Life Insurance & Big Data Analytics: Enterprise Architecture Author: Sudhir Patavardhan Vice President Engineering Feb 2013 Saxon Global Inc. 1320 Greenway Drive, Irving, TX 75038
Contents Contents...1 Introduction...1 Role of Analytics in Insurance Life Cycle...2 Marketing and Sales Force Management... 2 Underwriting... 4 In-force Management... 5 Claims Management... 6 Big Data Analytics Solution for Insurers...7 Adapters... 8 Metadata based configurations and toolsets... 8 Big Data Analytics Framework... 8 Conclusion...9 Introduction Emerging technologies under the umbrella of Big Data provide a unique opportunity for Insurance companies. Armed with large volumes of data collected over the decades, Insurers can now unleash the power of predictive analytics to fine tune productivity and profitability in multiple areas. A typical Insurer with about a million policies has tens to hundreds of Terabytes of billing and other policy administration data. This transactional data with the application data and medical records form the traditional internal data sources. Non-traditional data sources including financial, consumer information, driving records, prescription data, etc. can now be accessed and analyzed in conjunction with traditional data to build predictive models for critical business decisions. In the Analytics arms race, Insurers who are restricted to traditional OLAP systems and reporting tools providing retrospective reporting will be faced with important questions and tough choices. Early adopters of Big Data technologies are moving towards Analytics based predictive models to unleash the power of their data. The movement towards data driven business management will differentiate the smart Insurers from the ordinary. This movement is driven by the requirements in running the business. IT teams now have the ability to respond with the plethora of tools available in the Big Data space. This whitepaper analyses the opportunities that have opened up for Life Insurance Companies and also provides a solution framework to implement Big Data technologies. Feb 2013
Role of Analytics in Insurance Life Cycle Product Development Marketing and Sales Force Management Underwriting In-force Management Claims Management Product Development/Actuarial and Underwriting have been traditional candidates for investment in Analytics for Insurers. Data Analytics for Underwriting leads the charts in terms of IT investment; closely followed by Product Development * 1. Claims management is the next biggest area in which Analytics is already playing a huge role. Most of the Analytics implementations in these areas are based on data generated and available within the organization. Now, there are non-traditional sources of data available like DMV records, Rx records, Credit Scores, Social Media data etc. that can be harnessed for better decision making. These non-traditional sources are both high volume and high velocity with an added complexity of being unstructured many times. These challenges demand the context for using Big Data technologies for implementing Analytics in the areas of Product Development, Risk assessment, Inforce and claim management. Use of analytics will accrue benefits to both Insurers and customers due to improvements in risk section and efficiency. Marketing and Sales Force Management Field Underwriting Agent Effectiveness Sales Requirement Turnaround Feb 2013
Agent Effectiveness: Measuring Agent effectiveness involves analyzing sales, field underwriting, requirements turnaround data. When this is coupled with policy in-force experience, new retention models can be developed for Agents. At-risk producers can be recognized to provide targeted assistance. Predictors like Agent longterm effectiveness scoring can be used to manage attrition more effectively. Customer Relationship Mindset: External Policyholder data Internal Policyholder data Move from Conversion Mindset to Relationship Mindset Analytics can also help the sales force to move from a conversion mindset to a relationship mindset by enabling agents to respond to real-life events. Big Data technologies can be used to merge in house policyholder information with data from external sources like social media, credit scoring, etc. to deliver tailored and unique customer experiences. Big Data can bring a significant improvement in the way Insurers plan and enable their sales force. Marketing is a significant part of the Insurers budget and this new approach can increase the ROI effectively.
Underwriting Requirement Optimization Application Interview Insurers Underwriting Algorithm Underwriting score Traditional Underwriting Rx Database DMV Records Credit scoring Third Party Data The key to improving efficiency and productivity of underwriting decisions is to blend human decision making with algorithmic support. The Analytical algorithms traditionally use the application information to make decisions on further requirements like medical checkup, etc. With the emergence of commercially available prescription databases and driving records, these algorithms can make more informed suggestions. The underwriting score derived from the Insurer s algorithm can be used to make better underwriting decisions of mortality/morbidity/incidence and drastically reduce requirements in many cases.
In-force Management Risk Monitoring In-force Policies In-force Risk Assessment Risk Profile Rx Database DMV Records Credit scoring Third Party Data Insurance is about management of Risk. Underwriting process is a risk based filtering and pricing activity. Policies after being accepted can degrade or improve in risk based on real life events. In automobile insurance the policy risk gets re-evaluated frequently die to short term renewals. In life insurance management of risk is more difficult owing to the longer renewal periods. Assessing the present risk factor of a policy can help tremendously in managing attrition, reinstatements, etc. At a higher level using In-force Risk assessment techniques can help in making better Re-Insurance decisions. In-force risk monitoring and management is as important as underwriting in the insurance business. New sources of external data like social media, prescription database, credit scoring can be coupled with application information to create real time and dynamic risk profiles at various levels in the Inforce business. Alerts and trends can be used on this profile for more proactive Risk Management.
Claims Management Claim Complexity Identification Previous Claims New Claim Extract Characteristics Build Dynamic Claim Experience Social Network / Financial Predict claim complexity Identification of problematic claims at an early stage offers tremendous advantage in terms of applying the right kind of resources. Determining the complexity of a claim based on claim experience derived from analyzing characteristics of previous claims can be instrumental in recognizing fraudulent and litigation prone claims. Claims being the biggest expense for Insurers must be the primary candidate for Big Data analytics. Coupled with social network analysis, Big Data can be used to build strong fraud detection algorithms. Many organizations may already have static claim complexity prediction models. With Big Data one can build dynamic models based on the characteristics of the incoming claim. Another candidate in Claims Management for Analytics is the activity optimization analytics based on previous claim processing logs.
Scripts and Adapters for external Data Big Data Analytics Solution for Insurers External Data Sources Social Media DMV recor ds Prescripti on Credit Score Data Simulator Collation Layer Metadata Based Configurations and Toolsets Aggregation Rules Big Data Analytics Framework Big Data Indexes Policy Billing Claims People Q u e r y i n g Model Parameters Data Utilization Adapters for internal data sources Internal Data Sources Policy Maintenance Billing Claims Applicatio n
Adapters In its simplest form, any adapter pulls data from a source and forwards it to the destination application. When we are talking of Big Data, the data sources are wide and varied. The different types of adapters needed for this solution are: File Based adapters for handling file inputs TCP/UDP adapters for handling streams Script based to handle databases Building standalone adapters for different data sources is one thing but configuring these adapters for enterprise level is a different ball game altogether. Some adapters need to be configured for just forwarding; others need to be configured for parsing and then forwarding. A topology of adapters to handle local caching, load balancing and throughout management have to be designed based on all the characteristics of the data sources. Metadata based configurations and toolsets The Big Data Analytics solution implementation needs to decouple the technology and domain related aspects very carefully. We propose a metadata based configuration layer that handles all domain related information needed for the Analytics Engine to work. Some of the configurators and toolsets needed would be: Data simulators aid in developing sample data to do what if analysis and also for testing the robustness of the models Aggregation Rules are the algorithms behind the metrics that need to be extracted from the raw data Model Parameters are the co-efficients that drive the statistical analysis of the calculated metrics Big Data Analytics Framework This is the main technology framework that is cluster enabled. It has all the pieces required to Collate and Process the data from the forwarders, extract metrics at real-time and index the results, Querying engine and a Data utilization layer. The collation layer has multiple utilities like parsing, lookups to external and internal sources, transformation options, deduplication tools and data relationship management.
The Big Data index layer gets built by the aggregation engines running the aggregation rules. These indexes will have replication, archival and other data management utilities built into it. The query layer will allow user to build standard and complex knowledge objects that can be reused by different algorithms. The data utilization layer will be a set of utilities for performing statistical analysis, patter extraction and dashboarding. Conclusion Big Data Solutioning is an amalgam of Data Science: The solution needs to cater to the needs of pattern extraction through data mining with sufficient room to simulate data for various statistical modeling. Engineering: The solution needs to be scalable both horizontally and vertically to allow enhancements and additions of new models and accommodate data velocity and volumes. Care needs to be taken to design for technology to handle spikes and ROI Innovation: Every business organization has its own inherent knowledge that has been traditionally managed as human resource assets. This unique knowledge needs to be absorbed and cherished by the solution. No solution will succeed without the integration of this knowledge into the Solutioning both at the design, execution and usage phase. The proposed solution is comprehensive in its application. It can be applied to all analytical needs of the Insurer including sales force analytics, underwriting analytics, In-force analytics and claims analytics. The solution implementation is based on try first, enhance and integrate methodology. Insurers can implement small POC s without much infrastructure and licensing costs. These POC s and implementations can be done in parallel by different departments, sharing the platform and the knowledge on the go. The implementation methodology merits a separate whitepaper, but these are the basic principles: Create a grand vision for the analytics framework, but start implementation in small phases The analytics solution need not replace existing MIS applications in use, but rather enhance them Implement usable POC s with live data. Augment it simulated datasets for testing predictive models Define KPI s and measure ROI strictly If ROI is demonstrable in the POC s, create a detailed integration plan and roll out in phases through business lines or product segments.