Business Intelligence Network - The Vision for BI and Beyond

Measuring Data Quality in the Eyes of an Actuary

Published: February 6, 2007

Visual data measurement coupled with a baseline assessment and continuing monitoring will drive quality improvement and increase an insurance company's book of business.

There’s no such thing as a “bad risk” in insurance,
there is such a thing as a “poorly priced” risk. – Anonymous Actuary

Value Proposition
The insurance industry’s foundation is data quality manifested in the actuary department. Data has become one of the most valuable corporate assets, particularly when quoting rates and chartering new insurance products. Data supports sophisticated actuarial analyses and decision-making processes that make insurance companies more profitable and make insurance products and services more desirable for consumers. The true value of data, regardless of the encompassing system, depends upon quality. Insurance products based on flawed data are suspect and can cost the insurance provider its highly valued customers. According to INFORMATION IMPACT International, Inc., "It has been demonstrated that non-quality data can cause business losses in excess of 20 percent of revenue and can cause business failure.” Historical data cleansing is a necessity if data quality and ultimately decision quality and profitable insurance products are the goals.

Approach
The lack of data quality must be identified as a risk when deploying any insurance industry business intelligence environment. Data quality is the state of completeness, validity, consistency and timeliness that makes data appropriate for a specific business decisions such as insurance product premium determination and rating. There are two imperative steps to understanding data quality in the insurance industry:

1. Baseline Assessment – Conduct a baseline assessment of data quality using the top 20 most valuable policy and claim data elements. These should include data elements such as policy holder SSN, address, date of birth, effective state of policy issuance and insured property category. The assessment’s first purpose is to quantify the quality level of the data fields destined for the insurance business intelligence environment. The data quality assessment sample size should be .02% of the total policy population. The best practice is to calculate data quality on each data element and report the findings in graphical form.
   
   
2. Continued Monitoring – Data quality should be monitored on a scheduled basis. Monitoring should have two samples: new data and repeated quality measurement of historical data. The same top 20 data elements included in the baseline assessment should be included in the continued monitoring.

Measurement Best Practices

The baseline assessment and continued monitoring should include the following measurements for each data element:

• Population – Examination of data for missing values, defaulted values (known or commonly used), or non-default values with frequency statistics for blank (null), default, non-default values.
  
  
• Validity – Test of non-default data against known acceptable or known unacceptable values with frequency statistics for invalid non-default, valid non-default values only.
  
  
• Consistency – Examination of non-default data for appearance as expected by comparing the field to other related fields with frequency statistics for inconsistent non-default, consistent non-default values only.
  
  
• Completeness – Examination of full record for the number of data fields that are populated with non-default, valid values with frequency statistics for complete (valid and populated), incomplete records.
  
  
• Duplication – Rate of record redundancy using business keys such as customer ID, SSN or policy ID.

Caution is needed in discussing validity. The best way to understand validity is in comparison to reliability:

• Validity – A data element is true and is what it purports to be. If a data element is valid, it is also reliable.
  
  
• Reliability – A data element consistently returns the same value, which may or may not be valid. A reliable data element is not necessarily a valid data element.

Insurance business users and actuaries want a quick visual of the data quality measurement rather than a row-by-row assessment report. The best practice approach is to adopt a repeatable actual measurement against an industry standard, a compliance standard and a company desired improvement target. All of these measurements are displayed in graphical comparative form. There are two primary graphs: 1) A line graph that compares the actual to targets, and 2) A column graph that displays the population distribution over time. Here are examples of each:

Actual vs. Target

• All graphs show normalized measures, 0%-100%
  
  
• Measures include Validity, Completeness and Consistency (shown one at a time)
  
  
• Graphs contain Compliance UCL (Upper Control Limit), Experienced-Based LCL (Lower Control Limit), Business-Defined Improvement Target and Actual
  
  
• Graphs can display a single data element, composite data elements, data elements in compliance groups or business-defined data groups
  
  
• Graphs can show repeated data quality samples or repeated strata such as new customers tracked over time

Figure 1: Line Graph Comparing Actuals to Targets

Population Distribution

• All graphs show normalized measures, 0%-100%
  
  
• Measure is population distribution of valid, invalid, default and no data against target

Figure 2: Column Graph Displaying Population Distribution Over Time

Drive Quality Improvement
Utilization of these two data quality graphics coupled with a baseline assessment and continuing monitoring will drive quality improvement and increase the insurance company’s book of business. Please watch for more articles on best practices in deploying business intelligence environments in the insurance industry.

If you found this article helpful and would like to receive the latest insights each week from the experts featured on the Business Intelligence Network, please subscribe to the Business Intelligence Network Newsletter.