IBM & Allstate Show Quantum Value for Portfolio Optimization

Allstate and IBM are tackling a core computer science challenge, the knapsack problem, to address increasingly complex risks within insurance portfolios. The collaboration seeks to improve how insurers balance customer needs with responsible risk management, a task made significantly harder by interconnected homeowner risks; if a tornado impacts a community, a home insurer may end up paying to replace your house, your neighbor’s house, and every building in the area. “On the homeowner side, it really requires us to be thinking from a portfolio perspective and not just from an individual risk perspective,” explained Eric Huls, Allstate’s Chief Analytics Officer and Chief Data Officer, underscoring a recognition of interconnected risks. Published to arXiv in May, the joint work explores how quantum computing might offer a solution to computational hurdles previously insurmountable for classical computers when dealing with vast numbers of correlated risks.

Knapsack Problem Applied to Insurance Portfolio Risk

The seemingly abstract challenge of the knapsack problem from computer science is now at the heart of a collaboration between Allstate and IBM, aimed at fundamentally reshaping how insurance portfolios are optimized and risk is managed. This isn’t merely an academic exercise; the core issue, finding the best combination of items to maximize value within constraints, directly mirrors the complexities insurers face when balancing risk across potentially correlated homeowner policies. Many formulations of the knapsack problem have no practical solution in classical computing, particularly when the number of items gets very large, a situation increasingly common for large insurance providers. Allstate is thinking from a portfolio perspective due to the increasing recognition of interconnected risks. This interconnectedness dramatically increases the computational burden of accurately assessing portfolio risk.

Traditional methods, like running 100,000 simulations of possible futures, struggle with the accuracy of estimating rare, high-cost events. “The challenge is that when you’re looking at 1% of tail events out of 100,000 simulations, you’re down to 1,000 events. And there’s a fair amount of uncertainty within those estimates, particularly when you’re looking at many different types of peril across very large geographies,” notes Jean Utke, Data Scientist and Technical Director in Allstate’s Insurance Product organization. The Allstate-IBM team is tackling this complex knapsack problem with a hybrid quantum-classical approach.

Vaibhaw Kumar, IBM quantum researcher and co-author of the paper, highlights the limitations of current classical methods: “Current classical approaches either rely on simulating multiple scenarios based on the uncertainties (accurate but computationally expensive) or by considering the worst-case range (safe but overly cautious).” Their workflow leverages IBM Quantum Heron to generate candidate solutions, prioritizing combinations that balance value and adherence to risk tolerance. A classical step then refines this list, learning from successful combinations to improve subsequent quantum computations. This iterative process, coupled with a technique of training on smaller problems before scaling up, demonstrates promising results, tying with the best results on problems up to 75 items and remaining competitive with established classical heuristics. This workflow isn’t yet ready for scales where exact solvers are too slow to be useful, but these findings suggest a pathway toward leveraging quantum computing for real-world insurance applications as hardware continues to advance.

Allstate & IBM’s Quantum-Classical Workflow

Allstate and IBM are developing a hybrid computational approach to address the escalating complexity of modern insurance portfolio optimization. The collaboration, detailed in research published to arXiv in May, centers on adapting the knapsack problem, a notoriously difficult challenge in computer science, to the specific demands of balancing risk and reward within a large insurance book of business. Unlike simpler problems where a definitive solution exists, many real-world instances of the knapsack problem become intractable for classical computers as the number of variables increases. This computational burden stems from the increasingly correlated nature of risks faced by insurers. If a tornado impacts a community, a home insurer may end up paying to replace your house, your neighbor’s house, and several buildings in the area. The quantum component, executed on an IBM Quantum Heron processor, generates potential solutions, prioritizing combinations that offer high value while remaining within acceptable risk parameters. This initial output is then refined by classical algorithms that correct any solutions exceeding the budgetary constraints and learn from successful combinations to improve subsequent iterations.

Quantum Circuit Training and Scalability

Allstate’s pursuit of quantum solutions extends beyond simply identifying potential applications; the company is actively refining the techniques needed to train quantum circuits for complex real-world problems. Researchers at Allstate and IBM detailed a method for improving quantum circuit performance through a transfer learning approach, addressing a critical hurdle in scaling these systems to tackle increasingly large and intricate datasets. The team’s work, published to arXiv in May, focuses on optimizing quantum algorithms for the knapsack problem, a computer science challenge directly applicable to building balanced insurance portfolios. A key difficulty lies in the fact that the losses insurers face aren’t static figures, but rather uncertain ranges, reflecting the unpredictable nature of disasters. This uncertainty, combined with the need to stay within a defined risk tolerance, creates a classically challenging problem.

The quantum circuit, running on IBM Quantum Heron, generates a batch of candidate answers, and a classical algorithm refines the results. This iterative process, coupled with a technique of training on smaller problems before scaling up, demonstrates promising results, tying with the accuracy of exact solvers on problems up to 75 items and remaining competitive with established classical heuristics. There’s reason to expect this workflow to get more powerful and efficient as hardware improves.

Benchmarking Quantum Results Against Classical Heuristics

Allstate’s exploration of quantum computing isn’t a speculative venture into future technology; it’s a direct response to escalating computational demands in modern insurance portfolio optimization. The core challenge lies in accurately assessing interconnected risks, a problem rooted in the knapsack problem from computer science. Unlike scenarios with independent events, like individual car accidents, homeowner insurance faces correlated risks. If a tornado impacts a community, a home insurer may end up paying to replace your house, your neighbor’s house, and every building in the area. Classical methods currently employed by Allstate involve running up to 100,000 simulations of possible futures to understand potential outcomes. However, accurately modeling the financial impact of rare, high-cost events presents a significant hurdle. To benchmark the quantum approach, the team utilized an exact solver, a computationally intensive method that guarantees the optimal solution, as a baseline.

This allowed for direct comparison against four established classical approximation methods: parallel tempering, tabu search, simulated annealing, and a genetic algorithm. Each method was allocated 30 minutes to solve the same problems. Results showed that on problems with up to 75 items, the quantum-classical method tied with the best classical heuristics, and remained competitive with established classical heuristics. This suggests a pathway toward leveraging quantum computing for real-world insurance applications as hardware continues to advance.