Insurers are beginning to demonstrate measurable efficiency gains from AI across underwriting, claims and operations, but executives say proving whether those improvements translate into meaningful business returns remains one of the industry’s biggest unresolved challenges.
The issue was raised during a panel discussion at the InsurTech Connect (ITC) Asia 2026, where insurance executives debated how insurers should assess the value of AI investments as deployments become more widespread.
“The measure of ROI is difficult,” said Darren Ma, Chief Actuary for Asia Pacific at Berkley Insurance Asia. “People learning, people going on courses, people attending meetings… all of that is investment time, and that’s a little bit harder to capture,” he said.
Ma said insurers often underestimate AI’s true implementation costs, which extend well beyond software and hardware to include staff training, governance, and security reviews. Many companies also lack visibility into the per-person cost of providing staff with access to AI tools, which is likely to increase over the coming years.
The benefits can be equally difficult to quantify.
While AI can significantly reduce the time required to review submissions, whether those time savings translate into improved underwriting performance depends on how insurers use the additional capacity.
One way to isolate AI’s impact would be to compare teams with and without access to AI tools over time, but described such an experiment as “nearly impossible” in practice.
Instead, insurers often overstate AI’s benefits by failing to account for implementation costs.
“We need to compare apples and apples,” said Olivier Michel, Founder and CEO of Ancileo. “People tend to inflate the impact and kind of decrease the cost.”
Michel said one of Ancileo’s early AI claims initiatives delivered clear productivity improvements, but required not only software investment but also a dedicated implementation team and ongoing human oversight to ensure compliance.
He added that AI projects frequently exceed their planned implementation timelines, delaying expected returns.
To avoid over-investing in AI projects with uncertain returns, Michel said Ancileo evaluates initiatives against three criteria: proprietary data, the AI tools required, and the workflows needed to make the technology usable.
Insurers should also evaluate AI investments by comparing the cost of implementation against the cost of inaction and tying each initiative to a clear underwriting, claims or operational objective, rather than treating technology adoption as an end in itself.
“Tech is the biggest component of the cost in driving transformation, but honestly, it’s about the business value or the outcome,” said Sourabh Chitrachar, Chief Technology Officer at MS First Capital Insurance.
From pilots to proof of value
While executives acknowledged the difficulties of measuring financial returns, they pointed to a growing number of operational AI deployments already producing measurable efficiency gains.
At Zurich, Chief Growth and Propositions Officer for Japan P&C Roopa Malhotra said the insurer has piloted an agentic workflow for commercial insurance that reduced quote-to-bind times to just a few minutes. After initially deploying the system in one market and one line of business, Zurich is expanding it to three additional markets and two more business lines by the end of the year.
At MSIG, Senior Vice President and Head of Transformation Rajnish Pal said they are tracking operational KPIs across claims, contact centres, policy renewals and endorsements, while redesigning commercial underwriting workflows so AI can handle more administrative work and underwriters can focus on portfolio management, risk selection, pricing and better risk decisions.
Pal said one commercial underwriting initiative has moved beyond the pilot stage into what MSIG internally calls “proof of value,” with scaling across multiple markets expected in the second half of the year.
Even so, executives cautioned against applying the same measurement framework to every AI initiative.
Operational AI initiatives naturally lend themselves to quantifiable metrics such as time savings, process automation and speed to market. AI investments aimed at developing future business models, however, require longer time horizons and greater tolerance for experimentation.
“We need to be careful not to get lost in KPIs, especially when we are thinking about future models, because this is a very evolving space,” Malhotra said. “We have to allow for a bit of experimentation.”
Some of AI’s value may only become apparent over time through better risk selection, stronger portfolio quality and improved loss ratios rather than immediate productivity gains.
Pal said AI could help insurers filter submissions more effectively and support longer-term portfolio management, while cautioning that additional underwriting capacity should not automatically translate into more business.
“Distribution will bring in more business, but just because we have additional capacity doesn’t mean that we take on more business,” he said.
AI could also help insurers write the business they prefer, which should support growth at stable or improved loss ratios over time, though returns from new business or embedded insurance may take “three or four years” to materialise, Ma added.
