As companies across industries increasingly look to artificial intelligence to automate entry-level work, there are growing fears that they may be eliminating the very roles that once served as training grounds for future experts.
In insurance, where professional judgment is built over years of handling complex claims and client interactions, that trend could have unintended consequences.
For Joel Raedeke (pictured), chief technology officer at Crawford & Company, the technology is exposing a paradox. While AI can dramatically increase productivity, he believes that it delivers the greatest value when used by seasoned professionals rather than inexperienced staff. In some cases, relying too heavily on AI early in a career may even create new risks.
"When you're new at something, you don't know what good looks like," he told Insurance Business. "There's a strong chance of doing a lot of bad things at scale without recognizing it."
As insurers accelerate AI adoption, Raedeke believes that technology is unlikely to replace expertise. Instead, it may make expertise more valuable than ever. The principle particularly applies to generative AI
“Someone new may look at the output and think, 'This is amazing,'” the CTO said. “But only someone more experienced can immediately recognise what's weak or sloppy about it."
Crawford has been treating AI not a standalone initiative, but the latest stage in a long-running effort to embed advanced technology into claims operations. Raedeke said the company subjects AI-powered workflows, "including… machine learning, predictive modelling and now large language models,” to extensive testing before they reach adjusters.
But while generative AI can produce impressive summaries and insights, Raedeke stressed that an increased level of AI scrutiny is important in claims, where small errors can influence outcomes for policyholders and carriers alike.
"If you've used ChatGPT to summarise something, you'll often find it's amazing in terms of what it can do. It's almost mind-numbing," he said. "But when you look at the details, it may not be summarizing things exactly the way you would want.
“We'll put these tools through rigorous testing to ensure whatever we're putting in front of adjusters doesn't diminish their ownership of the claim strategy. There can be a strong temptation to let the technology do the work for you, but if it isn't meeting our standards for accuracy and comprehensiveness, it can negatively impact the claim outcome.”
Notably, Raedeke also revealed that early concerns about generative AI led some clients to prohibit AI use entirely in contracts. The irony, he added, was that insurers had already been using forms of artificial intelligence for decades.
“Early on, about a year and a half to two years ago, one challenge involved the language around AI and what it actually means,” Raedeke said. “A lot of clients, concerned that AI might expose their data to outside parties, quickly inserted language into contracts saying things like, ‘You can't use any AI in our claims process.’ But if you go back to the early 2000s, technologies such as OCR (optical character recognition) were already being used to turn scanned images into text. That's a type of AI.
“There was a disconnect where people would prohibit something in contracts that had effectively been in use for decades. AI became this thing that everyone wanted, but that many people were also afraid of.”
As a result, claims firms often found themselves navigating conflicting expectations: clients wanted innovation but were wary of perceived risks around privacy and data security. Today, Raedeke said those conversations are becoming more sophisticated. "We're not fully free of (the distrust), but clients are becoming more nuanced,” he observed. “The conversation is less about ‘You can't use AI’ and more about data privacy, security and how it's being used.”
Despite growing automation, the CTO does not envision a future where claims professionals disappear.
"The analogy I use is that if we once worked with hand tools, now we have power tools," he said. "If we once had shovels, now we have cranes and bulldozers."
The goal, he explained, is to automate the administrative tasks surrounding key decisions rather than the decisions themselves.
The industry is confronting both efficiency pressures and growing complexity. AI-powered fraud is already emerging as a significant challenge. UK insurer Aviva recently reported detecting a record £233 million in fraudulent claims during 2025, citing increasing use of AI-generated documents and manipulated evidence by fraudsters.
At the same time, insurers are investing heavily in AI capabilities. A recent KPMG survey found that 73% of insurance CEOs now rank AI as their top investment priority.
For Raedeke, however, the industry's future advantage may not come from the technology itself but from the professionals using it. One unexpected lesson from Crawford's AI journey was the effectiveness of AI-generated code when paired with senior engineers.
"We eventually found an effective approach, but it requires very senior software engineers," he said. "They know what to look for and can recognise when something isn't right. Putting AI into the hands of a junior developer isn't something we've found to be a good practice. Putting it into the hands of a very senior engineer can be extremely valuable.”
