Artificial intelligence, all the rage in popular culture, is not a new fad in the insurance market, where it quietly powers operations that make businesses run more efficiently, at lower costs and with fewer errors.

Still, AI is in its infancy and its rapid development means insurers, brokers and their customers should see the technology continue to broaden throughout customer service, claims handling and sales, experts agree.

Thanks to the viral popularity of ChatGPT, an AI chatbot developed by OpenAI that allows users to perform tasks that range from frivolous to serious, AI is suddenly mainstream. But having a chatbot write a poem about a pet is not the same as having it handle claims or steer buyers to coverage, which is why AI in insurance is being carefully implemented in select areas of commercial and personal lines.

“Six months ago, nobody was talking about this,” said John Cottongim, New York-based chief technology officer at Roots Automation, which provides tech services to the insurance industry. “We are in the toddler phase of the capabilities” of AI, he said. “It’s going to be a wild ride over the next 10 years.”

By then, it’s likely that “we’ll be able to just point the platform at your website or knowledge base, click a few buttons and you can have a really good virtual agent live and supporting it,” said Bill Schwaab, San Francisco-based vice president of North America at Boost AI AS, a Norwegian company that trades as boost.ai and develops conversational AI platforms for insurers. Even with such advancements, however, back-end systems will still require work to maintain, he added.

AI is well-suited to claims handling, sources agree, partly because of the wealth of insurance data available for it to pore through to help determine which claims need immediate attention or how they can be processed with less human interaction. 

“Insurance is kind of the original database business; we have relied on data and data science capabilities since we existed as an industry,” said Mano Mannoochahr, chief data and analytics officer at Travelers Cos. Inc. in Hartford, Connecticut, in a recent webinar. “The opportunity that we have, generally from an AI perspective … is to be able to reimagine and rethink all parts of our business.”

Employers have long clamored for a system that would streamline the workers compensation claims process, said Dennis Tierney, Marsh LLC’s Norwalk, Connecticut-based national director of workers compensation claims. In January, the broker introduced its AI-powered process to identify claims that may not have received proper attention based on the old approach of selecting files largely based on reserve amounts, he said.

Using AI, “we’re helping identify claims that are not on everyone’s radar,” Mr. Tierney said, and prioritizing them by considering such factors as jurisdiction and type of injury.

At Travelers, an AI model is trained on millions of high-resolution images of insured U.S. properties, Mr. Mannoochahr said. After a natural disaster, the model can quickly assess damage and begin the claims process, in some cases before the owner returns to the property.

“That allows us to make better decisions about where we deploy our adjusters and claims handlers and how we prepare them for the onslaught of calls that we may get,” he said.

Beyond claims, virtual agents in an AI setting can be trained to sell, Mr. Schwaab said. “The data from the virtual agent can be leveraged to do a little bit of selling on behalf of the company” by encouraging customers to consider additional products that it senses they may need, he said.

Chatbot technology can use real-time analytics to gauge customer sentiment and help virtual agents make decisions, including customized product suggestions, said Mamta Rodrigues, New York-based divisional president of banking, financial services and insurance at Teleperformance SE, a Paris-based digital business services company.

“AI is becoming increasingly accurate and further augmenting the digital experience, making that conversational bot so much more intelligent and personal,” Ms. Rodrigues said.

Complex commercial lines may take some time to address with AI, though, said Mr. Cottongim of Roots Automation. “Personal lines are suited toward the common natural language that we speak every day, which these algorithms are trained on,” he said.

But while the platforms can help with auto damage claims, interpreting a clause in an errors and omissions policy would be another story, Mr. Cottongim said. 

“That’s quite esoteric and it’s unlikely that the algorithms have been trained on that,” he said, adding that fine-tuned models are expected to eventually address complex commercial issues.

It would be a mistake, though, to discard the human touch completely in favor of AI platforms, Ms. Rodrigues said.

“We’re doing quite a bit of proof-of-concepts and demonstrations for some of our major clients to show how a bot platform can generate stronger efficiencies,” she said, “while not taking away from the personal touch that comes from a human.”

Employers need guardrails to reduce risks that come with chatbot use

Artificial intelligence presents both benefits and threats to risk management.

On the one hand, AI can assist risk managers in identifying potential risks and automating routine tasks. However, AI also poses risks such as algorithmic bias, lack of transparency, and cybersecurity vulnerabilities.

At least, that’s how ChatGPT sees it. The preceding paragraphs were written by the bot when asked whether AI is a threat or a benefit to risk management.

A self-assessment that’s not far off the mark, according to Karla Grossenbacher, a partner with Seyfarth Shaw LLP in Washington. 

While it is unlikely that employers are sanctioning the use of ChatGPT solely to provide advice or make business decisions, they need strong policies that ensure it won’t happen, she said.

“Employers may not be necessarily thinking about what they need to do to make sure it’s not being misused,” Ms. Grossenbacher said. “There are things that employees could ask ChatGPT to do and those could be leveraged, but there are certainly situations where you would want to prohibit it.”

When developing AI workplace guidelines, employers should address the risk of release of confidential or proprietary information, Ms. Grossenbacher advised. And there is the danger that information gleaned from publicly available bots could be inaccurate or that it could be subject to copyright protections and if used without proper credit charges of plagiarism could result, she said.

“It would be prudent to call upon your risk manager to weigh in on these types of issues,” Ms. Grossenbacher said.

The U.S. Department of Commerce’s National Institute of Standards and Technology offers some guidance in its 42-page “Artificial Intelligence Risk Management Framework,” issued in January for organizations designing, developing or using AI systems.

The voluntary framework promotes changes in institutional culture and suggests ways organizations can measure and monitor AI risks and consider the potential benefits and threats of the technology.