The promises arrived first. Underwriting decisions in seconds. Claims settled without human touch. Fraud detected before it happened. For insurance executives who spent the better part of the last three years approving nine-figure technology budgets, the return on those promises has proven maddeningly elusive.
Now a blunt new body of research - and a cautionary tale playing out in real time at one of the world's most prominent technology companies - suggests the industry may have been asking the wrong question entirely.
The question was never whether AI could make people faster. It can, and it does. The question is whether making people faster, in isolation, constitutes transformation at all
In March of this year, Atlassian - the Australian software giant behind workplace tools used by millions of professionals globally - announced it was cutting approximately 1,600 employees, or roughly 10% of its workforce, in order to "self-fund further investment in AI." The move, reported by hrdmag.com and confirmed in a letter from chief executive Mike Cannon-Brookes, was framed as an act of strategic adaptation: reshaping the skill mix to build for an AI-first future.
The timing carried a particular irony. Just weeks later, Atlassian's own Teamwork Lab published its State of Teams 2026 report - a survey of more than 12,000 knowledge workers and 172 Fortune 1,000 executives - that arrived at a finding both counterintuitive and damning. AI, the report concluded, is making individuals faster while quietly making teams more dysfunctional. The researchers gave this dynamic a name: the fragmentation tax.
"AI has made us faster and it has made us more speedy, but it hasn't changed how we work together," Alicia Lenart, Atlassian's vice president of HR, told Human Resources Director magazine. "And I think that's really interesting to HR folks because that's our job = to help people work better together. It's just that hidden cost that we're seeing."
The fragmentation tax, as Atlassian defines it, is the erosion of organisational value that occurs when AI accelerates individual output without redesigning the collaborative workflows that surround it. The report estimates this cost at approximately US$161 billion annually across the Fortune 500 - a figure driven by duplicated work, misaligned priorities, and coordination failures that compound every time a faster individual meets a slower, unchanged system.
For the insurance industry, the implications are not abstract. They are already visible in the data.
Grant Thornton's 2026 AI Impact Survey, which drew on responses from 950 senior business leaders including 100 from the insurance sector, found that 52% of insurance executives are reporting AI-enabled revenue growth, and 62% say the technology is improving their decision-making. Those are not trivial numbers. The investment is producing results - somewhere.
But the same survey found that 44% of insurance executives say governance or compliance challenges have contributed to AI projects failing or underperforming. Only 24% expressed strong confidence that their organisation could pass an independent review of AI governance and controls within 90 days. The evidence of what AI is actually doing, and how it is being governed, is fragmented across teams and tools - sometimes invisibly so.
"Most insurers already have AI governance policies in place, but they haven't built the operational infrastructure to prove and test them," Grant Thornton noted in its findings. Boards have approved the investments. Committees have been convened. Yet the gap between ambition and execution remains wide.
A separate AutoRek 2026 Insurance Report found that 82% of insurers agree AI will define their future - but only 14% have meaningfully integrated it. The BCG analysis cited in Insurance Business found that only 7% of insurance AI initiatives make it beyond the pilot stage.
What is happening in that gap? Atlassian's research offers one of the clearest explanations yet. Most organisations - insurance firms very much included - are pointing AI at the easiest layer: individual productivity. Summarising meeting notes. Automating data entry. Generating policy documents. As Lenart told HRD, "that's the easiest place to point because you can kind of just buy the tools, do a little bit of upskilling and get people using them." But Atlassian's research found that 80% of knowledge work actually happens at the collaborative level - the layer where workflows intersect, priorities are negotiated, and decisions are made. That is precisely the layer that AI has, so far, been least applied to.
In the insurance context, this plays out with particular force. Underwriters, claims adjusters, and risk managers do not work in isolation; their value lies in judgement applied within intricate human systems. Speed up one node without redesigning the network, and you do not necessarily get faster outcomes - you get faster bottlenecks.
The most instructive voices are not the technology vendors. They are the insurance executives who have gone furthest and are now reporting what they have learned.
At AXA XL, artificial intelligence is now handling the administrative elements of underwriting - loss run ingestion, motor vehicle report summaries, submission data entry. Kathleen Ziegler, the company's chief operating officer for the Americas, described both the opportunity and the complication: "AI is taking over the administrative elements of underwriting. This is valuable because it frees people to focus on the insights those documents provide. But it also changes the training ground."
That last phrase is worth pausing on. The tasks AI is absorbing in insurance are not merely tedious; they were historically the training ground for professional expertise. The junior underwriter who spent years ingesting loss runs was developing pattern recognition that no prompt can replicate. Removing those tasks without replacing that developmental pathway creates a capability gap that may not become visible for years.
Troy Dehmann, chief operating officer at Beazley, speaking at a recent Financial Times webinar, put the point directly: "I think one of the bigger transformations will be skills and how we think about skills in our industry, within financial services more broadly." He argued that every insurance role will be affected in some way - not by replacement, but by redefinition.
This view was echoed across the industry in the CEO Voices Report 2026, produced by Sollers Consulting from interviews with senior leaders at carriers across Europe, North America, and Asia-Pacific. The finding, as reported by Insurance Business: executives largely rejected the notion that AI inevitably destroys jobs. Instead, roles are evolving toward more technological, analytical, and strategic responsibilities. Front-line positions are shifting away from repetitive administration toward higher-value case management and advisory work - where human judgement, empathy, and contextual reasoning genuinely matter.
"Several CEOs identify workforce transformation as one of the most pressing challenges ahead," said Marcin Pluta of Sollers Consulting. The challenge is not headcount arithmetic. It is the harder work of redesigning what people do, how they develop, and how they collaborate around the tools that are now doing more of the routine work.
Here Atlassian's research offers the insurance industry something practical, not merely diagnostic.
Lenart is unequivocal on one point: AI mandates do not work. "We don't believe in AI mandates, and you will never see an AI mandate from us," she told HRD. "AI mandates drive fear, and then you also get compliance. People will click into the tool because they're worried." Compliance without genuine adoption produces the worst of all outcomes: the appearance of transformation without the substance of it. Organisations acquire the cost of change without capturing its value.
The alternative Atlassian has pursued is a "learning loop" framework - an approach built on encouraging employees to learn, play, and then share with AI tools, building capability through peer influence rather than executive decree. The company runs hackathons, maintains champion networks, and treats AI fluency as a social practice. "The first time you use AI for a task, it's not always very good," Lenart noted. "You need to sharpen your prompt. Or maybe you need to try a different tool."
Safety National, one of the insurance carriers that has received notice for its thoughtful approach to AI adoption, has arrived at a strikingly similar philosophy. "By educating the average user in this organisation as to what AI is for that user and, more importantly, how they can utilise AI, has been the most important decision we have made," a senior executive told Insurance Business. The company describes its approach as "crawl-walk-run" - a deliberate sequencing designed to build genuine confidence rather than mandated compliance.
The parallel is not coincidental. Both organisations have learned the same lesson that the broader research literature is now confirming: the technology is the easy part. The cultural transformation is the hard part, and it cannot be compelled.
There is a deeper issue that neither hackathons nor learning loops can resolve alone, and Lenart's framing of it carries particular weight for insurance.
Most insurance carriers are currently managing their AI transformation through a familiar split: the chief information officer or chief technology officer drives the technology investment, while the people organisation sits alongside it - informed, perhaps, but rarely integrated. The result, Lenart argues, is that the two essential inputs to transformation - technological capability and cultural change - never actually compound each other.
"If you continue to split tech and have the IT person driving it, and then you have the people thing on the side, you're never going to get the value together," she said. "The tech has gone really far, but the people bit hasn't."
Atlassian's response was structural: the creation of a combined chief people and AI enablement officer role. Whether that specific model is right for every carrier is debatable. What is less debatable is the underlying logic. Organisations that treat AI transformation as a technology project with a people dimension, rather than a people transformation enabled by technology, are structurally likely to produce the fragmentation tax Atlassian's research describes.
The insurance industry is not short of evidence for this. Grant Thornton's survey found that three-quarters of boards have approved major AI investments, yet only 52% have set clear AI governance expectations, and just 54% have integrated AI risk into ongoing board-level oversight. The enthusiasm is there; the organisational integration is not.
Grant Thornton's UK consulting arm made the point with characteristic directness in a recent insight: "If the underlying process is fragmented, manual or unclear, AI simply creates expensive inefficiency at scale." In a sector whose core processes - policy administration, claims handling, underwriting, regulatory compliance - have in many cases accumulated decades of complexity and workarounds, that is a structural warning as much as a tactical one.
Any honest accounting of the insurance industry's AI moment must grapple with a fact that the technology's optimists sometimes obscure: jobs are being affected, and the direction of that effect is not uniformly positive.
A Q1 2026 Insurance Labour Market Study by The Jacobson Group and Aon's Strategy and Technology Group found that job openings in finance and insurance fell to their lowest monthly level in a decade by December 2025, dropping from an annual average of 281,000 openings to approximately 138,000 in a single month. Automation improvements requiring fewer staff were the most common reason cited by companies that reduced headcount. Involuntary turnover rose 0.6 percentage points year-on-year, attributed in part to technology advances.
The roles growing, by contrast, require judgement rather than processing: experienced underwriters, compliance specialists, analytics professionals, and technologists. The jobs shrinking are those in financial reporting, data synthesis, call centres, data entry, and transactional operations - roles that have provided career entry points for generations of insurance professionals.
This is, in miniature, the same dynamic that Atlassian's layoffs made visible at the organisational level: AI does not merely transform work; it restructures who gets to do it, and on what terms. Insurance executives who frame this purely as a productivity story are missing a material risk - not only to their people, but to their regulatory relationships and their reputations with customers who are already watching how the industry uses these tools.
Consumer trust, while improving, remains fragile. According to Insurity's 2026 AI in Insurance Report, cited in Insurance Business, 39% of policyholders now say it is a good idea for their insurer to use AI to improve services - nearly double the figure from 2025. But that still leaves the majority unconvinced. And confidence drops sharply when AI moves from assisting humans to replacing them in decision-making.
None of this argues against AI transformation. The opportunity is genuine and the competitive pressure is already visible. A McKinsey analysis reported by Insurance Business found that early AI leaders in insurance are generating roughly six times the total shareholder returns of their AI-laggard peers. Grant Thornton found that organisations with fully integrated AI are nearly four times more likely to report revenue growth than those still piloting. The gap is not narrowing - it is widening.
The lesson of the Atlassian research is not that AI fails. It is that AI applied narrowly - to individual speed rather than collective effectiveness - produces benefits at the individual level while incurring costs at the organisational level that are easy to miss until they are large.
For insurance leaders, the practical implications of that lesson are threefold.
First, the measurement question deserves serious attention. Most organisations tracking AI return on investment are measuring what is easy: processing speed, cost per transaction, claims cycle time. Fewer are measuring what matters: whether the quality of underwriting judgement has improved, whether claims decisions are more consistently sound, whether cross-functional coordination around complex risks has become more effective. If the fragmentation tax is real - and the evidence suggests it is - then the metrics that matter most are the ones being measured least.
Second, the cultural infrastructure of adoption deserves investment that is at least proportionate to the technology investment. Training without integration, tools without workflow redesign, and mandates without genuine capability-building are the three most reliable paths to expensive disappointment.
Third, the structural question of who owns AI transformation within an insurance organisation is not a human resources question. It is a strategy question. The carriers that are generating disproportionate returns from AI are not doing so because they bought better tools. They are doing so because they redesigned how people work around those tools — and that work requires organisational authority that sits above the IT function and the HR function separately.
Atlassian cut 1,600 jobs to fund its AI future. The insurance industry is watching a version of that same reckoning play out across its own workforce in slower motion. The firms that will navigate it well are not the ones moving fastest. They are the ones moving most thoughtfully - asking not just what AI can do, but what kind of organisation they want to be when it has done it.
Related reading: AI is accelerating in insurance – are you ready? | When the algorithm knocks: how AI is quietly rewriting insurance jobs | AI push puts human judgement back at the centre of insurance | Rewriting insurance transformation by putting human experience at the centre of AI | Widening 'AI proof gap' exposes weak governance behind board-level enthusiasm
