After years of hype and pilot projects, the insurance industry is entering a more demanding phase of its relationship with generative AI. Brokers and carriers are still investing heavily, but the conversation has shifted: what, exactly, is the return on investment, and how long will it take to materialize?
That question has become sharper as evidence from other industries tempers expectations. An MIT paper published earlier this year suggested that as many as 95% of firms have yet to realize measurable ROI from AI initiatives.
At least one expert argues the insurance industry may ultimately outperform that benchmark, but only if firms abandon “silver bullet” thinking and anchor AI deployment to specific operational problems.
“Insurance is an interesting industry because it usually is a little bit behind in adopting technologies,” said Anurag Shah (pictured), chief data officer at SIAA, the independent insurance agency alliance. “But when it comes to data and AI, it is probably one of the industries that can have the maximum impact, because the fundamental asset of this industry is data. So, there is a lot of potential for AI to play a role.”
Across both brokers and carriers, the strongest early ROI signals are emerging in efficiency-driven use cases rather than transformative reinvention. Faster quote generation, improved renewal rates, reduced claims cycle times and higher employee productivity are the most common areas where insurers can point to tangible gains.
On the broker and agency side, generative AI is being applied to renewals, policy comparisons, client communications and document processing. While they are relatively narrow workflows, they have direct financial implications, said Shah.
“If you improve renewals by even a small percentage, you can show the math very clearly,” Shah said. “You know what you spent, and you can see the uplift. It’s not going to cut costs by 90%, but it can still deliver a strong return.”
Carriers, operating at a greater scale, are targeting more complex parts of the value chain. Underwriting support, claims triage, fraud detection, pricing optimization and customer service automation are all high on the agenda. In many cases, generative AI is layered onto existing machine learning systems rather than deployed in isolation, accelerating decision-making rather than replacing it.
Those priorities can be seen in intellectual property filings. Data from Evident, an AI benchmarking and intelligence platform for financial services, shows that AI patenting in insurance is highly concentrated among a small group of US property and casualty insurers.
State Farm, USAA and Allstate account for 77% of all insurer AI patents tracked, many of which focus on claims automation, telematics and customer service. Generative AI has surged from just 4% of filings to 31% since 2023, according to Evident, particularly in claims and customer-facing processes.
Structural differences between brokers and carriers are shaping how each group approaches AI and how quickly they see returns.
Brokers, especially independents, tend to favour smaller, more tactical deployments. Their focus is on helping staff work faster and retain clients, rather than rebuilding core systems. These projects are cheaper, quicker to implement and easier to justify internally. The trade-off is that gains are incremental rather than transformational, Shah said.
Carriers, by contrast, are more likely to pursue ambitious, capital-intensive programmes that touch underwriting, pricing and claims. These initiatives offer larger long-term upside, but they also carry greater execution risk and longer timelines. That dynamic partly explains why carrier-led innovation dominates patent filings, while many brokers focus on commercial tools rather than proprietary platforms.
Compared with sectors such as retail or manufacturing, insurance’s vast historical datasets and skilled workforce should make AI adoption more natural. In practice, however, brokers and carriers face many of the same barriers seen elsewhere.
Data quality remains uneven, systems are siloed, and cultural resistance can slow adoption. Deloitte found that only around one in five organizations across industries qualify as “AI ROI leaders,” defined as those achieving strong financial returns, operational savings and rapid impact.
One reason ROI remains elusive is that insurers often struggle to define the “I” in ROI, according to Shah. AI costs are variable and opaque, ranging from cloud compute and data engineering to vendor licences and internal change management.
“People don’t know whether it’s going to cost $100,000 or a million dollars,” Shah said. "Because there’s so much uncertainty around the cost, there’s also uncertainty around ROI. People don’t know what to expect. To me, this feels very similar to the early days of the internet. If you had asked someone back then, “What is the ROI of building (a website) on the internet?” they wouldn’t have had a clear answer. People questioned the value of having a website at all. We’re back in that phase now, where it’s difficult to quantify value while you’re in the middle of the transition."
As a result, many firms are moving away from traditional technology ROI models and towards use-case-level measurement. Instead of asking whether AI as a whole is paying off, they ask whether a specific deployment reduced handling time, improved retention, lowered error rates or increased throughput.
Global claims firm Davies, for instance, has noted early gains from applying generative AI to targeted parts of the claims journey. Technology and AI are core to the organization’s growth plan over the next five years, but CEO Dan Saulter has admitted that there’s no clear way to measure the immediate impact of the digital transformation.
“It’s too early to be very precise, or it’s too early to be precise about the effect it has on the whole business,” Saulter told Insurance Business. “Right now, you do have to throw quite a lot of stuff at the wall and see what sticks… not everything is going to work, and I think you do have to be aware of that.”
The CEO is instead measuring ROI in terms of tangible benefits. “If you measure teammates’ time, you can start to get a picture,” said Saulter.
“If you scale that piece of the automation, how many five-, ten-, twenty-minute pieces of time can we save for this handler or for this adjuster? You can start to see that this (initiative) should really unlock a lot of productivity gains, and therefore it should unlock more profitable business for us.”
This reflects findings from Deloitte’s 2025 survey of nearly 2,000 executives across Europe and the Middle East. The survey showed that most organizations expect AI investments to take two to four years to deliver satisfactory returns, far longer than the typical seven- to 12-month payback expected from conventional IT projects. Only 6% of respondents reported AI payback within a year.
Crucially, Deloitte noted that AI “rarely delivers value in isolation.” Gains are often entangled with parallel initiatives such as data clean-up, process redesign and workforce reskilling, making precise attribution difficult. In insurance, where legacy systems and fragmented data are common, this challenge can be particularly acute.
Beyond insurance, the rapid expansion in AI technologies and investment has sparked growing concern among investors, economists and regulators that markets may be experiencing an AI-driven financial bubble.
Market volatility and investor anxiety have intensified this week after major tech company shares, including Oracle, fell sharply following weaker-than-expected earnings and heavy AI-related spending, stoking fears that AI valuations may be unsustainably high. Nasdaq and other tech indexes dipped amid this wobble.
However, Shah rejected that framing and affirmed that AI would create real results and benefits. “What’s happening is a correction in expectations,” he said.
One driver of unrealistic expectations, he argued, is the outsized attention paid to generative AI chat interfaces. While conversational tools are visible and intuitive, they represent only a small part of what AI can do.
“If your expectation was, ‘We are going to put in the chat interface with generative AI and my claims experience is going to become X times better,’ that is the kind of unrealistic,” Shah explained. “There is a series of things you need to do across the workflow to make the overall experience better.”
That insight is increasingly shaping investment decisions. Organizations are becoming more selective, doubling down on projects that demonstrate value and abandoning those that do not. According to Deloitte, investment in AI continues to rise despite unclear ROI, driven by competitive pressure and belief in long-term impact.
Over the next two years, generative AI is expected to continue delivering short-term productivity gains, while more advanced “agentic” AI (systems capable of managing multi-step processes autonomously) remains a longer-term bet. Deloitte found that only 10% of organizations using agentic AI currently see significant ROI, though many expect returns within three to five years.
Looking ahead, Shah said the insurance industry’s AI ROI story is likely to be defined by high-quality, well-governed data, careful selection of use cases tied to business pain points, and, ultimately, cultural adoption by employees.
“Adopting AI requires a new way of working, and it’s natural for people to resist that change,” he said. “The negative concerns are easy to focus on: will it replace people? Will it replace jobs? To some extent, those effects may occur.
“However, the broader outcome is likely to be far more beneficial overall. AI has the potential to create more opportunities than it eliminates, even if that’s difficult to see in the short term.”
