Will AI prove useful in risk management?

The AI hype crowd promises the technology will transform just about everything—even raising a newborn, apparently. For the risk-minded finance leaders thinking through the implications of all this promised AI disruption, we have news for you: AI is coming for your risk management function, too, if it hasn’t already.

Nearly two of every five (38%) organizations use AI in risk assessment and analysis, according to insurance broker Gallagher’s 2026 Business Owners Survey, which polled 1,000 US-based owners in late January and early February. Similarly, 36% of respondents said they’re using AI for mitigating risk “within their insurance and risk management programs.”

As with AI investments in other functions, like corporate dealmaking, the CFO needs to monitor whether RM applications of AI are paying off. According to experts, that typically means tracking cost avoidance.

It’s similar to expert advice on ERM programs at large: Find appropriate metrics that tie investments to financial returns and monitor whether they improve after implementation. The metric could be something like a measurable reduction in insurance claims, accident rates, or certain workplace injuries.

The “primary ways” to demonstrate ROI in these cases also happen to be “hidden ways,” since the technology is used to minimize risky events or at least reduce their likelihood of happening, Arpan Podduturi, VP of product at Samsara, a safety technology platform, told CFO Brew.

When measuring these changes in risk-event occurrences over a period, organizations can “create this ROI that is very, very obvious,” Podduturi added.

For the skeptical CFOs who think they can just sit this one out: Doing nothing also has a cost, according to Andrew Zarkowsky, head of AI underwriting at The Hartford.

“Doing nothing is risky, because not developing is going to put you behind,” Zarkowsky said during a presentation at the Riskworld conference in May. “There’s not a no-risk scenario.”

Where the money’s going. Some larger businesses, such as retailers, are using AI to help identify potentially fraudulent activities or predict what type of workplace incidents are likely to result in costly insurance claims or litigation, according to Stephen Rhee, global chief digital officer of Gallagher.

For example, AI can look at the history of claims and identify commonalities in those that usually go to litigation, Rhee said, and pinpoint the average settlement amount the company has to pay in those cases versus for claims it settles before they go to court.

“That’s where you see a lot of large enterprises with high claim activity really focused, because that is managing cash flow literally every day. How much am I paying in claims?…So we see a lot of use cases there,” he said.

When incorporated into vehicle cameras, AI can help spot risky behavior like distracted driving or external risk factors like poor road conditions, Podduturi noted. AI can also analyze vast sums of data to help companies predict when they’ll need to perform equipment maintenance.

Start here. The opportunities may feel overwhelming, especially for organizations that are only beginning to explore their options, according to Podduturi. Rather than getting hung up on all the technical capabilities of AI models, Podduturi recommended that organizations view AI options through the lens of desired results—such as reducing accidents or maintaining truck schedules.

“These are very practical, concrete uses of the technology,” he said.

Spread the wealth. AI investments in risk management could benefit other functions. Zarkowsky shared an example of a globally operating “large metal manufacturer” that installed AI technology for predictive maintenance. As it turned out, that manufacturer also found the technology helped with forecasting demand.

“That type of use case, when you bring that to a CFO, you get funding, and you get funding real quick,” he said. “This is a really interesting idea about how you can take some risk management concepts, pair them with production ideas, and actually be able to have a dual use case.”