Insurers spent a decade building friendlier screens on old systems. Now that artificial intelligence has arrived, many are discovering those screens hid the problem rather than solved it. Legacy insurance systems remain the first transformation challenge leaders' name, and the ones who have fixed them describe a process that is slower, more disciplined and more organizational than most technology roadmaps admit.
Few carriers set out to accumulate technical debt. At Encova Insurance in Columbus, Ohio, it came through growth. "Through that model, we accumulated 16 companies across 26 states," John Kessler, executive vice president and chief strategy officer, said in an interview on how Encova built one modern carrier. "Because we allowed each to maintain their identity, we layered infrastructure on top of infrastructure as we grew. Long story short, we became an organization that was bloated and deep in technical debt."
Anthony Slimowicz, executive vice president and chief operating officer of MGA and Specialty Risk at Amynta Group, sees that pattern across the market. "Most traditional insurance organizations are still operating legacy greenscreen back-end systems and have simply built more user-friendly outgoing platforms that sit in front of or around them," he said.
The cost shows up in people's time. Hasani Jess, chief technology officer at London-based Aventum Group, calls it "human glue": the double-keying and manual transfers that fill the gaps wherever two systems were never designed to work together. Aaron DiCaprio, US head of programs at Accelerant, put the AI consequence plainly. "AI doesn't solve fragmented, inconsistent data. You need the infrastructure to ingest, standardize, validate, and understand that data first," he said.
Encova chose a replacement. Kessler's team rationalized more than 30 core systems under three pillars – standardize, modernize and rationalize and selected Guidewire Insurance Suite as its target platform for property-casualty lines. "We took a true greenfield approach," he said. "We pulled leaders from every discipline across the organization into an integrated project team, and we burned the boats; everything was net new going forward."
Aventum rebuilt. Rather than stitch point solutions together, Aventum built its ATOMX platform entirely in-house, retaining the intellectual property and bringing in partner teams that embed within its organization. Jess understood the trade-off. "Moving really, really fast in a domain as robust and regulated as insurance brings its own challenges," he said.
Amynta protects. For a group running very different specialty businesses, Slimowicz said the MGA edge lies in judgment, not platforms. "The biggest challenge is understanding what can be modernized or moved to a common platform versus the things that make the business unique and different," he said. After an acquisition, the warning is sharper: "The worst thing you can do is come in, put your letterhead on the front door, and say, 'Okay, now you're on our system' and lose the entire culture and expertise you purchased in the first place."
"We burned the boats; everything was net new going forward."
The leaders who have made progress describe three disciplines rather than one technology choice.
Write the destination down before choosing the vendor. Encova's then-chief executive wrote a 31-page vision of what the company should look like in 10 years and read it aloud to employees across all 16 affiliated companies before execution began. That plan was broken into 10 strategic directions and 350 to 400 objectives of 12 months or less.
Decide what to standardize and what to protect. Slimowicz's test is whether a line's systems serve needs that are genuinely different, as workers' comp, surety and trade credit do, or whether difference is simply habit.
Organize integrated teams, not IT projects. "Being siloed off from the business is a perpetual challenge in this industry," Kessler said. "We solved that by adopting a guiding principle: everything we do goes through integrated teaming."
The payoff is in the next phase. Encova now runs on two core platforms, down from more than 30, and Kessler said that changes its starting point. "Because we've built modern platforms, we're now well positioned to take on the challenges and opportunities that come with AI," he said. "We don't have to solve the operational challenges of the past first."
Carriers still running on layered infrastructure face the opposite position: every AI project begins by working around the core. Part 2 of this series examines why so many of those project's stall at the pilot stage, and what leaders are doing to get them into production.
The AI pilot is the easiest part of insurance transformation. Getting it into daily work is the hardest. Clayton Fuchs, vice president of casualty at Austin, Texas-based wholesale brokerage QuoteWell, pointed to a widely cited MIT-affiliated study finding that most enterprise generative artificial intelligence (AI) pilots fail to produce a measurable return, largely because the tools are never properly built into how people work. In an interview on why AI adoption will split wholesale brokers, he said loading a chatbot into an email inbox does not count as using AI effectively.
The reasons pilots stall are mostly human, and so are the fixes.
Rajeev Khanna, chief information officer at Lindon, Utah-based brokerage Trucordia, has led transformation programs across several industries and explained why many transformation programs fail before launch. "Any large transformation program doesn't usually fail because of technology," he said. "It fails because of the organizational change management. It's typically a lack of adoption, a lack of engagement."
Momentum is the second problem. Hasani Jess, chief technology officer at London-based Aventum Group, said belief in a program fade unless people can see it taking shape. "Vision casting only lasts so long. People buy into the vision, but then they need to see the hallway. The kitchen isn't ready yet, but they need to see something more," he said.
The third is automating the wrong thing. Clayton Matthews, director of claims at Georgia-based Oakbridge Insurance, put it bluntly. "Don't automate a disaster faster," he said. "If it's a bad process, why would you speed it into automation?"
The firms getting past the pilot stage involve end users from the first day. "I don't want to guess what the business needs," Khanna said. "I want to understand it from the people who are living it on a day-to-day basis." At Trucordia, business stakeholders take part in requirements, reviews and testing rather than receiving a finished product.
Atlanta-based Crawford & Company goes further. David Wright, chief AI, data, and cloud officer, said Crawford tests every AI tool with adjusters first, and they can reject it. "They could say no," he said. "Or: is this accurate? That's actually a really big one. We have to repetitively test and make sure that across multiple uses, it's stable and accurate." If adjusters flag problems, the tool is shut down and refined until they approve it.
QuoteWell embeds technology staff inside daily brokerage operations, using the forward-deployed engineer model several of its founders used at Palantir. "Adoption is a goal – not a nice-to-have outcome," Fuchs said. Each fix is tested against whether producers trust it enough to stop doing the task themselves, then killed or scaled.
"Adoption is a goal – not a nice-to-have outcome."
Start with a defined problem and clean data. David Fishel, chief growth officer at Texas-based Higginbotham, described a crawl, walk, sprint approach to AI in which the firm first understands the technology and confirms it fits the business strategy before committing, because any AI tool is only as good as the data it receives and is a tool, not a strategy.
Fix the process before automating it. Matthews' warning applies to every function: a pilot built on a broken workflow will scale the breakage.
Co-design and give users a veto. Khanna and Wright both treat frontline staff as the gatekeepers of adoption, not its audience.
Measure behavior, not demos. QuoteWell's test is whether people stop doing the work by hand. Anurag Batta, president of Bridge Specialty Group, described a similar standard. "When the metrics demonstrate results and actually help people, buy-in follows naturally," he said.
Show progress in small pieces. Jess's "hallway" is the visible proof that keeps a long program funded and believed.
Keep the vendor list short. Trucordia deliberately limits its strategic partners to avoid the overhead of managing too many relationships, using proofs of concept to decide which ones deepen.
The goal, Wright said, is to turn scaling into a habit rather than a project. "What we really want is to enable hundreds and hundreds of workers to be the sources of ideas and innovation," he said.
That depends on people who know the business well enough to judge what AI produces. Part 3 of this series examines the human capital side of transformation: how insurers are capturing expert knowledge and passing it on.
Six surety underwriting trainees sit outside Brock Masterson's office in Morristown, New Jersey, beside four executive underwriters averaging 25 years of experience. For Masterson, senior vice president and chief operating officer of the surety division at Crum & Forster, that seating plan is the transformation that matters most. Knowledge capture and sharing has become the human capital challenge at the center of insurance transformation: AI can speed up how people learn, but only if someone has captured what the veterans know before they leave.
Insurance Business America's coverage has cited a projected shortfall of roughly 400,000 insurance jobs by 2035, a U.S. Bureau of Labor Statistics projection reported in a 2025 ReSource Pro analysis.
"Nobody goes to university to study surety. There are very few transferable skills from other industries," Masterson said. "This is an industry of trial and error – your surety knowledge is based on your experiences." Software can measure a contractor's capital and capacity, he said, but not character. "That is where underwriting cannot be replaced by an AI model – and it's the most prevalent part of the work."
Automation adds a new risk. Kristen Skender of Jimcor Agencies has warned that if AI absorbs the entry-level submission work that once trained new underwriters, the first rung of the career ladder may disappear. Clayton Matthews, director of claims at Georgia-based Oakbridge Insurance, worries about what replaces it. "What happens when an AI tool gives what appears to be a mostly right answer and a young adjuster takes it as gospel because it sounds authoritative? That concerns me deeply," he said.
Douglas Turk, president of specialty at Chicago-based Relation Insurance Services, described how Relation records specialist knowledge for AI recall. The brokerage interviewed several of its specialists for six to seven hours each, asking what to look for when placing cyber insurance and what normal limits look like, then turned the transcripts into a specialty advisor model any employee can query. "That shortens cycle time," Turk said.
Turk was clear the model supports specialists rather than standing in for them. "You never replace that," he said. "There's always a specialist in the process."
Masterson built his own daily AI briefing on court cases, project awards and bankruptcies, with an explanation of why each matters to surety, and shares it with trainees. "That's how we use AI to boost less experienced underwriters – accelerating their training and the value of their conversations by giving them access to data that previously took a 25-plus-year career to accumulate," he said.
"Like any tool, AI is most powerful when paired with experience."
At Texas-based Higginbotham, chief growth officer David Fishel said AI accelerates knowledge transfer to the next generation, while the apprenticeship model stays intact. "AI is accelerating how quickly someone can learn the technical side of this business, but it doesn't replace wisdom, judgment or relationships," he said. "Like any tool, AI is most powerful when paired with experience."
Put the generations side by side. Masterson's seating plan is deliberate, and Fishel describes pairing digital-native hires with seasoned producers as the fastest way to move expertise.
Record the veterans before they retire. Relation's interview-and-transcribe method is simple to copy, and it turns one specialist's time into a resource the whole firm can use.
Teach the why, not just the process. "We need to be laying out the pitfalls and the 'whys' – why we do certain things – for the younger generation," Masterson said. Jennie Carr, vice president of commercial brokerage at Arlington/Roe, agreed. "I don't want to just train people to be good at processes. I want them to understand how the business works," she said.
Train people to question the machine. Carr said part of that training is knowing, when AI gives you an answer, to ask whether the question was framed correctly. That habit is the safeguard against the authoritative-sounding answer Matthews worries about.
The payoff, leaders said, is time handed back for the work only people can do. "If technology can give a broker back an hour in their day that they can spend talking with an agent, learning a new coverage, negotiating with an underwriter, mentoring a younger associate, or finding a solution for a difficult account – that's the real return," Carr said.
Across this series, the pattern holds: fix the core, build AI with the people who will use it and capture what experienced staff know while they are still there to teach it.
