Practical Considerations for Applying AI to Pricing in the P&C Insurance
Authored by Sunil Yoon, Principal Actuarial Consultant, RNA Analytics
1. Introduction
In recent years, the insurance industry has increasingly focused on digital transformation, with significant changes underway across underwriting, claims, sales, and marketing operations. The actuarial function is no exception. Many actuaries are now incorporating artificial intelligence (AI) into their work to enhance efficiency and improve outcomes. In fact, the International Actuarial Association (IAA) launched an AI Task Force last year, which has been extended into a second phase this year, reflecting the global momentum.
Actuaries around the world are working to integrate AI with traditional actuarial methodologies to drive operational improvements. This article highlights practical considerations in applying AI to pricing within the property and casualty (P&C) insurance domain.
2. P&C Insurance Pricing with AI
Effective pricing in P&C insurance must accurately reflect risk factors across diverse policyholders informations. While actuaries have traditionally relied on Generalized Linear Models (GLMs), machine learning (ML) and AI-based techniques are now emerging as superior alternatives in terms of predictive performance.
These newer approaches leverage modern computing power to model data relationships without the need for manual variable selection, allowing for faster and more flexible modeling processes. For instance, Gradient Boosting Machines (GBMs) are often used in pricing due to their ability to handle nonlinear interactions and robustness to overfitting. Similarly, neural networks(NN) are applied to identify complex patterns in high-dimensional datasets. However, there are challenges when it comes to using these models directly in setting prices for customers.
3. Challenges and Considerations
Despite the potential, applying AI in actuarial pricing comes with several important considerations:
Model Interpretability and Explainability: Many ML algorithms, such as ensemble models or deep learning, are inherently opaque. This lack of transparency can conflict with regulatory expectations for interpretability in insurance pricing.
Data Bias and Fairness: High-quality, unbiased training data is crucial for effective model development. If biased data is used, the resulting model may produce unfair or inaccurate pricing, potentially harming the insurer’s long-term sustainability.
Actuaries should therefore not rely solely on AI-generated results. Instead, they must play a critical role in validating and interpreting these outcomes—tools such as SHAP (SHapley Additive exPlanations) assign an importance value to each feature for a given prediction, helping actuaries understand which variables are driving pricing decisions and ensuring responsible use of AI.
4. Applications of AI Pricing in Underwriting
While AI-generated prices may not be directly suitable for use as final policy premiums, they can support underwriting decisions. For instance, comparing AI-based pricing—developed using techniques such as Gradient Boosting Machines (GBMs) or Neural Networks (NNs)—to current pricing can help flag policies where the current premium falls significantly below the AI estimate. These cases may warrant stricter underwriting or even rejection, in order to improve overall portfolio profitability.
Using the French motor insurance dataset provided by the Casualty Actuarial Society (CAS), a simulation was conducted to model pricing and underwriting decisions. The results showed that by rejecting the lowest 10% of policies (in terms of AI-to-current premium ratio), the portfolio's loss ratio could be reduced by approximately 5%. However, this result only accounts for test data profitability. Therefore, broader stakeholder discussions would be essential before applying such strategies in practice.
5. Conclusion
This article provided an overview of AI applications in pricing within the P&C insurance sector. Due to the relatively short policy terms and simpler product structures, general insurance is well-suited for AI-based analysis and backtesting. As insurers continue exploring the integration of AI into actuarial work, and as international bodies such as the IAA expand discussions around AI's role in insurance, actuaries must actively contribute to digital transformation efforts.