Insurance is a business built on trust. Policy represents a promise that insurers must be able to explain, defend and ultimately fulfill.
When you add in AI becoming more embedded in underwriting, pricing, claims and customer engagement, that promise is being mediated by data and algorithms.
An IDC report, commissioned by SAS, uncovers a striking imbalance across the insurance industry. AI adoption is accelerating, but the safeguards needed to ensure trust – governance, transparency and explainability – are lagging behind. Only one in five insurers currently operates at the highest level of AI trust maturity.
As host of Brewing Curiosity: Insurance Unfiltered, I’m tasked with asking industry experts the hard truths about AI in the insurance industry.
Watch the video to hear the top 5 takeaways from Episode 1 of Brewing Curiosity: Insurance Unfiltered
AI safety isn’t a training problem – it’s a leadership problem
According to Chris Parrish, a data scientist and industry veteran: “A common assumption in the market is that AI risk stems from a lack of technical expertise. In reality, most data scientists and analytics practitioners already understand AI techniques and their inherent trade‑offs.”
Parrish knows the bigger issue is not whether teams know how to build models safely, but whether organizations know how to operationalize them responsibly.
“AI safety ultimately sits with leadership,” he explains. “Without clear expectations, governance structures and accountability, even the most capable teams will struggle to align AI initiatives with enterprise risk appetite and regulatory requirements,” Parrish says that while policies, documentation standards, impact assessments and data safeguards form the science of AI governance, they are only part of the equation.
The art lies in culture. Insurers that succeed foster an environment of experimentation balanced with challenge. AI development cannot happen in silos; it must be a collaborative, cross‑functional effort that invites scrutiny from risk, compliance, legal and business stakeholders. When diverse perspectives are built into the process, AI outcomes are stronger, safer and more innovative.
Where insurers miss the mark on AI goals
If insurers are serious about AI, why do so many initiatives stall after the pilot phase? asks Thorsten Hein, an industry veteran and global risk technology expert. Hein points to a fundamental misalignment where AI is often treated as a technology project rather than a business transformation.
“Many insurers prove that a model works, then attempt to layer it onto legacy workflows designed for manual, sequential decision making,” Hein says. “But AI changes how decisions are made – earlier in the process, faster, and with different information. Without redesigning workflows, AI insights arrive too late to influence outcomes in underwriting, claims or customer experience.”
Hein shares that barriers like fragmented data, disconnected processes and legacy technology exist throughout an organization’s ecosystem. As a result, AI initiatives operate parallel to the business rather than being embedded within it. To scale AI, insurers must make sure that insights flow seamlessly across different functions and are delivered at the precise moment decisions are made.
“Insurers often underestimate the human and regulatory dimensions of AI. Governance, explainability and accountability must be built directly into redesigned workflows – not bolted on after deployment,” Hein says. “At the same time, roles like underwriting and claims management must evolve from task execution to supervision and refinement of AI‑supported decisions.”
What does AI success look like in practice?
Despite the challenges, some insurers successfully scale AI. Lemonade, a digital insurer built on AI, is a frequently cited example. By automating large portions of the policy and claims life cycle, Lemonade can evaluate and resolve many claims in seconds, significantly reducing administrative costs and improving customer experience.
Fortunately, success is not limited to digital‑only insurers. Hein describes how established carriers like Allstate are integrating generative AI into existing workflows. In claims communications, AI drafts most customer emails while humans review and finalize them. This approach improves clarity and consistency at scale while keeping people firmly in control.
Most notably, effective AI amplifies human expertise rather than replacing it.
The trust gap – and why it matters now
When only 20% of insurers operate at the highest level of trustworthy AI maturity, the implications go far beyond technology. As Parrish notes, the push for trustworthy AI often exposes deeper organizational issues – inefficiency, weak execution and an inability to operationalize innovation. “AI can accelerate outcomes, but it is not an 'easy' button.”
Hein emphasizes the point. “Weak AI trust erodes customer confidence, slows regulatory approval and creates internal hesitation that ultimately stifles innovation. In an industry already under trust pressure, this compounds risk at exactly the wrong time.”
SAS helps insurers with trustworthy AI transformation
AI can make or break an insurer’s promise in a single instance. By embedding governance, explainability and fairness across the AI life cycle, SAS helps insurers turn trust into an operational capability. In turn, trusted AI can be an enabler of growth, enabling insurers to innovate faster while remaining prepared to meet actuarial and compliance expectations.
