pict

A Fraud Detection Management Solution

HUGIN TECHNOLOGY

HUGIN software is based on state-of-the-art artificial intelligence techniques for reasoning and decision-making under uncertainty, and has a wide range of applications for improving Business Intelligence in the financial service sector. This white paper illustrates the use of HUGIN EXPERT's advanced decision support software for fraud prediction in the insurance claim handling process.

Fraud is a significant problem for insurance companies today, with an estimated 10-12 % of all insurance claims involving some kind of fraudulent activity. Insurance fraud occurs in all areas of insurance and includes application fraud, pre-existing damage fraud, claim padding, falsification of estimates and documents, etc. Fraud compels insurance companies to raise their premiums with the result that honest customers must pay too much for their insurance.

Among the challenges confronting insurance companies with regard to fraud:

  • Most people make insurance claims that are higher than the value of their loss.

  • Attitudes among customers that some forms of insurance fraud are acceptable, and that falsifying insurance claims does not constitute breaking the law.

  • The belief that fraudsters are only looking for a fair return.

Advanced detection methods such as HUGIN software play an essential role in the fight against insurance fraud and changing the mindset of customers. Reducing fraud using advanced detection techniques can lead to significant savings, both for the customer and the insurance company.

One solution is to implement a warning system that can compute the probability of a fraudulent claim in real time. This system will identify suspicious claims that have a high probability of involving fraud. Here we illustrate how HUGIN software can be used to fight insurance fraud related to claim padding.Using HUGIN software every insurance company can develop and integrate a fraud prediction model that quantifies the risk that a claim is fraudulent even in cases where data is missing or incomplete. With compelling financial benefits as a result.

fraud_wp_fig1

Figure 1: HUGIN software improves efficiency in the claims handling process by identifying fraudulent claims.

HUGIN software is an advanced tool for model-based decision support. It can be used to compute the probability of events such as fraud even when customer information is incomplete. Figure 1 illustrates how HUGIN software may be integrated into an existing claims handling process. HUGIN software is flexible, allowing the advanced decision support capabilities of the tool to be integrated into IT environments in various ways. A traffic light is only one example of how the probability of fraud may be made known to a claim handler.

A central component in a model-based decision support system is the model. A fraud detection model specifies the dependence relations between a fraudulent claim and fraud indicators and risk characteristics, including customer behavior and customer characteristics.

In particular, a fraud detection model specifies which behavior and customer characteristics differentiate fraudulent claims from legitimate claims. In the example, the traffic light allows the claims handler to classify cases as fast track, regular track, or as cases requiring additional consideration or investigation. This type of system enables insurance companies to detect fraudulent claims at an early stage, giving them more time to investigate cases with a high probability of fraud.

Exploiting advanced model-based decision support in claims handling improves the ability to identify both high risk claims and claims for fast track handling with improved results for the company: identifying high-risk claims reduces costs, identifying low risk claims saves man-time hours and increases customer satisfaction.

Figure 2 illustrates a simple fraud prediction model. The model describes dependence relations between claim type, fraud, claim history, age, etc. In the case of a burglary claim made by a male with a claims history of more than one case, the probability of fraud according to this model is 85.17%. Even with limited information the model can assess the probability of fraud. In this case the received information indicates a high probability of fraud.

fraud_model_evidence

Figure 2: The probability of fraud given a male person with a claim history of more than one claim reporting a burglary is 85.17%.

FLEXIBLE ANALYTICS

In addition to computing the probability of fraud, HUGIN software supports a variety of methods for analyzing a model and the results it produces. Given the information that the customer is a male with a claims history of more than one burglary claim, the most informative observation is the customers age. If customer age is between 25 and 40 years, it is very likely the claim is fraudulent. On the other hand, if customer age is more than 60 years, the probability of fraud is reduced to 33%.

In addition to this type of value of information analysis, a sensitivity analysis reveals that the probability of fraud is most sensitive to the observation on claims history. Sensitivity analysis can also identify the parameter of the model that has the largest impact on the probability of fraud.

The intuitive graphics of the models make it easy for analysts and subject matter experts to communicate about the properties of a model. HUGIN software enables the flexible combination of multiple sources of information such as previous claim cases, expert knowledge, background information or a mixture of these information sources. Furthermore, the model computes with missing observations in real time.

The result of implementing a system based on HUGIN software is improved customer service including a more efficient claim handling process with an improved fraud detection rate producing a reduction in the insurance claim cycle time. This implies lower costs and increased customer satisfaction. In addition, a fast and efficient claim handling process will increase customer loyalty.

This type of knowledge sharing solution for claim handlers increases the precision, efficiency, and consistency when handling an insurance claim.

The example illustrates the use of HUGIN software for fraud detection. This is not the only possible use of HUGIN software. In general, HUGIN software is well suited to problems involving reasoning and decision making under uncertainty.

Using HUGIN software any insurance company can develop and integrate a fraud prediction model that quantifies fraud insurance risk even in cases where data is incomplete or missing. The risk quantification measure is the probability of a claim being fraudulent.

The flexibility of HUGIN software makes it possible to exploit the advanced model-based decision support capabilities of the tool on a step-by-step basis. It is not necessary to develop a complete solution in the first step.

Initial models can be extended in subsequent revisions in order to cover a wider range of types of claims. In addition to detecting fraud, models may be developed to increase the efficiency of claims investigators.

Benefits of the HUGIN Fraud Detection Management Solution

The experience from a large customer reveals a number of significant benefits:

  • More accurate identification of risk claims - over 90% of red-flag claims involve fraud

  • Faster processing of legitimate insurance claims with fewer errors

  • Fewer claims processing hours per case, i.e., reduced claim cycle time

  • Reduction in overall losses due to insurance fraud

  • Better utilization of expert knowledge within the insurance fraud area

Model Development

The model development process proceeds in two steps. The first step is to identify model variables and their dependence relations while the second step is to quantify the dependence relations using (conditional) probability distributions. Figure 3 illustrates the structure of a simple fraud detection model. The model consists of six variables where fraud? is the variable of interest. The purpose of the model is to compute the probability of fraud given observations on a subset of the remaining variables. The other variables specify information that may or may not be known about the customer making an insurance claim.

fraud_model_structure

Figure 3: A simple model relating fraud to indicators such as claim history, claim type, age of customer, etc.

The graphical structure of the model specifies dependence (and independence) relations between pairs of variables. Often the arrows of a model reflect cause-effect relations of the domain. The strength of a dependence relation is specified using a (conditional) probability distribution. The model represents a joint probability distribution of its variables.

Once the qualitative part of the model, i.e., the structure, has been specified, the conditional probability distributions as defined by the structure of the model should be specified. Figure 4 specifies the quantification of the specified dependence relation between a fraudulent claim and age of the customer. The probabilities may be assessed by subject matter experts, estimated from data or a combination of the two.

fraud_model_age_cpt

Figure 4: Quantification of the relation between age and fraud.

The same model supports both fraud detection and fraud investigation. The model may be used in real-time to assess the probability of fraud in a given case, to continuously monitor all claims to identify fraudulent cases, to analyze properties of fraudulent cases, etc.

To support the model development phase, HUGIN Expert A/S offers professional training in HUGIN technology and how to use advanced HUGIN software.

System Integration

One of the major advantages of HUGIN software is that it integrates trouble-free with existing IT environments. The advanced decision support capabilities of HUGIN software integrate into existing as well as new environments.

fraud_wp_fig5_small

Figure 5: Trouble-free system integration.

HUGIN Application Programming Interfaces make the system integration process simple and efficient. Figure 5 illustrates the typical setup. The model development process proceeds on standard PC platforms using the HUGIN Graphical User Interface. Each model is stored in an ASCII text file. The file is then transferred to the production system. On the production side existing IT systems interact with the model through the HUGIN API feeding information into the model and receiving results of inference using a well-defined set of flexible functions.

System maintenance and model updates are separated. Once system integration has been accomplished, model update is simple. Given the same set of input and output data, model update consists of replacing a single ASCII text file on the production side. This enables analysts and knowledge engineers to adjust models without the support of IT experts.

Key Advantages of HUGIN Software

HUGIN software has a number of key advantages over competing systems:

  • Integrates different sources of information, including historical data and knowledge of subject-matter experts

  • Computes with incomplete (missing) observations

  • Calculations are highly efficient supporting real time inference

  • Model integration and maintenance is easy

  • Intuitive, visual graphics make the model an efficient tool for communication between analysts, subject matter experts, claims handlers, decision makers, etc.

  • Flexible knowledge representation. Models can easily be extended with additional variables, and parameters of a model can be adjusted with minor effort

Products & Services

HUGIN software consists of model development and deployment tools. The HUGIN Graphical User Interface supports analysts in the model development phase, while the HUGIN Decision Engine comes with Application Programming Interfaces for major programming languages enabling HUGINs efficient integration into new and existing IT systems.

HUGIN software has been implemented on wide range of software and hardware platforms including servers, desktop & laptop PCs and PDAs running on Microsoft Windows, UNIX and Linux operating systems.

HUGIN offers professional training, consulting and technical support services. Basic training in Hugin technology and tools provides analysts with the knowledge they need to develop sophisticated models using HUGIN software. To assist in the model development and integration phase, HUGIN Expert A/S offers professional consultancy services.

Key Product Features

Figure 6 below is an example of the HUGIN Graphical User Interface. This interface supports analysts and knowledge engineers in the model development process.

fraud_developer_small

Figure 6: HUGIN Graphical User Interface illustrating the fraud model.

  • User-friendly graphical knowledge engineering tool

  • Highly efficient inference engine

  • Integrates easily into existing systems

  • Easy to extend, modify, implement and maintain

  • Flexible, scalable technology with sophisticated analytical tools

  • Flexible Application Programming Interfaces to major programming languages: C, C++, Java, .NET and VBA for Applications