SOMPO group taps Dataiku to standardize AI model testing

SOMPO Holdings reorganized its business into two areas, "SOMPO P&C" and "SOMPO Wellbeing", starting in 2025, aiming to combine its customer base, risk management know-how, and digital, data, and AI transformation. As AI use spread, questions about fairness, transparency, and continuous operation became pressing, which is why the group formalized AI risk governance in March 2025. The body frames governance not as a brake on AI but as a base for drawing out its value safely.

The group selected Deloitte Tohmatsu for its insurance and financial industry knowledge and its ability to cover test design, execution, and improvement proposals end to end. Dataiku was chosen as the platform, in part because it can be used continuously over the long term and has a global track record. Its traceability, process logging, and version management of test conditions and results support explainability and reproducibility.

Testing was structured around three viewpoints: normal-case testing of output accuracy and evidence, robustness testing of how answers shift when wording changes, and security testing of publicly facing systems against prompt injection. Results were compiled as dashboards, letting group companies check benchmark attainment and underlying test data. Since initial tests often fall short, retesting after revisions is where accuracy gains have shown up, and the body treats those numbers as key for release decisions. The framework is already deployed across domestic group companies, and the group plans to internalize repetitive tests while keeping outside experts for advanced areas. The ultimate test is whether this discipline lets SOMPO expand AI use without raising risk.