Traditional carriers remain cautious about generative AI, while insurtechs adopt it

ChatGPT took the world by storm a few months ago, making artificial intelligence technologies a kitchen table topic. Insurers have been using forms of AI for years. However, the generative aspect of GPT won’t likely be deployed by legacy insurers until there is explainability in the models. 

Michael Nadel, a senior director at Simon-Kucher, says generative AI as a concept has taken off and there are many potential use cases. 

“In the insurance industry, nothing moves quickly,” Nadel said. “So, in terms of where it’s going to land in the insurance ecosystem, I think there’s still so much whitespace.”

Nadel does see the technology increasing productivity and revenue for insurance agents by standardizing manual tasks like writing emails, sales pitches and social media posts. 

“I think the other aspect from a productivity standpoint is this ability to be on the other side of the equation when you’re talking to somebody about a policy,” Nadel added. “When it gets to these more complex policies, whether it’s life insurance or a more substantial commercial policy, the ability to use generative AI to get in there and really answer a lot of these product questions — What are my limits? What are my coverages like? What are the deductibles? How could that change if I tweak something? — this makes it a lot easier for the agent.”

Additional areas where generative AI could be useful include potential personalized product recommendation engines and helping agents with sales strategy, Nadel said. 

Potential bias

Despite their potential usefulness, AI technologies come with a number of risks. Datasets can often reflect current racial, gender and societal biases.

At Monitaur, an AI governance software company, CEO Anthony Habayeb says insurance as an industry is already motivated to be effective at risk management, especially with the increase in more complex and automated decisioning technologies. Consumer conversations and fears around AI have driven regulators to step in.

“I like the fact that we’re talking about risk management frameworks and governance as the methodology for regulating this technology,” Habayeb said. “And the reason is, asking carriers to demonstrate good risk management frameworks gives them an ability to operationalize a process that works for their size and scale, but gives regulators an ability to have confidence or visibility into how aware companies are of their technology, the data that they’re using, the people building these systems, and build good processes over them. So, I think we’re moving into a place that hopefully will have good risk management frameworks and enable responsible innovation.”

Insurtechs plunge ahead

Such challenges haven’t stopped insurtechs like Roots Automation, Cowbell Insurance and Capitola from releasing GPT technologies. 

Roots Automation, an insurtech with digital coworkers, released InsureGPT, a generative AI model designed for the insurance market, in May.

Chaz Perera, CEO and co-founder of Roots Automation, said, “Large language models have been able to allow people or companies to take unstructured information and essentially try to synopsize what it’s saying, potentially extract relevant information out from these massive documents. The fundamental problem with just using ChatGPT, or just using Bard or just using some other open source tool is that those platforms are exposed to the world. So any data that you put on it, particularly for an insurance company, private information about their customers and the company, they don’t want that stuff out in the wild.”

The second reason has to do with expertise, or really the lack of expertise that a ChatGPT or a Bard has in a particular industry, he said.

“These things are trained only on publicly available information,” Perera said. “They understand some insurance language, in the context that insurance companies have websites, and they’ve put some information out to the public about their products and the services they provide, And these technologies have used that to learn. But they don’t have an intimate understanding of the idiosyncrasies of insurance language.”

Roots’ InsureGPT is an insurance expert large language model, he said.

Rajeev Gupta, co-founder and CPO of Cowbell Insurance, says the company released its MooGPT to cater to insurance agents and policyholders so they can engage in more of a conversation with chatbots. 

“We provide chat capabilities for our policyholders and agents to connect. With most of the older chatbots conversations are more predefined, it’s very dry and transactional and not fluid. It’s not a human-like interaction and [MooGPT] brings this amazing ability for us to interact in a more conversational manner, to present data in a format that humans can easily digest.”

Sivan Iram, co-founder and CEO of Capitola, a digital marketplace for commercial insurers, says the introduction of GPT technology into its platform because the technology is now more reliable.

“We’ve spent months making sure of the reliability of the information, the security and its ability to stay contained,” Iram said. 

The company released its Capitola Co-Pilot in June, the technology uses GPT to offer brokers productivity enhancements. Iram sees GPT-tech being helpful in bridging the talent gap as it can help broker teams produce more but it is also necessary to retain workers. 

“One of the biggest trends in the workplace is consumerization of the enterprise,” Iram said. “It really means to put the user at the center of the user experience and it’s a game changer in terms of a company’s ability to retain people and give them tools that optimize their flows. When we talk to brokers, what we hear is, they use the most advanced technologies at home but then they go to the office and they use software that feels like it was built 20 years ago.” 

What’s ahead

Insurers are largely staying the course on digital transformation. Customer experience is the top priority by 48% of carrier and broker respondents for the second straight year; 85% rank it in the top three, according to recent research from Arizent. The survey was conducted during the rise of conversational AI and LLMS, and so may represent the interest around these emerging offerings.

Habayeb says that while people are talking about LLMs a lot, carriers aren’t actively deploying them. He believes there are a few things at play here including a need for insurtechs to develop robust governance.

“As I think about the insurtech ecosystem, if you are an insurtech that is attempting to sell a solution or value to a larger insurance enterprise, and that solution or value is based on machine learning or AI, you should really take a pause and look at how do I tell that customer that I have good governance and quality around my application?” he said. “Because they are already starting to ask vendors: ‘Hey, I can’t buy your AI unless you give me proof that you have some good control around your system.'” 

Good AI needs great governance, Habayeb said. 

“So I think over the next 10 to 15 years we’re going to see the data providers, the insurtechs, the carriers, the reinsurers, the regulators, there’s going to be a healthy normalization that proof of good governance is critical throughout the entire value chain. It’ll become table stakes. You can’t do business with AI and insurance without evidence of good governance.”

Nadel believes insurtechs will continue to have challenges in partnering with large carriers to deploy the tech. 

“Insurtechs are going to face the general risk averse nature of the insurance industry and commercial carriers specifically,” he said. “Ironically, the legacy insurer is at a huge advantage in deploying these types of models over insurtechs. Generative AI models require so much data to actually be valuable.” – data that legacy insurance companies have.

However, he noted, many traditional carriers aren’t equipped with data warehouses that allow easy access to data.

“A lot of them are going on these journeys to the cloud and migrating from these legacy data stores and making sense of redundant data and quantifying it in a way, storing it in a way that’s easy to get at,” Nadel said. “If the Allstates and the State Farms of the world can figure that out, they’re at a huge advantage.”