Insurance Institute of Ontario- Bucket vs. Searchlight – a Scientific Approach to Investigating Exceptional & Suspicious Claims

As part of an overall strategy response for cost containment, insurance systems often apply some form of a procedural If/Then claim triage algorithm. These types of If/Then decision models are akin to what is known in systems theory as a Forward Chaining inference engine. These models can rely on the assumption of a stochastic environment. In probability theory, a purely stochastic system is one whose state is non-deterministic (or “random”) so that the subsequent state of the system can be determined probabilistically.

In contrast, a deterministic state is one where the same inputs can yield different outcomes, and is the world where we often find ourselves when investigating exceptional (high-severity) and suspicious (potentially fraudulent) claims, the ‘Black Swans’ of the insurance world. Forward Chaining starts with the available data and moves forward using inference rules to extract more data or obtain more information until a goal is reached and can be metaphorically visualized as a Bucket. Complex investigations applying this type of rule system can lead to the ratio of false-positive or irrelevant data (noise) overwhelming useful information (signal), decision petrifaction, and seemingly costly and interminable investigations.

In response to the challenges presented by suspicious and exceptional claims, insurance systems can potentially consider the use of Backward Chaining methodology, which employs a Then/If strategy for knowledge discovery, by moving backward from a goal. This approach can also be is described as the Searchlight approach.

This webinar will be in-depth and delivered with technical terminology.

Webinar Objectives

•Learn to apply deductive reasoning and Backward Chaining techniques to the investigation of exceptional [suspicious and high-severity] claims.
• Explore real-world examples on the effectiveness of these techniques in detecting fraudulent claims.
• Understand the limitations of inductive and data-driven fraud detection tools.