The number that changes the conversation
Goldman Sachs published a U.S. Daily note in 2026 ranking occupations by AI substitution risk, a combined score drawing on standard displacement analysis and the International Monetary Fund's (IMF) complementarity index. Insurance claims clerks ranked among the occupations with the highest substitution risk, alongside telephone operators, bill collectors, customer service representatives, and data entry staff.
The Goldman framing is direct: these are roles where "AI can handle most of its core tasks." That is a substitution argument, and it has been made before. What makes the current moment different is not the AI capability; it is what is happening to the supply of people available to do those tasks at all.
The insurance workforce shortage in the markets is not primarily about AI. It is about demographics. As Gabriel De Dominicis, Kapto's MD & Head of AI, has argued: the industry is not choosing between humans and AI. It is facing a workforce that is aging out of the labor pool faster than it can be replaced. AI is the answer to a question that demography is forcing.
The demographic cliff: what the numbers actually say
The European Commission's 2024 Ageing Report sets the baseline. The EU's old-age dependency ratio (the number of people aged 65+ relative to the working-age population) stood at 36% in 2022. By 2050, that reaches 55%. By 2100, 65%. The ratio of workers to elderly people falls from 2.7 today to approximately 1.8 by 2050, and to approximately 1.5 by 2100.
The OECD's Pensions at a Glance 2025 tightens the timeline even more: EU populations will age almost twice as fast over the next 25 years as over the last 25. This isn't a gradual drift. It's an acceleration.
Country-level data reinforces this. Germany's baby-boom cohort enters full retirement between 2025 and 2035. Working-age population is projected to shrink by 4 to 6 million in that window alone. Italy and Spain face among the steepest demographic transitions in the developed world. Swiss Re, citing WHO 2024 data, notes that in Italy, Germany, and France more than 40% of doctors are aged 55 or older, with Japan and Spain above 30%. The same age distribution is echoed in insurance back-office functions.
The manufacturing sector has hit this wall first. German and Italian procurement administrators, quality documentation staff, and accounts payable teams face the same demographic cliff as insurance claims clerks. The insurance industry is watching this play out in adjacent sectors and, in most cases, not yet acting on what it means.
What this means specifically for insurance
The insurance workforce skews older than the broader economy. Industry estimates place the median age at approximately 45, with roughly one quarter of the workforce already 55 or older. That cohort was trained for paper-based processes, manual exception handling, and relationship-based workflow management. Within ten years, a significant fraction of that institutional knowledge retires.
The Jacobson Group and Aon's Q1 2026 Insurance Labor Market Study found that technology roles remain the insurance industry's top talent need overall, while analytics roles have ranked among the hardest positions to fill for five consecutive studies. Carriers are competing for that talent against fintech, consulting, and technology companies that pay more and carry more professional prestige with younger workers.
The structural mismatch is clear: the cohort retiring held paper-era skills, and the cohort that needs to replace them either doesn't exist in sufficient numbers or doesn't find insurance careers appealing enough to enter.
Goldman's own data on Gen Z is instructive here. In AI-exposed occupations, the unemployment rate gap between entry-level and experienced workers has widened significantly. KPMG's 2025 Insurance CEO Outlook puts numbers on the leadership-level response: 77% of insurance CEOs agree that workforce readiness and upskilling is a top growth constraint. 79% say AI is already changing the skills required for entry-level roles.
The US Bureau of Labor Statistics projects a 5.1% decline in insurance claims and policy processing clerk job openings by 2034. That's the demand side softening while the supply side collapses. The two curves are moving in the same direction, but for opposite reasons, and neither is temporary.
Swiss Re's Sonar 2025 report identifies workforce shortages as a structural risk. The EU lists 42 shortage professions; the UK lists 39 shortage occupations. Insurance back-office functions appear in neither list by name, but the skills those functions require map directly onto occupations that do.
This is the bind. You cannot recruit your way out of a demographic shortage. You cannot train your way out of it fast enough. And unlike a cyclical talent crunch, it does not self-correct over time, and the gap between BPO decline and AI readiness is widening.
Steelmanning the counter-arguments
Anyone building an operations case for automation needs to take the objections seriously, not dismiss them. Four deserve direct attention.
Regulatory requirement for human oversight. The EU AI Act and EIOPA's AI Opinion both require meaningful human oversight for high-risk AI applications. This is real and it applies to insurance. The counter is that human oversight is not the same as human execution. A team of ten reviewers overseeing AI-processed outputs at scale is a fundamentally different staffing model than a team of fifty executing those tasks manually, and it's achievable against a shrinking talent pool.
Customer preference for human interaction. In complex claims, particularly personal injury or contested liability, customers want a human involved. This is also real. But it applies to a specific subset of claims, not to the full document intake and processing chain. Most of what a claims clerk does day-to-day is not customer-facing.
Skill evolution, not elimination. The World Economic Forum’s (WEF) Future of Jobs Report 2025 indicates that routine processing roles are declining while oversight and human-AI collaboration roles are growing. The argument is that roles evolve from execution to exception management. This is correct in principle. The challenge is the transition timeline. Retraining a 52-year-old claims specialist to manage AI exception queues is a different proposition than hiring a 24-year-old into that role from the start, and there are not enough 24-year-olds choosing insurance careers to absorb the volume.
Union and regulatory protections slow workforce reduction. True in several European markets. But demographic attrition is not a workforce reduction programme; it is natural retirement. Unions and regulators don't prevent people from retiring. They don't extend careers by a decade. What they do create is a constraint on how fast you can replace exiting workers with automated capacity, which argues for starting earlier, not for not starting.
New roles will emerge: AI trainers, data quality specialists, model oversight functions. That's also true. But it doesn't answer the binding constraint: where do the workers come from when 25% of the current workforce retires within the next ten years?
The demographic argument doesn't require believing AI replaces everyone. It only requires believing that the people who currently do the work won't be there to do it, and that their replacements aren't entering the pipeline in sufficient numbers.
Why "we'll just hire" is not a plan
Understanding what "execution-grade AI" actually means matters here because the framing often gets reversed. The common assumption is that automation is adopted when it becomes compelling enough to displace existing workers. That framing implies choice and timing: that organisations will evaluate automation when ready, on their own schedule.
The demographic data suggests the actual sequence is different. The workers doing document intake, policy processing, and claims administration are leaving the workforce on a schedule that isn't negotiable. The replacement pipeline (younger workers with different skills, different career preferences, and higher wage expectations) is not keeping pace.
The automation question for most European insurers isn't "should we automate?" It's "at what point does operational continuity require it?"
Frequently asked questions
Why does this matter more for European insurers than US insurers?
European demographic projections are steeper. Germany loses 4-6 million working-age people by 2035. Italy and Spain are among the fastest-aging populations in the developed world. The countries with the highest insurance market concentration (Germany, Italy, France, Spain) face the most acute transitions.
How many insurance workers are retiring in the next decade?
Industry estimates place roughly 25% of the current insurance workforce at 55 or older, with a median age of approximately 45. Within ten years, a significant fraction of institutional knowledge retires with that cohort. The replacement pipeline is not keeping pace: 72% of insurers report difficulty finding people with the right technology and data skills (CII 2024).
If AI is replacing entry-level roles, why is youth unemployment rising in those occupations?
Automation is absorbing the tasks that previously provided entry-level work, while experienced workers retain roles requiring judgment and oversight. Goldman’s data shows the unemployment rate gap between entry-level and experienced workers has widened in AI-exposed occupations. This is a transition problem that affects who gets displaced first.
Can AI replace insurance claims clerks?
For structured, repetitive document processing tasks, yes. Goldman Sachs ranked insurance claims clerks among the occupations with the highest AI substitution risk. For judgment-intensive edge cases (disputed liability, multi-party litigation, emotionally charged interactions), human operators remain the right answer. The practical model is not full replacement but a shift from execution to oversight.
Is automation the right response to an insurance labor shortage?
The demographic data leaves no alternative. Hiring cannot solve a structural shrinkage in the labor pool. Managed AI operations, delivered by digital workers (such as what Kapto provides) that absorb the full document processing chain, add operational capacity without requiring headcount that does not exist.
What skills will insurance workers need in the future?
The WEF’s Future of Jobs Report 2025 indicates that routine processing roles are declining while oversight and human-AI collaboration roles are growing. New roles will emerge: AI trainers, data quality specialists, model oversight functions. The challenge is the transition timeline, not the destination.
Is this different from previous technology transitions?
Yes, because previous transitions (internet, mobile) created more roles than they eliminated. The current transition is happening against a backdrop of demographic contraction: there are structurally fewer working-age people entering the pipeline. The technology shift and the demographic shift are moving in the same direction simultaneously.
What is the EU old-age dependency ratio and why does it matter for insurance?
The ratio of people aged 65+ to the working-age population. It stood at 36% in 2022 and is projected to reach 55% by 2050 (European Commission 2024 Ageing Report). For insurance, this means fewer people available to do document processing, claims handling, and back-office work, while the volume of insurance documents generated by an aging population is not declining.
The constraint described here, a structural insurance workforce shortage with no plausible hiring solution, is exactly the operating condition execution-grade AI operations are built for. If your document processing chain depends on a talent pool that is shrinking faster than it can be replaced, that is an operational risk, not an HR problem. Digital workers do not retire, do not require upskilling, and scale with document volume rather than headcount. Kapto has processed over 10,000,000 pages for European insurers, absorbing the full intake and processing chain: per-page pricing, deployed in six weeks, over 95% accuracy with confidence-calibrated uncertainty, without adding to the headcount you do not have. See how it works.
.jpeg)
