www.RiskNoodle.com

Text Mining for Fraudulent Patterns

Using an approach of text mining of claim diaries, Magnify introduces a new product called, FraudFocus. 

FraudFocus^TM, Magnify's predictive modeling-based claims fraud product, continuously reviews claims, beginning at first notice of loss and thereafter whenever a claim is updated throughout the entire life of the claim. The "heart" of the FraudFocus^TM approach is predictive modeling, which relies on advanced statistical analysis of claims data to develop algorithms (models) that can accurately identify claims likely to be fraudulent. Predictive modeling is a proven fraud detection technology. In the credit card industry, predictive modeling has been utilized for years to reduce fraud losses by 30% or more. Opportunities exist in Worker's Comp to apply predictive models in a similar manner to identify claimant fraud, provider fraud and policy fraud simultaneously.

Couple this approach with a streaming feed of medical bill analysis, and the possibilities are endless.

"I solemnly swear to tell the truth": Voice Fraud Detectors

Imagine asking your teenager the following, "Who are going out with tonight?", or "Where will you be going?". Typically the response is "with friends" or "nowhere special", yet you instinctively know that trouble is a-brewing. You can tell by the manner in which they answer.

Intuitively, you know that hestitation means trouble. Reseachers have found that a truthful hesitation is 0.5 seconds, whereas the average hesitation for liars is 1.5 seconds. Technology today can utilize algorithms to meaure this hesitation against a baseline of always true questions.

...a decrease in rate often indicates that the subject is editing information from the account, or fabricating information within the account. Both of these mental activities (editing and fabrication) require time and to buy time, the subject slows down his response rate.

Call Center Fraud Detection

Advanced insurers are turning toward more sophisticated methods of fraud detection. It goes without saying the earlier fraud is identified, the sooner aggressive action can be taken against the claimant.

Fraud Detection – First Notice has worked with clients to develop a series of “fraud” questions to be inserted into auto call flows. Based on the answers to those questions and data (e.g., policy inception date) pulled from a client's policy administration system and from 3rd-party sources, First Notice generates a fraud-probability score. If the score exceeds a client’s threshold, it is flagged for SIU or fraud units.

Can Voice Technology Silence Fraud?

The Pursuit Of Predictive Analysis
Detecting claims fraud early in the game--at the prepayment stage--saves insurers not only a lot of pain, but a lot of money.

Fighting fraud is becoming more meaningful to most insurers, research indicates. The Washington, D.C.-based Coalition Against Insurance Fraud (CAIF), for instance, found that 86% of insurers track the percentage of claims

they refer to their SIUs and that 39% refer between one percent to three percent of claims.

"British insurers have had success with VSA," Quigley says. "It's new on the horizon. It functions like a lie detector test, registering stress within a person's voice. It suggests that there might be deception occurring. It's one more tool to refer to an SIU for follow up."

But the downside of VSA is that it's very "Big Brotherish," Quigley adds. "People might consider it an invasion of privacy. And it would not bode well if VSA implicated someone for possibly committing fraud, only later to be found innocent. These are all thing that insurers have to think about when they expand their programs."

How Voice Stress Analysis Works

Stress can impact someone in many ways. We are all familiar with neck tension, nervousness and other physical attributes of stress, but have you considered how your voice reacts?

In both principle and execution VSA is a simple technology. Researchers found that frequencies in the human voice in the 8 to 12 Hz range are sensitive to honesty. When a person is being honest the average sound in that range is generally below 10 Hz, but is usually above 10 Hz in dishonest situations.

A Research Guide to Voice Stress Analysis

Just what you've always wanted...a definitive listing of resources for voice inflection fraud detection. It is a terrific resource!