DATA QUALITY IN ACTION: CHALLENGE IN AN INSURANCE COMPANY

1. rkh:For example purposes only—some content has been removed • DATA QUALITY IN ACTION: • CHALLENGE IN AN INSURANCE COMPANY Miguel FeldensGoDigital Precision Marketingmiguel@godigital.com.br Dalvani LimaGoDigital Precision Marketingdalvani@godigital.com.br Luis Francisco LimaGoDigital Precision Marketingfrancisco@godigital.com.br Executive Summary/Abstract: It is obvious to say that data quality management must be a major concern in all industries. Strategies for customer relationship management (CRM), and business intelligence (BI), for example, can only be as good as the quality of data. Both internal and external data stores should have quality standards defined, measured, analyzed, and improved, to succeed.

2. Objectives of this presentation • Contribute to the research on Information Quality (IQ), providing information about a real world implementation in a challenging scenario • Analyze this scenario under the perspective of Total Data Quality Management (TDQM) • Identify contributions of TDQM to the insurance business • Move towards a replicable methodology for TDQM in insurance

3. 40000 35000 30000 25000 20000 15000 10000 5000 0 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 all male female Scenario ExampleHow bad data can corrupt good information • Histogram of “AgeXGender” attributes – abnormal curve, with distortions caused by incorrect data entry. “Lazy typing”, for example: 05/05/55, 06/06/66... This kind of data entry may corrupt customer profile analysis, for example. age

4. 16000 14000 12000 10000 8000 6000 4000 2000 0 6 1 36 51 61 66 76 11 16 21 26 31 41 46 56 71 81 86 91 96 101 age all male female Scenario Alternatives to correct- Alternative sources (assistance)- Direct contact with the customer- Inference from profileThese cases had to be addressed in individual basis Quality Evaluation DataQuality Detect Distortion Discard incorrect information Correct data AgeXGender Analysys

5. Methodology Unique Customer View (UCV)Integrates internal data sources with sales channel, and assistance data. John DoeSSN 7127067003011th, Cauduro St Suite93ZIP 90000 Sales Channels / Banks Unique Customer View • Insurance Company • Products • Risk • Otheroperational data sources John DoeSSN 7127068003045 Raphael Zippin POAZIP 9128 5200 Behavior + Performance Assistance John Doe11, Cauduro St – Room 93Porto Alegre CityPhone Number 3311 2962 John DoeSSN 7127068003045 Raphael Zippin POAPhone 51 3311 2962

6. Methodology Example of Insurance UCVA customer-centric view of insured person Customer record must be unique, and preserve relationships with all related entities from internal and external sources. Retrospective (KPI) and Predictive Models IBGE*/Census LifeInsurance Customer(Insured) Proposalsand Sales Insured Group(company) Assistance AutoInsurance OtherInsurances Accident FireInsurance Claim

7. Methodology Steps Towards UCV in InsuranceOnce IQ of internal data stores is measured and analysed, there are four steps to build the UCV 1 2 4 3 + + + InformationIntegrationMapping different schemas, buildingand propagatingkeys to link externaldata sources. Extract, transform,and load operationaldata, into the UCV. QualityImprovementexternal data. Inference, and validation. • Modelling key performance indicators (KPIs) andstrategic information for each customer, channel, product • Retrospective • Total revenue • Total accidents • Total cost of sales • Total cost to serve • Contribution margin Structure Customer ViewUnderstanding IP consumer needs,and business strategies. Modelling the necessary datastructures to supportthem. Unique Customer View

8. Results Increased Analytical PowerAbnormal behaviors and outliers in the KPIs raise “red flags”, and can reveal potential frauds • Regions – a specific state had very poor results • Distortion in policies adopted for that particular state • Insured good – specific brand of auto with very high rate of accidents • Potential fraud or anomaly in a large fleet of cabs • ... Some actual discoveries:

9. Results Increased Analytical PowerTreatment and enrichment of customer data plus Data Mining, revealing more than U$ 10 M in cross-sales opportunities just for the active customer base. Assessment of active customers using discovered rules: * Actual products and numbers where changed due to non-disclosure agreement

10. Summary • Scenario • Value chain, culture, and technology • IP suppliers and Quality Evaluation • Corrupted Information: an Example • Methodology: Unique Customer View (UCV) • Measuring and Analyzing Data Quality • Data Reengineering • Key Performance Indicators (KPIs) • Results • Fraud / Anomalies Detection • Analytical Results & Opportunities • Further work • What’s next?

11. Next StepsAssessment, measurament, analysis, cleansing, and enrichment were large steps, but it is just the beginning... FurtherWork  On-line Data Quality • Integrate Data Quality operations with Assistance Call • Company website • Operational applications  Value-added information in the touch-points with the Integrate KPIs, predictive models, and differentiated Call center • Website • APIs (application programming interfaces) available as services to banks and financial services partners

12. References [1] Andersen. Brazilian Insurance Market 2000/2001. Andersen, 2001. [2] Eyler, T. et all. Reinventing Insurance Sales. The Forrester Report. Forrester Research, 2001. [3] Wang, R. Y. A “Product Perspective on Data Quality Management”. Communications of the ACM, Vol. 41, No. 2, February 1998, p. 58-65