Allstate shows how quantum computing could help build better insurance portfolios

Allstate and IBM have collaborated on research demonstrating how quantum computing can optimize insurance portfolios by solving the complex knapsack problem for highly correlated risks. The study, published in May 2026, utilizes a hybrid quantum-classical workflow to manage homeowners insurance scenarios where traditional classical simulations often struggle with uncertainty and tail events. This research is significant for the quantum computing sector as it proves that current hardware can remain competitive with, and occasionally surpass, established classical heuristics in high-stakes financial modeling.

Allstate and IBM researchers published a paper to arXiv in May 2026 exploring how quantum computing can address the knapsack problem in the insurance industry. This mathematical challenge involves selecting the most valuable combination of items—in this case, insurance policies—without exceeding a specific risk limit or weight capacity. Unlike car accidents, which are largely independent, homeowners insurance involves highly correlated risks like wildfires and hurricanes that can impact entire regions simultaneously. Allstate’s Chief Analytics Officer Eric Huls and Data Scientist Jean Utke noted that traditional classical simulations, which often run 100,000 scenarios, struggle with uncertainty and rare tail events when dealing with large-scale, complex perils across diverse geographies.

The joint team developed a hybrid workflow that utilizes the IBM Quantum Heron processor to generate batches of candidate portfolio combinations nudged toward valuable and budget-compliant solutions. This quantum step is followed by a classical refinement process that repairs answers exceeding the risk budget and identifies patterns of successful solutions to guide subsequent quantum computations. To overcome the vanishing signal problem as problem sizes grow, the researchers employed a technique of training the circuit on smaller versions of the problem and transferring that learning to larger instances. IBM quantum researcher Vaibhaw Kumar indicated that as hardware noise decreases, the computational burden on the classical side of this hybrid system is expected to diminish.

To benchmark the performance of their quantum-classical method, the team compared it against an exact solver and four established classical approximation heuristics: parallel tempering, tabu search, simulated annealing, and genetic algorithms. In tests involving up to 75 items, all methods achieved the provably best answer within a 30-minute time limit. However, as the problems scaled and constraints tightened, the quantum-classical approach remained competitive with the strongest classical heuristics and marginally surpassed them under tight constraints. While the workflow is not yet ready for the largest industry scales, the results suggest that quantum-enabled methods offer a viable path for managing the uncertain weights and flexible budgets inherent in modern insurance portfolios.