GPT-Assisted Predictive Pricing Model Build and Validation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing large language models are not effectively utilized for building, testing, and validating predictive pricing models in the insurance industry, requiring manual and time-consuming processes involving actuarial involvement, and lack automation for generating supporting documentation.

Innovation Solution

A network-based system using large language models, particularly GPT models, to analyze predictive pricing models, identify issues, generate code changes, simulate new models, and provide supporting documentation, thereby automating the building, simulating, and validating process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual processes are used for building, testing, and validating predictive pricing models with actuarial involvement, then model accuracy and reliability are maintained, but the process is extremely time-consuming and requires significant manual effort

Engineering Contradiction:

Improvemodel validation reliabilityVSAvoidmodel development time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent introduces an intermediary system comprising a code generator, simulation environment, and validator that acts as a mediator between actuaries and predictive pricing models. This intermediary automates the building, testing, and validation processes while maintaining actuarial oversight, thereby reducing manual effort and time without compromising model reliability

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system enables semi-automated model development where the code generator automatically creates model code based on actuarial specifications, and the simulation environment automatically tests and validates the models. This self-service capability reduces the need for manual actuarial involvement in routine tasks while preserving actuarial control over critical decisions

Inventive Principle:

Principle #25Self-service

2Adaptability or versatility

If standard large language models are used for analyzing predictive pricing models, then general language processing capabilities are available, but they cannot effectively handle insurance industry-specific features and requirements

Engineering Contradiction:

Improvelanguage processing capabilityVSAvoidindustry-specific analysis accuracy

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent applies local quality by customizing the large language model specifically for insurance industry tasks. The code generator and validator are trained on insurance-specific data and terminology, enabling them to handle industry-specific features effectively while maintaining the general language processing capabilities of the underlying model

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system changes the parameters of the large language model by fine-tuning it with insurance industry data and adjusting its behavior to meet specific regulatory and analytical requirements. This enables the model to effectively analyze predictive pricing models while maintaining accuracy for industry-specific features

Inventive Principle:

Principle #35Parameter changes

3Reliability

If comprehensive model validation and documentation processes are implemented, then regulatory compliance and model reliability are improved, but the complexity and time required for model deployment increases significantly

Engineering Contradiction:

Improveregulatory complianceVSAvoidmodel deployment complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges the model building, testing, validation, and documentation processes into an integrated automated system. The code generator simultaneously creates model code and documentation, while the simulation environment performs comprehensive validation tests, thereby maintaining regulatory compliance without increasing deployment complexity

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system performs preliminary validation and documentation generation during the model building phase itself, rather than as separate subsequent steps. The validator automatically checks compliance requirements and generates necessary documentation upfront, reducing the complexity of later deployment stages

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250265624A1Large language modeling systems and methods for building, testing, and validating a predictive model

Publication Date: 2025.08.21 STATE FARM MUTAL AUTOMOBILE INSURANCE COMPANY

AI summary

A computer system for building, simulating, and/or validating a predictive model. The computer system programmed to (i) analyze one or more predictive pricing sub-models to detect one or more issues; (ii) execute a GPT model on the one or more predictive pricing sub-models, the GPT model trained to identify differences between predicted pricing and actual pricing for certain predefined events; (iii) compare the one or more issues to one or more outputs of the GPT model; (iv) in response to the comparison, generate a new model software template including one or more code changes to the one or more predictive pricing sub-models based upon the comparison; (v) generate a simulation environment based upon a plurality of data; (vi) execute the new model software template in the simulation environment; and/or (vii) update the new model software template based upon one or more outputs of the execution.