RiSIC does not bring only the technology for detecting WAF in claims, but also the experience on how to apply this into existing complex environments. The methodology we developed leverages our experience in this domain and maximizes the benefits of the innovative technology.
Basic hypothesis of the methodology is that fighting WAF is a continuous process. Insurance companies themselves are not able to address the motivations because of which people do WAF. Therefore, WAF will continue to exist taking different forms: blocking WAF one way without removing the original incentive will lead to individuals looking for other ways to get the same benefits.
The goal
To understand behavior (members, clinicians, providers, processors) and identify new patterns in the data.
“Assess” is a critical step in establishing the base line: the system gets to understand the data and identify the population behavior. In the same way, each new cycle of WAF detection and prevention is triggered by identifying new patterns during the data assessment.
The outcome
Identification of small enough patterns that can be accurately identified. This output is then the input for the following “Automate” phase.
The goal
To automate the decisions based on behavioral patterns.
The patterns identified in the Assess step are translated into business rules that can fit into the existing operational environment of the insurance company.
The outcome
The system can detect the patterns with high accuracy and make decisions on behalf of the claim processors.
The goal
To look at new types of WAF.
Due to the data-driven nature, the system generates alerts for suspicious behavior patterns which have not been identified before. These need human attention, as these are not trivial cases but rather need more detailed investigation.
The outcome
Identification of new patterns which are not trivial and not easily detectable in the data.
When kicking off new implementation, the assessment is a critical step in establishing the base line. Once completed, we immediately identify the population behavior. We can distinguish outliers but we also detect and adjust positive and negative behavior. The second step is to automate patterns that the system accurately detects. In the case of claims processing, we identify immediate wins of claims whose decision can be automated.
We were able to identify a set of claim items for denial, with accuracy higher than 90%. Once identified, we managed to successfully deny 60% of them. The 3rd phase, we call it augment, looks at new patterns which are not trivial and not easily detectable in the data.