6. Fraud Detection – Data Mining / Predictive Modeling –
Business Insight
Reference :
Appendix ‘A’ – Claims & Payments-67 to 74
ISP Report – TS_CL03
Discovery Workshop Session – BI Core Business and Products
Original Requirement :
Fraud solution effectively predicts fraud and maintains a full history for reference
Fraud tools include:
• data mining
• rules engine
• predictive analysis
Fraud Analytics capabilities include:
• provides flagging on suspected fraudulent claims based on TLI past historical reference claims
• provides flagging on suspected fraudulent claims based on predictive modeling
The original assumption was to enable rule based fraud detection without predictive modelling and hence no effort was estimated.
Current Understanding :
• Based on discussion in workshops, we understand that the eBao out of the box solution will not suffice the Fraud Analytics requirements.
• Based on understanding of the requirement TLI requires a fraud detection system that is Rule based
• The business rules need to be defined to identify early claims, frequent claims, etc.
Revised Solution (if applicable) :
• Option 1:
“Rule engine is available for claims module, but not for Fraud claims at this point. The common usage of rule engine is that use parameter (if not exist, need to create parameter by coding) to define the rule through it. Fraud claim is same. But as the fraud related information is not in rule engine now, new parameters need to be added. This will be customized but no effort was estimated.
• Option 2 :
HCL recommendation is to build a rule driven system as a part of the NGIS
The rules required to flag a claim as potentially fraudulent will be built with many combinations and scenarios.
The rules can be defined and configured in rules engine.
• Option 3 :
Predictive analytics based solution post implementation
Original Estimation :
Effort was not originally estimated.
Effort Estimation :
This requirement was discussed in workshop and SME session and understood that it can suffice to have an extensive rule based analysis for fraud detection purposes. It has estimated that it would take 415 man days of development effort for this requirement based on the following assumptions:
Assumption: 5 simple rules, 10 medium level complex rules, 2 High complex rules
(E.g. Medium Complex rules: Large number of claims received on policies of the same agent + same hospital + same doctor + same region + same product + regular intervals combination)