Predictive Model Data Record Selection for Subrogation
Solution Overview
Problem
Insurance companies and self-insured employers face challenges in accurately identifying claims with subrogation potential, leading to missed recovery opportunities and inefficient resource allocation due to inadequate identification of suitable claims for subrogation.
Innovation Solution
A computer system and method that processes structured and unstructured data using predictive models and text mining to assess the suitability of insurance claims for subrogation, by extracting relevant words and phrases from communications and applying them to determine a subrogation score, thereby identifying claims with increased potential for subrogation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If claim handlers manually review claims using rules to identify subrogation opportunities, then resource allocation is simplified, but identification accuracy decreases leading to missed recovery opportunities
Solution Approach 1:
The patent introduces an intermediary system comprising text mining software and predictive modeling that acts as a mediator between raw claim data and human review. This intermediary automatically extracts relevant information from unstructured data, applies predictive models to identify subrogation opportunities, and presents prioritized results to claim handlers, thereby improving identification accuracy without requiring complex manual processes
Solution Approach 2:
The patent replaces the mechanical manual review process with an automated information processing system. Text mining software automatically extracts data from unstructured sources, computational algorithms apply predictive models to assess subrogation likelihood, and the system generates automated reports - substituting human cognitive mechanics with computational mechanics to enhance precision
2Reliability
If claim handlers refer excessive numbers of claims to claims recovery operation, then subrogation opportunities are not missed, but resource allocation becomes inefficient
Solution Approach 1:
The patent applies partial action by having the predictive model identify and prioritize only the most promising subrogation opportunities rather than reviewing all claims. The system processes a subset of claims with highest predicted subrogation likelihood, achieving reliable recovery outcomes without the excessive resource consumption of reviewing every claim
Solution Approach 2:
The patent changes the parameter of claim selection from manual judgment to quantitative predictive scoring. The predictive model transforms multiple data parameters into a single subrogation likelihood score, enabling the claims recovery operation to focus on claims exceeding a threshold score, thereby optimizing the balance between recovery reliability and operational productivity
3Productivity
If claim handlers use exclusion rules to filter claims, then resource allocation is improved, but subrogation opportunities are lost due to inaccurate identification
Solution Approach 1:
The patent performs preliminary action by having the predictive model pre-assess all claims before human review or recovery action. The model提前 identifies claims with subrogation potential and prioritizes them, so that subsequent processing focuses only on promising cases, improving both productivity and precision by avoiding premature exclusion of viable claims
Solution Approach 2:
The patent implements feedback loops where the predictive model continuously learns from outcomes of subrogation cases. The system receives feedback on actual subrogation recovery results and uses this to refine its predictive algorithms, progressively improving identification precision while maintaining high productivity through automated prioritization
Data Source
AI summary
A computer system for selection of data records for forwarding includes one or more data storage devices storing data records including structured and unstructured data, and an analysis hardware server configured to, in cycles, determine sets of data records for predictive model review, extract words and phrases from the unstructured data, apply the predictive model to determine current-cycle scores for the data records of the set, assign the data records to groups in accordance with the current-cycle score, determine a velocity of change in group, and select data records for forwarding to an operation system based at least in part on the determined velocity of change in group.