Using Predictive Analytics to Find Claims Leakage

Overpaying claims, known as claims leakage, is one of the major concerns for not only insurers but also for third-party administrators, self-administered organizations, and the policyholders. Every year, millions of dollars of cost are incurred by the industry due to claim leakage.

There are two general factors which pave the way for claims leakage– process issues, which are inefficient processing of claims and improper payments resulting in fines, penalties, or worker’s compensation issues, and human issues, such as poor decision making, low-quality customer service, and most importantly, fraud.

Determining Claims Leakage

It is essential for companies to identify and determine the magnitude of past and current claims leakage to determine if they are overpaying claims and develop a strategy to counter it. Determining claims leakage results in several benefits, including increased cost recovery, elimination of unnecessary or even fraudulent payments, and improved productivity of the workforce.

One way of identifying and determining claims leakage is through an audit of closed claim files, but when used alone, this may not be an optimal approach. An alternative and efficient approach is to incorporate predictive analytics in the process.

Use of Predictive Analytics for Determination of Claims Leakage

Predictive analytics, a branch of advanced analytics, makes use of data mining, modeling, statistics, machine learning, and artificial intelligence to make predictions about future events. Predictive analytics has experienced dramatic growth in use by the insurance industry. Using this technique, claims leakage can be analyzed and modeled at any stage of claim development. Claim assignment, coverage confirmation, investigation, negotiations, and recoveries are areas where predictive analytics helps.

Step 1: Data Gathering

The first and foremost step in determining claims leakage using predictive analytics is the gathering of data. Much of the data used is internal company data, i.e., claims data, or preferably, claims transactions data. Data from third party providers such as medical bill review data is also desirable.  External data, such as geodemographic data, are also often incorporated.  Questionnaires are also  a great way to augment a company’s internal structured data. The data is then cleaned and preprocessed. Feature engineering is used to transform the cleaned data so that it is more predictive.

Step 2: Data Analytics

The data related to claim leakage can be explored. Examples include:

  • Litigation rates
  • Closure rates
  • Reserve adequacy statistics
  • Salvage and Subrogation
  • SIU (special investigation unit) referrals
  • Large changes in paid amounts over time as the claim develops
  • Large changes in case reserves over time as the claim develops

These statistics are evaluated at various maturities of the claim cohort. The questionnaire date is used to identify additional claim leakage opportunities and identify claims where claim leakage occurred in the past. This data can be used to construct a dependent variable for a model that predicts the patterns associated with claim leakage.

The results of the data analytics step is typically summarized and presented in charts and tables.

Step 3: Reviewing of Analytics Results

Reviewing the results of the data analytics helps in concluding whether there is an excessive source of leakage, and based on this conclusion, whether an organization must take action.

It is essential for every company to control claims leakage to reduce their overall costs. Tools and techniques such as basic reporting analysis and predictive analytics are valuable allies in reducing this cost.

 

Authored by:

 

Grover M. Edie, MBA, FCAS, MAAA, CPCU, ARP, CERA, ARM

Consulting Actuary

Huggins Actuarial Services, Inc.

https://hugginsactuarial.com/grover-m-edie

 

Louise Francis, FCAS, CSPA, MAAA

Francis Analytics

 

Kim Piersol, FCAS, MAAA

Consulting Actuary

Huggins Actuarial Services, Inc.

https://hugginsactuarial.com/kim-e-piersol

 

 

 

Claim Fraud Detection and Predictive Analytics

Nearly one-third of insurers have reported that fraud was as high as 20 percent of claims costs! Fraud and abuse accounts for 5%-10% of claims costs EACH YEAR according to the Coalition Against Insurance Fraud. Many insurers are using predictive analytics to research their claims in order to discover fraudulent claims and excessive demands.

Predictive Analytics – Predicting Future Events

Predictive analytics is a branch of advanced analytics which helps in forecasting activities, trends, and behaviors, based on both new and historical data. It encompasses a combination of several technological techniques, including data mining, modeling, statistics, artificial intelligence, and machine learning. Predictive models are classified into two overall general types: supervised learning, where a dependent variable, i.e., an indicator of suspicious activity, is present in the data, and unsupervised learning, where a dependent variable is not available, and other approaches to finding patterns in data are used. Experts have classified predictive models into two general types; regression and classification; distinguishing between these two depends on the type of the target: numeric versus categorical.

The use of predictive analytics technique is extensive and can be applied to customer relationship management (CRM), healthcare, cross-selling, risk management, underwriting, and fraud detection.

Fraud Detection with Predictive Analytics

With technological advancements, fraud and abuse behaviors are evolving in complexity, which has made it difficult for businesses to prevent fraudulent activities from occurring or react when they do to minimize the financial damages. Predictive analytics is a technology-based approach which helps companies in detecting fraud before it takes place or after it takes place but before significant financial damages occur.

A variety of modeling approaches are used to develop fraud predictive models including:

Supervised learning
o Regression models
o Classification models

Unsupervised learning
o Clustering
o Principal components
o Network analysis

To use classification or regression models, a dependent variable related to suspicious behavior is identified, such as Special Investigation Unit referrals, or amount recovered by an SIU investigation.

Unsupervised learning does not require a dependent variable. However, a dataset rich in information that captures patterns associated with fraud or abuse is needed. This might be a detailed claims database, data from a third-party provider, such as medical bill review data, and might include external geodemographic data. The gathered data helps in generating behavioral patterns for all claims and can be summarized using data visualization techniques.

The second step of the process is to identify individual records such as policyholders or claims for follow-up. Using all the generated patterns and decision factors, a data analytics specialist scores each claim or policy. The score is then used to take further management actions such as referring claims to a specialist for further investigation, or alternatively quickly and efficiently issuing a check for the claim amount. The model then helps the company avoid further costs associated with suspicious activity.

All industries, including healthcare, insurance, and financial security are making use of this intelligent technique to predict fraudulent activities and working to create a fraud-free environment.

An Independent Actuarial Firm

Your company might be paying for claims it shouldn’t be. And if it isn’t, how do you prove it to the C-suite or the Board? You need an independent detective to provide the answer. Huggins is that independent detective.

How does Huggins do it?

Huggins has developed a technology-based approach which helps companies in detecting fraud before it takes place or after it takes place but before significant financial damages occur. Specifically, Huggins uses a variety of modeling approaches to develop fraud predictive models using your own company’s data and data format, including such predictive analytics tools mentioned above.

Other insurance companies and financial services organizations are already making use of this intelligent technique to predict fraudulent activities and working to create a fraud-free environment. Shouldn’t your company?

[1]   https://www.insurancefraud.org/statistics.htm#1

[1]  Ibid.

 

Authored by:

Grover M. Edie, MBA, FCAS, MAAA, CPCU, ARP, CERA, ARM

Consulting Actuary

Huggins Actuarial Services, Inc.

https://hugginsactuarial.com/grover-m-edie

 

Louise Francis, FCAS, CSPA, MAAA

Francis Analytics

 

Kim Piersol, FCAS, MAAA

Consulting Actuary

Huggins Actuarial Services, Inc.

https://hugginsactuarial.com/kim-e-piersol