Using insurance claim data, Break Through Tech AI fellows worked with Allstate to develop machine learning models that forecast the ultimate cost of insurance claims.
Published
03/25/2025
How can we predict the final cost of an insurance claim at the moment it’s filed? That’s the question Allstate posed to our fellows during their AI Studio Challenge Project. Using insurance claim data, fellows worked to develop machine learning models that forecast the ultimate cost of claims while uncovering the factors that influence those costs.
For Allstate, claim resolution is a time-intensive process. Determining fault, processing third-party claims, and assessing injury-related medical expenses can take months—leaving uncertainty around the financial impact of each claim. If Allstate can estimate claim costs upfront, it could set more accurate reserves, improve financial planning, and gain deeper insights into what drives higher or lower claims. This Challenge Project gave fellows a unique opportunity to tackle a real-world industry problem with direct business implications.
To build predictive models, fellows analyzed a dataset containing both categorical and continuous variables, including the target field: final claim loss. They applied a range of machine learning techniques, from linear regression and decision trees to more advanced approaches like random forests and gradient boosting. Using Python libraries such as Scikit-Learn, XGBoost, TensorFlow, and PyTorch, they iterated on different models, optimizing for accuracy and interpretability. Key deliverables included an analysis of the most influential cost-driving factors, performance metrics for each model, and visualizations to support their findings.
The impact? Not only did fellows refine AI models to predict claim costs with greater precision, but they also proposed insights that could help Allstate refine its claims process. From feature importance rankings to innovative data visualizations, their work highlights the power of AI in transforming insurance analytics.
By taking on this Challenge Project, Break Through Tech’s AI fellows gained hands-on experience with predictive modeling—while helping Allstate explore new ways to enhance efficiency and decision-making.
Your company can do the same. If you have a real-world business challenge that could benefit from fresh AI-driven solutions, join us for AI Studio and collaborate with our talented students to drive innovation in your industry. Click here for more information.