AXA’s rapid AI rollouts raise testing stakes

Michiel Willems
August 25, 2026

The growing testing, governance and resilience challenges surrounding AI in insurance will be discussed at the QA Insurance Forum London 2026 on November 25. The event will bring together insurance technology, quality engineering and testing leaders to examine how firms can deploy agentic AI safely and reliably across regulated operations. View the agenda and register here.

Global insurer AXA has embedded artificial intelligence across its core insurance operations, creating new demands for software testing, quality assurance and digital resilience across the highly regulated insurance sector.

Its strategy spans predictive, generative and agentic AI, extending beyond experimentation into underwriting, pricing, claims processing and customer service.

The transformation has taken more than four years of testing, trials and phased implementation, as AXA worked to introduce AI-powered tools into increasingly important business processes.

The expansion creates new requirements around model validation, accuracy, robustness and regulatory compliance, particularly as AI moves closer to core decision-making systems.

Agentic AI use cases

AXA reported more than 60 agentic AI use cases in testing or partial deployment, spanning underwriting, contact centres, and claims processing, according to a recent case study.

For testing teams, that scale of deployment implied ongoing evaluation, structured QA oversight, and controls around how AI-driven automation performed in live financial workflows.

Dr Andreas Schertzinger, AXA Group Chief Data, AI & Innovation Officer, described the insurer’s ambitions as a shift beyond automation toward “augmented intelligence,” where human expertise teams with generative and predictive models.

He said that AXA was “designing and deploying AI solutions across the entire value chain,” and emphasised that the aim was not to replace human judgement but to assist and enhance it, a goal that carries direct implications for how models are tested, validated and governed.

*AXA’s global headquarters in Paris, France*

A central focus for testing teams involves tools that support dynamic extraction, summarisation and presentation of guidance materials.

AXA highlighted its use of Retrieval Augmented Generation (RAG) to help underwriters sift through large reference documents.

Pilots reduced average research time from “10 minutes to under three minutes per query,” with 86 % of users rating the tool eight out of ten and expressing confidence they had considered all relevant guidance.

“We are on a bold journey to shape the future of AI in insurance.”
– CTO Matthieu Caillat

That performance metric resembled end-user validation data more than a headline claim.

The metrics, speed improvements and end-user confidence, resemble key validation data for QA teams, but they also expose critical testing requirements: ensuring that summaries are accurate, that outputs are robust under varied conditions, and that model decisions remain compliant with regulatory standards.

Schertzinger also pointed in the case study to AXA’s approach to building a common AI platform that would be “open, secure and compliant,” stressing that governance and safe integration were core to the company’s strategy.

For QA and testing professionals, such an ecosystem underscored the need for rigorous integration testing where AI components interacted with established underwriting engines, policy systems and regulatory controls.

Broader strategy

AXA’s leaders framed the rollout as part of a broader platform strategy. The insurer said it was building one common foundation to “power agentic AI across AXA,” designed around “open market standards” and intended to remain “secure and compliant.”

“AI will shape the way we engage, interact, learn, work and live and it strongly impacts the way we deliver on our mission as AXA,” explained Matthieu Caillat, AXA Group’s chief technology & AI officer, in the recent Computing case study.

For software testing teams in banking and insurance, the shift has been one to watch closely as it went far beyond experimentation. In fact, it took AXA’s teams more than four years to test, trial and roll out the new AI-powered features and capabilities.

Predictive models and agentic AI workflows introduced new demands for validation, accuracy testing, robustness, and compliance assurance, particularly as these tools moved closer to core decision-making systems.

To celebrate the launch, Caillat brought AXA’s Tech, Data and AI teams together with some of the firm’s business teams. During the gathering in Paris, in March of this year, Caillat declared that “we are on a bold journey to shape the future of AI in insurance.”

He added: “Now, the focus is about accelerating end-to-end transformation, impactful execution, scaling across the group, and delivering tangible value.”

For software testing teams, that direction brought familiar enterprise demands. Integrations with underwriting and claims environments still required functional and integration testing, particularly where AI tools interacted with legacy policy systems and regulated decision chains.

Document summarisation and workflow automation also required quality controls to avoid misclassification, hallucinated outputs, or missed regulatory guidance, risks that became more acute when AI systems replaced manual review steps.

AXA’s approach underscored that AI adoption in insurance was not simply an innovation story. It became a testing and resilience challenge, where QA teams played a central role in validating fairness, accuracy, and compliance, while ensuring AI systems operated reliably under real supervisory scrutiny.