| Home | Software Providers | Consultants | Articles | Columns | Reviews | Headlines |

-

Sidebar Article: "The Case of the $800,000 Man"

Entire contents Copyright © 1999 RISK & INSURANCE

A new category of software combines the expertise of the most experienced human being with a computerÕs ability to process thousands of variables simultaneously. The result is better decision-making, improved profitability, and more efficient claims operations.

Neural networks are no longer limited to science fiction. A new category of software that uses these networks is rapidly being deployed within the insurance, health care, and workers' compensation industries. These applications effectively address critical provider, payer, and risk management profitability issues such as risk reduction, fraud detection, business process efficiency, and customer relationship management.

Working much like the human brain, these solutions can predict the outcome of a situation earlier and more accurately than any other method currently in use, extracting value from a data stream that is otherwise too unwieldy to be useful. They offer an automated, faster and more flexible alternative to traditional approaches to analyzing data. The result is better decision-making, improved profitability, and more efficient operations that benefit risk managers, insurance carriers, third-party administrators, and employers.

These new applications are commonly referred to as predictive software solutions. They combine a traditional "rules-based" software approach with advanced forms of artificial intelligence and pattern recognition technology to deliver "predictive value."

Before the introduction of predictive software solutions, data analysis techniques were primarily rules- or red flag-based, identifying transactions that differed from specific rules. But, as we all know, human behavior is complex and constantly changing. The pattern of insurance fraud, for example, resembles more the ever-changing amoeba than the "box" of suspects identified by rules-based programs.

Predictive software solutions are based on a combination of rules-based and neural networks. A neural network is software that builds and implements mathematical equations called models. Neural networks are built from historical data; the network is then "trained" so that it recognizes which inputs have a significant effect on the prediction. The neural network automatically discovers not only additive linear relationships among data, but also iterative, nonlinear relationships.

From the historical data, the network learns how to combine independent variables to produce the desired outcome. The network begins training by making predictions with randomly adjusted weights. It then compares its predictions with the known results and adjusts each weight so that it causes less error. After reviewing thousands of examples hundreds of times, the network learns patterns and trends that enable it to make accurate predictions.

Fraud Detection

What is the scope of potential savings from predictive solutions? In the United States, losses due to workers' compensation fraud are estimated at more than $5 billion; but only about 20 percent of this fraud is detected. Losses from automobile claimant fraud are even greater;estimated at $13 billion to $18 billion in unnecessary claims costs.

These are just two areas in which predictive software can generate dramatic savings through earlier and more accurate fraud detection. In the financial services sector, the use of predictive software solutions has already decreased credit card fraud by up to 50 percent in just a few years.

The overall demand and growth for predictive technology is enormous and growing exponentially. The 1998 market for predictive technology was estimated at $1.6 billion; the market is expected to more than double by the year 2000, to $3.7 million. Within the last decade, predictive software solutions have become the industry standard in financial services; 84 percent of the credit card transaction of major U.S. carriers now use the industry's leading predictive software fraud-detection product. It is expected that the risk management and insurance industries, as well as other transaction-intensive fields, will quickly follow suit.

Better Claims Processes

An example of the value of predictive technology is its application in analyzing and interpreting claims data. In the insurance and health care industries, there is a tremendous wealth of untapped potential in the voluminous raw data associated with claims information. The application of predictive solutions to these data streams enables risk managers and insurance and health care decision-makers to obtain answers to such pressing questions as:

How much of a claim's cost can be reduced if a nurse case manager actively manages a specific case?

How likely is it that case management will reduce the claim cost by the forecasted amount?

Which claims are misrepresented by claimants or fraudulently handled by the provider?

Which claims have suspect quality of care variance, such as excessive physical medicine treatments?

Which companies are most likely to need a particular product or service to enhance their operations?

Answers to these and other critical questions can be found only through the intervention of a software system than can analyze massive amounts of complex data. Predictive software solutions combine the expertise of the most experienced human being with the ability to process thousands of variables simultaneously. These predictive solutions can update individual profiles in real-time as each new transaction presents itself and process hundreds of variables simultaneously to determine decisions and outcomes-;a feat far beyond the capability of the human mind.

Predictive software solutions can be deployed in risk management and insurance in most situations in which decisions need to be made based on a large volume of data. For example, predictive software solutions can be developed to:

Determine the critical manage/don't manage decision for case management of group health and workers' compensation claims and quantifying the value of the case management process. Nurses and adjusters spend their time more efficiently, managing only those cases for which it is cost-effective to do so.

Determine the potential for subrogation on medical, auto, and other types of claims.

Detect fraud earlier and more accurately than any other method currently available.

Provide an objective and automated means of assessing risk for insurance underwriting.

Identify what customers can use which products, based on the customer's behavior in real-time.

The Next Generation

Context vector analysis is considered the next generation in predictive software solutions. Context vector analysis is new, powerful technology that provides the ability to characterize the content of free-text information;for example, an e-mail, a set of nurses' notes, an adjuster's report;and interpret that text in mathematical terms. These mathematical representations, called "vectors," can be matched to libraries of other mathematically represented information, such as specific health care directives, customer service information and new product offerings. Context vectors can be applied to:

Automate responses to incoming electronic communications like e-mails, an important step in improving efficiencies as interactive communications and transactions over the Web increase;

Interpret case management notes to target specific communication to the nurse in real-time about his or her patients and about treatment options;

Automatically identify exceptions in claims processing based on the text of written notes and reports; and

Target customers for new products based on their information requests and purchase patterns.

In summary, predictive software solutions combine a unique set of technologies that allow insurers to deliver better service while increasing profitability through risk reduction, process efficiency and the ability to target products more effectively. Risk managers and the companies that employ them will also be beneficiaries of these improved efficiencies and will share in the increased profitability overall as risk and loss are reduced.

John Mutch is president of HNC Insurance Solutions in Costa Mesa, Calif. HNC Insurance Solutions is a division of HNC Software Inc., a developer of predictive software solutions in client-server environments.