US insurers are racing to deploy artificial intelligence across underwriting, claims, and customer engagement – but without proper governance, they risk eroding trust faster than they gain efficiency. That’s the warning from Naveen Dhar (pictured), director of insurance strategy at Microsoft, who said that AI’s promise comes with real peril.
“My background has been insurance,” Dhar said. “I'm actually very fascinated by the intersection of technology with insurance and the changes that we can do in insurance.” That fascination now centers on governance – especially as carriers accelerate AI adoption in fraud detection and claims processing.
But speed without safeguards is risky. “In late 2023, we saw multiple lawsuits being filed on the claims side in healthcare,” Dhar said. With regulators circling, a structured strategy is no longer optional.
Dhar outlined a seven-part framework for responsible deployment, beginning with strong governance aligned to NIST and AIC principles. He emphasized the importance of using diverse datasets, integrating domain expert review, ensuring model explainability, and maintaining robust privacy protections. Continuous monitoring of performance and transparency with regulators round out the list. “Sometimes to go fast, you have to make sure your foundation is strong,” Dhar said.
Despite strong executive interest, few insurers have AI systems in full production. “Executives said that 90% would identify AI as a top initiative,” Dhar said, “but only 22% reported having AI solutions in production.”
Legacy systems and data quality remain major roadblocks. Even where AI offers measurable gains – such as reducing claims costs or accelerating processing – integration hurdles persist.
“The challenge is integrating it back into the legacy systems and the quality of data that we're collecting right now,” Dhar said.
Privacy concerns are also slowing adoption. Just 17% of insurers currently offer personalized insurance, and only 18% use AI-based models for wildfire risk – despite AI’s ability to integrate imagery, climatology, building permits, and loss history.
Dhar also sees untapped potential in predictive analytics – especially around customer retention. “The industry can still do a better job identifying patterns in customer data,” he said. That could help carriers act early to reduce churn.
But again, legacy tech holds them back. “What's really holding them back is data integration and the lack of modern tools around analytics.”
Customer-facing AI, meanwhile, introduces reputational risk. “There’s also the branding part of it,” Dhar said. If an AI-powered underwriting decision is flagged for human review, it must be clear why.
Transparency is critical. “If you have a virtual agent... and the customer thinks it’s a human, that’s going to be a challenge,” he said. Pilot testing should be standard. “These programs will identify edge cases, ensure compliance, and verify performance in the real world.”
And when AI fails? “No system is infallible,” Dhar said. He urged insurers to deploy contingency plans and PR strategies alongside their tech stack.
When asked how to modernize legacy infrastructure without sacrificing customer experience, Dhar recommended targeted investments.
“My first step would be assessing and prioritizing ruthlessly,” he said. Instead of full-scale replacement – “too expensive and too risky” – he advised focusing on low-complexity, high-impact areas like customer-facing portals.
“What I’m trying to do is spend the minimum amount of money and get the maximum returns. And... I would start with customer-facing digital wins.”
Cloud platforms and SaaS solutions can help insurers move fast without ballooning costs. These early wins should be tied to measurable outcomes. “It’s important to build that goodwill to generate enough money for your back-end conversion,” Dhar said.
He recommended a phased approach. “Don’t do a big bang here. Start with non-critical systems and reinvest savings.” Over time, carriers can shift to microservices and cloud-native models.
Success hinges on staying current. “There are ways and techniques that you can do things faster, cheaper, and better,” Dhar said. “Because I work for Microsoft, and we keep having all these new tools coming out.”
Transformation doesn’t require revolution – but it does demand discipline. “What we need is precision, pragmatism, and a relentless focus on value.”
