Terabyte 12 Clients routinely pass my team 5 Gigabyte files representing a month s worth of customer data Colleague described their UBI database as 7 terabytes -- Library of Congress is 235 Petabyte 15 Sensor readouts for oil exploration companies Exabyte 18 5 of these would equal the number of seconds since the Big Bang Avogadrobyte* *term I invented 24 2 times the number of molecules in L of any gas 10

11 Trends Affecting Big Data Shrinking: Cost to process and store data Process - Moore s Law Storage Growing: Data NewAge.com (Q4 2011) 1.2b social media users over the age of 15 are on for long periods every day 800 exabytes in 2009 (60 times storage) Annual growth rate of 45% 11

12 Trends Affecting Big Data Shrinking: Cost to process and store data Process - Moore s Law Storage Growing: Data NewAge.com (Q4 2011) 1.2b social media users over the age of 15 are on for long periods every day 800 exabytes in 2009 (60 times storage) Annual growth rate of 45% 12

13 Data Growing array of sensors that connect machine-tomachine create 30% more data each year Satellite imagery Telematics Radio Frequency Identifiers (RFIDs) Event Date Recorders (EDRs) Electronic Health Records (EHRs) Not all digital text, video, speech Unclear how to reconcile 30% and 45% annual growth estimates for data. Both represent an avalanche. As the data set grows organization and flexible tools become more critical. Enterprise Data Manager. 13

14 Data Growing array of sensors that connect machine-tomachine create 30% more data each year Satellite imagery Telematics Radio Frequency Identifiers (RFIDs) Event Date Recorders (EDRs) Electronic Health Records (EHRs) Not all digital text, video, speech Unclear how to reconcile 30% and 45% annual growth estimates for data. Both represent an avalanche. As the data set grows organization and flexible tools become more critical. Enterprise Data Manager. 14

15 Data Growing array of sensors that connect machine-tomachine create 30% more data each year Satellite imagery Telematics Radio Frequency Identifiers (RFIDs) Event Date Recorders (EDRs) Electronic Health Records (EHRs) Not all digital text, video, speech Unclear how to reconcile 30% and 45% annual growth estimates for data. Both represent an avalanche As the data set grows organization and flexible tools become more critical. Enterprise Data Manager. 15

16 Data Growing array of sensors that connect machine-tomachine create 30% more data each year Satellite imagery Telematics Radio Frequency Identifiers (RFIDs) Event Date Recorders (EDRs) Electronic Health Records (EHRs) Not all digital text, video, speech Unclear how to reconcile 30% and 45% annual growth estimates for data. Both represent an avalanche As the data set grows organization and flexible tools become more critical Enterprise Data Manager. 16

17 Data Growing array of sensors that connect machine-tomachine create 30% more data each year Satellite imagery Telematics Radio Frequency Identifiers (RFIDs) Event Date Recorders (EDRs) Electronic Health Records (EHRs) Not all digital text, video, speech Unclear how to reconcile 30% and 45% annual growth estimates for data Both represent an avalanche. As the data set grows organization and flexible tools become more critical Enterprise Data Manager 17

18 Data Prioritizing what to collect is a key skill Decisions to retain or discard should be periodically reexamined. 18

19 Data Prioritizing what to collect is a key skill Decisions to retain or discard should be periodically reexamined 19

20 Machine-to-Machine Data 20

21 Mae West Movie Star Early 20 th Century Too much of a good thing is wonderful 21

22 Mae West Movie Star Early 20 th Century Too much of a good thing is wonderful. 22

23 Herbert Simon Economist at Carnegie Mellon University A wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it. 23

24 Big Data s Effect on the Organization Functions and Their Roles Information Technologists Data Management Actuaries, Statisticians and Product Developers Management 24

28 Key Tasks Identify: Data to collect Technological and analytic capabilities necessary to access/analyze data. Management processes to evaluate and use the results. 28

29 Key Tasks Identify: Data to collect Technological and analytic capabilities necessary to access/analyze data Management processes to evaluate and use the results. 29

30 Key Tasks Identify: Data to collect Technological and analytic capabilities necessary to access/analyze data Management processes to evaluate and use the results 30

31 Challenges Lead time for the collection of data. Process to identify what needs to be collected including data from 3 rd parties Data base organization. Metadata easy to access the data for an analysis. Recruit and manage analysts with the necessary skills. 31

32 Challenges Lead time for the collection of data. Process to identify what needs to be collected including data from 3 rd parties Data base organization. Metadata easy to access the data for an analysis Recruit and manage analysts with the necessary skills. 32

33 Challenges Lead time for the collection of data. Process to identify what needs to be collected including data from 3 rd parties Data base organization. Metadata easy to access the data for an analysis Recruit and manage analysts with the necessary skills 33

34 Big Data IT Requirements Large database storage capabilities Big Table Cassandra Cloud Computing Distributed processing strategies Hadoop Stream Processing 34

35 Big Data IT Requirements Large database storage capabilities Big Table Cassandra Cloud Computing Distributed processing strategies Hadoop Stream Processing 35

36 Big Data Technologies McKinsey Study June 2011 Big Table Business Intelligence Cassandra Cloud Computing Datamart Data Warehouse Distributed System Dynamo Extract, Transform and Load Google File System Hadoop HBase MapReduce Mashup Metadata Non-relational Database R Relational Database Semi-structured Data SQL Stream Processing Structured Data Unstructured Data Visualization 36

37 Big Data Analytical Techniques McKinsey Study June 2011 A/B Testing Association Rule Learning Classification Crowdsourcing Data Fusion and Data Integration Data Mining Ensemble Learning Genetic Algorithms Machine Learning Natural Language Processing Neural Networks Network Analysis Optimization Pattern Recognition Predictive Modeling Regression Sentiment Analysis Signal Processing Spatial Analysis Supervised Learning Simulation Time Series Analysis Unsupervised Learning Visualization 37

38 Management Set priorities for data and analysis Supervise main actors Approve changes supported by analyses Involvement is key Fast turnarounds and willingness to work with data. Recognize that views may change as additional data emerges. Slick is not always a positive attribute. Learning to Presentation ratio. 38

41 Management Set priorities for data and analysis Supervise main actors Approve changes supported by analyses Involvement is key Fast turnarounds and willingness to work with data Recognize that views may change as additional data emerges Slick is not always a positive attribute Learning to Presentation ratio 41

42 Actuaries Identify the data elements to collect and reevaluate these decisions at regular intervals 42

43 Actuaries Identify the data elements to collect and reevaluate these decisions at regular intervals Integrate Big Data into rate making and reserving Participate as technical experts on company crossfunctional teams for process improvement. Help decision makers quantify their decisions. For example, the asset value of renewals. 43

44 Actuaries Identify the data elements to collect and reevaluate these decisions at regular intervals Integrate Big Data into rate making and reserving Participate as technical experts on company crossfunctional teams for process improvement Help decision makers quantify their decisions. For example, the asset value of renewals. 44

45 Actuaries Identify the data elements to collect and reevaluate these decisions at regular intervals Integrate Big Data into rate making and reserving Participate as technical experts on company crossfunctional teams for process improvement Help decision makers quantify their decisions. For example, the asset value of renewals 45

46 Big Data Impact on Other Industries Airlines Oil Exploration Retailers 46

47 Airlines mature applications Southwest and Continental Goal is to maximize ticket revenue while minimizing overbooking. Lots of 3 rd party tools to choose from: Teradata Airline Decisions PROS and PROS Cargo Teradata s features: Leverages your Integrated Passenger Name Record (ipnr) data warehouse with sophisticated analytics on overbooking and passenger no-shows Highlights the impact of customer behavior on demand and show-rate trends hidden in PNR details Provides detailed descriptions of each metric and report layout, including the value and audience for the report PROS was a pioneer in the field: Aligning your RM (Revenue Management) department with other departments Forecasting fluctuations in passenger demand Managing revenue in markets without fences or price restrictions Differentiating customer value in real-time Accepting group requests without displacing high-value demand Pinpointing the areas of your business that make or break revenue and profit 47

50 Oil Exploration enormous data Operational Control Systems: Distributed sensors and High-speed communications Explored with data-mining techniques to monitor and fine-tune remote drilling operations. The aim is to use real-time data to make better decisions and predict Chevron glitches reported: Industry wide estimates of 8 percent higher production rates and 6 percent higher recovery from a "fully optimized" digital oil field. Internal IT traffic of 1.5 terabytes a day Seismic exploration develops huge datasets: Petabyte Elastic wave propagation is measured through Land based acoustic receivers Marine based hydrophones 50

51 Oil Exploration enormous data Operational Control Systems: Distributed sensors and High-speed communications Explored with data-mining techniques to monitor and fine-tune remote drilling operations. The aim is to use real-time data to make better decisions and predict glitches Chevron reported: Industry wide estimates of 8 percent higher production rates and 6 percent higher recovery from a "fully optimized" digital oil field. Internal IT traffic of 1.5 terabytes a day Seismic exploration develops huge datasets: Petabyte Elastic wave propagation is measured through Land based acoustic receivers Marine based hydrophones 51

52 Oil Exploration enormous data Operational Control Systems: Distributed sensors and High-speed communications Explored with data-mining techniques to monitor and fine-tune remote drilling operations. The aim is to use real-time data to make better decisions and predict glitches Chevron reported: Industry wide estimates of 8 percent higher production rates and 6 percent higher recovery from a "fully optimized" digital oil field. Internal IT traffic of 1.5 terabytes a day Seismic exploration develops huge datasets: Petabyte Elastic wave propagation is measured through Land based acoustic receivers Marine based hydrophones 52

53 Retailers emphasis on customers Amazon Largest online retailer in the US. Collects huge volumes of customer choice data elements Products researched Products purchased Use this data to develop their recommendation engine You might be interested in recommendations Timing during the transactions when recommendations are made. 53

54 We Have Spent Our Work Lives Knee Deep in Data What s Really New Here? New data sources and dramatically increased volumes Allow the use of innovative analytical techniques Applicable to a wider array of issues and processes Actuarial experience with data and analysis is a key resource to every insurance company Each of us knows someone like M. Jourdain in the Middleclass Aristocrat (Bourgeois Gentilhomme) who earns our scorn for his smugness at the discovery that he has spoken prose all of his life. 54

59 Special Characteristics of Insurance Legacy Book is most valuable asset 59

60 Value of Renewal Book Combined Ratio of 94 20% New Business at 120 CR 80% Renewal at 88 CR U/W Asset Value (in Written Premium) of Renewals 80% of Premium is Renewal 12 point U/W Margin 2 Year Average Additional Life 19.2% of EP Each additional month of life adds.83%, e.g. 3 years 28.8% 60

61 Value of Renewal Book Combined Ratio of 94 20% New Business at 120 CR 80% Renewal at 88 CR U/W Asset Value (in Written Premium) of Renewals 80% of Premium is Renewal 12 point U/W Margin 2 Year Average Additional Life 19.2% of EP Each additional month of life adds.83%, e.g. 3 years 28.8% 61

62 Special Characteristics of Insurance Legacy Book is most valuable asset Like all relationship businesses client retention and cross-selling is key 62

63 Need for Customer Lifetime Value Individual client relationships represent dramatically different values to the company. Value range is more than an order of magnitude from lowest to highest Decompose renewal value and assign it at the individual customer relationship level PLE or Policy Life Expectancy Include the value from all insurance relationships Critical element for the objective function on many projects 63

66 Special Characteristics of Insurance Legacy Book is most valuable asset Like all relationship businesses client retention and cross-selling is key. Regulatory realities Limit intervals between changes More disclosure Identical risks get the same rate, no AB experiments 66

67 Usage Based Insurance Big Data Example in our Industry Auto Market ($166 B Personal; $24 B Commercial) 187 million personal vehicles Around 10% of vehicles because of age or other factors don t support telematics. 90% of US vehicle population is 168 million. Potential market is 29% of 168 million or 49 million vehicles UBI offered on opt-in basis. Market Research and Progressive experience 40% interested in UBI 75% enroll in Snapshot 67

68 Usage Based Insurance Big Data Example in our Industry Auto Market ($166 B Personal; $24 B Commercial) 187 million personal vehicles Around 10% of vehicles because of age or other factors don t support telematics. 90% of US vehicle population is 168 million. Potential market is 29% of 168 million or 49 million vehicles UBI offered on opt-in basis. Market Research and Progressive experience 40% interested in UBI 75% enroll in Snapshot 68

69 UBI Improves Risk Models 69

70 UBI Progressive Claim Most Powerful Rating Variable 70

71 UBI Tripsense Progressive Tripsense 2005 Safe Driver Neutral Unsafe Discount Before Mileage Adjustment 20% 15% 10% Discount After Mileage Adjustment 0 to 2, % 13.4% 8.4% 2,501 to 5, % 10.1% 5.1% 5,001 to 7, % 6.8% 1.8% 7,501 to 10, % 3.5% None 10,001 to 12, % 0.2% None 12,501 to 15, % None None 15,001 + None None None Minnesota Progressive Northwestern Filed July 2,

72 UBI Data Records Once Per Second trip_id update_time latitude longitude altitude gps_speed obd_speed engine_speed event_code coolant_temp trip_odometer Four Times Per Second accel_lat accel_lon accel_vert Average vehicle develops 12 Meg per month 72

73 UBI Market Composition Program Segments: 1. Reusable Device with Limited Service Capabilities 2. Installed Device Provides Driver Feedback and Value Added Services 3. Specialized Programs For Unique High Service Markets Examples: 1. Progressive Snapshot 2. State Farm InDrive 3. Safeco Teensurance 73

74 Value of UBI Simplified Illustration Benefit: 20% improvement in pure premium Costs Average Discount: 12.5% Average Cost for Technology and Communication Costs: 3.5% Net Benefit: 4% 74

77 Impact of UBI on Conversion Elasticity example relevant to UBI Assume elasticity of 2, before intro of UBI program 15% Discount should result in a 30% increase in conversion Prior to the introduction of the discount quotes converted at a 20% rate The 30% increase in conversion will increase the conversion rate to 26% 77

78 Elasticity Depends on Expected Competitiveness Example: Applicant considers a quote from Company A that includes UBI Rate, before UBI, is 5% less than the applicant s renewal offer. Once the UBI data is collected it will likely result in an additional 10% saving. Key Question Is the conversion rate improvement like we would expect to observe at 5% or 15%? 78

79 Elasticity Depends on Expected Competitiveness Example: Applicant considers a quote from Company A that includes UBI Rate, before UBI, is 5% less than the applicant s renewal offer Once the UBI data is collected it will likely result in an additional 10% saving Key Question Is the conversion rate improvement like we would expect to observe at 5% or 15%? 79

80 Elasticity Depends on Expected Competitiveness Example: Applicant considers a quote from Company A that includes UBI. Rate, before UBI, is 5% less than the applicant s renewal offer. Once the UBI data is collected it will likely result in an additional 10% saving. Key Question Is the conversion rate improvement like we would expect to observe at 5% or 15%? 80

81 Elasticity Depends on Expected Competitiveness Example: Applicant considers a quote from Company A that includes UBI. Rate, before UBI, is 5% less than the applicant s renewal offer. Once the UBI data is collected it will likely result in an additional 10% saving. Key Question Is the improvement in conversion rate commensurate with a 5% or 15% premium savings? 81

82 Big Data Numerous insurance applications Actuaries Increase their impact Skills and experience to play a leadership role in insurance 82

83 Big Data Numerous insurance applications Actuaries Increase their impact Skills and experience to play a leadership role in insurance 83

84 Ed Combs Ed Combs is currently the principal at G. Edward Combs Consulting, LLC. Ed has had over 30 years experience in the Personal Lines Insurance Industry. He has consulted for many of the major Property and Casualty insurers and has been an executive at both Progressive and 21 st Century Insurance. His certifications include: CPCU - Chartered Property and Casualty Underwriter, and ARM - Associate in Risk Management. He is a former board member of the Insurance Institute for Highway Safety, the Highway Loss Data Institute and the Advocates for Highway and Auto Safety. Ed s blog can be found at: 84