Forget fully automated cars. How about a fully automated MGA?

Can you imagine a completely automated Managing General Agent (MGA) using artificial intelligence to serve its insurance company and brokerages partners?

That day is coming with the rise in agentic AI, says Sridhar Manyem, senior director at AM Best.

Mayhem was speaking about the applications and implications of agentic AI on the insurance industry at AM Best’s Insurance Market Briefing, held in Toronto last Thursday.

“In terms of insurance, [people] are starting to talk about agentic AI,” Manyem said. “Agentic AI is not just one AI engine, but it is a artificial intelligence ecosystem that can perform a specific goal with limited supervision. So, for example, it can be a MGA, right?

“An automated MGA takes submissions from various clients, analyzes that submission, looks at that submission and says, ‘Here’s the risk posed by this particular submission.’

“And then it looks at a carrier and says, ‘You know what? This carrier has told me that this particular risk matches their risk appetite.’ And they can transfer that risk automatically to the carrier,” he said. “So, it’s like a whole system that can mimic human decision-making in order to solve problems in real time.”

Manyem said this kind of AI ecosystem is “starting to get traction” in small commercial underwriting and emerging risks like cyber.

Agentic AI, powered by Big Data

“The beautiful thing about agentic AI,” he notes, is that it’s able to integrate diverse data sources such as telematics, social media activity, behavioural insights, and more, in order to make a dynamic risk assessment in real time. He cited an actuarial paper he read, which noted AI can even use data sources that underwriters may not think to use in a risk assessment.

“In that particular actuarial paper, they were saying, ‘You know what? We can collect information like the time of collision, the kind of car, etc. But we don’t really think that a car’s power-to-weight ratio is critical in determining the severity of an accident,” Manyem said. “Because of machine learning and big data, the machine spit out variables like power-to-weight ratio as an important determinant in the severity of an accident. So now that becomes a variable in assisting underwriting and pricing.”

Also in the news: U.S. specialty insurance firm to acquire SSRU

The promise of agentic AI is predicated on large data sets, as Manyem noted.

To make the most out of agentic AI, the insurance industry will need to get ready to take advantage of the explosion of big data. For that, insurers will need to focus on collecting structured data.

As of now, insurance companies have disparate systems.

They may have systems that carried over from a merger or acquisition. Or they may have individual operational systems for different functions.

“They would have a system for claims, a system for underwriting, a system for policies, a system to talk to the regulators,” Maynem said. “All this data that they have collected over hundreds of years is really not ready for AI, because they are all different, they’re all in various places, and [insurers] need to make sure that [their data are] structured, they’re organized, so that they can gain proper insights.”

Plus the data needs to be updated to reflect current trends and patterns.

Data poisoning

This reliance on data makes it very important for insurance companies to collect, use and protect data responsibly.

“Maintaining the integrity of the data will be very important for the insurance industry in the future,” he said, and segued onto the topic of future cyber threats.

“One particular aspect of cyber threats I wanted to talk about is this thing called, ‘data poisoning,’ or ‘model poisoning.’ I’m not sure why they would do this, but bad actors apparently inject bad data into your system. And therefore, when the model is getting trained on that data, it spits out wrong results or wrong inferences, etc.,” he said.

“That’s becoming more common, so you need to make sure that the data that you’re using is protected and you don’t let any bad actors into the system.”