There's a new report doing the rounds. It's got graphs, economists and a great deal of hand-wringing about artificial intelligence.
The insurance industry, this new report says, has a talent crisis. AI might save it. Or might not. The authors aren't entirely sure. Which is, at least, honest.
Let's start with what we actually know, because the report - produced by a small army of academics at Yale, Stanford, the University of Pennsylvania and the Federal Reserve Bank of Chicago - does contain some genuine numbers worth paying attention to, even if it buries them under enough statistical caveats to wallpaper the Gherkin.
The London Market employs around 61,000 people. It needs 82,200 by 2034. Graduate job postings in insurance fell 18% last year. The share of the market's workforce aged under 30 is projected to collapse from 24% today to just 7% over the next decade. And only 4% of young people in Britain consider a career in insurance to be appealing.
Meanwhile, a survey by MS Amlin found that 68% of Britons would not consider working in the industry under any circumstances. Which raises an obvious question: what exactly do the other 32% think insurance involves? Presumably they haven't looked into it yet.
Hundreds of economists, AI professionals and so-called superforecasters - people who are professionally good at predicting things, which is a real job, apparently - were asked what artificial intelligence will do to the economy over the next few decades.
The answer, after several hundred pages, is: quite a lot, probably, but slower than you think, and nobody is entirely sure.
The median economist in the survey expects annual GDP growth of 2.5% by 2030. That is marginally above current baselines. Not exactly the robot apocalypse. The researchers assigned only a 14% probability to the scenario in which AI systems broadly surpass human performance on most cognitive tasks within the next five years. Even the scenario they call "rapid progress" - in which AI can outperform every software engineer, paralegal and customer service agent - only gets a 14% chance of happening by 2030.
So the machines are coming. They're just stuck in traffic.
The economists did identify which jobs are most at risk. At the bottom of the list - the roles expected to shrink fastest - were general clerks, customer service representatives and data processing staff. At the top, holding on with the greatest resilience, were roles requiring human judgment, personal relationships, physical presence and contextual reasoning.
Which, if you think about it for more than thirty seconds, is a reasonably accurate description of what a good insurance broker actually does.
The researchers found something that should concentrate the mind of anyone who has spent more than a week in broking. The roles most exposed to automation are precisely the entry-level and administrative positions that have traditionally served as the training ground for experienced professionals. The people who process the policies today become the underwriters and account executives of tomorrow. Automate away the bottom rungs of the ladder and you don't just save some money on headcount - you quietly remove the mechanism by which the next generation learns the trade.
Swiss Re, in a separate piece of research, made exactly this point. If entry-level claims and underwriting positions are automated to cut costs, the result further down the line is a shortage of people with sufficient job-specific experience - which increases errors and, ultimately, raises costs for insurers. So the short-term fix creates a long-term problem that is considerably more expensive than the one it solved.
This is the sort of thing that sounds blindingly obvious once someone has written it down, and yet here we are.
This is the part where the report becomes quietly less impressive. Because for all the talk of AI systems that can negotiate book contracts, write Pulitzer-calibre novels and collapse years of research into days, the economists do not actually forecast anything that looks like the wholesale replacement of experienced broking professionals.
Even in the most dramatic scenario, white-collar employment is expected to stagnate rather than collapse as a share of the overall workforce. The jobs at the top of the survival rankings - human judgment, relationship management, contextual expertise, being physically in the room when something important is happening - are the jobs that good brokers do. The AI is going to get very good at finding information, processing data and handling routine correspondence. It is considerably less likely to persuade a nervous client that their coverage is adequate, read a room during a difficult renewal, or know from twenty years of experience that a particular risk has a history that doesn't show up in the submission.
Read next: Are your staff any ‘good’ at AI?
The report also found something genuinely interesting about why experts disagree on AI's economic impact. It isn't, contrary to popular assumption, because they disagree about how fast the technology will develop. They mostly agree on that. What they disagree about is what happens next - how quickly organisations actually change their workflows, how institutions respond, how long it takes for capability to become productivity. Every transformative technology in history has had this gap between what it can do and what actually changes. The personal computer existed for fifteen years before it showed up meaningfully in productivity statistics. The internet took longer still.
Which means the honest answer to whether AI can do your job is: some of it, eventually, and probably not as much as the people selling AI software would like you to believe.
Here is the thing, though. The talent crisis is real, whatever AI does or does not turn out to be capable of.
The London Market Group's figures are not speculative. There are as many people over 50 as under 30 in the London market right now. The over-50s and under-30s each represent about a quarter of the workforce. That is not a pipeline. That is a blocked drain.
Graduate job postings across the sector fell 18% last year, at the precise moment the market needs to be hiring more people, not fewer. Only 904 graduates and apprentices were hired across the entire London market in 2023, and 43% of those went to just three large brokers. If you work for anyone other than a very large broker, the chances are you hired essentially nobody from the graduate market last year.
The Insurance Institute of Canada - and this is relevant because Canada faces an identical problem - found that 8.5% of the entire insurance workforce will retire within the next five years. The London market has no reason to think it is in better shape. Ten per cent (10%) of London Market property risk engineers are due to retire in the next couple of years alone. These are not people whose knowledge lives in a folder on a server somewhere. It lives in their heads, built over decades, and it leaves with them.
Meanwhile, 50% of the current workforce globally is expected to have retired within fifteen years. AI or no AI, that is an enormous amount of institutional knowledge heading for the golf course.
To give the academics their due - and they have worked very hard and produced something genuinely useful, even if it does occasionally read like it was written by someone who has never actually had to renew a fleet policy at 11pm - the core finding is correct.
The disagreement about AI's economic impact is not primarily about whether the technology will arrive. It is about whether organisations, institutions and regulatory systems can absorb it quickly enough to matter. And the answer, historically, is: slower than you think.
That is actually quite reassuring if you are a broker. It means you probably have more time than the headlines suggest. It does not mean you have forever. And it certainly does not mean the 18% drop in graduate postings is fine, or that the demographic cliff is going to sort itself out, or that the perception problem - only 4% of young people finding insurance appealing - is someone else's responsibility.
The report concludes, with rather more words than strictly necessary, that artificial intelligence will change the economics of insurance work, that the change will be slower than the technology's boosters predict, and that the roles most likely to survive are those requiring judgment, relationships and expertise accumulated over years of experience.
Which is, essentially, a description of broking done well.
So no, a robot probably cannot do your job. Not the important bits, anyway, and not yet. But the industry it operates in is quietly running out of the people who know how to train the next generation, and the ones coming up behind are not arriving in sufficient numbers. That is the crisis. AI is a subplot.
The clock, to borrow a phrase, is running. And unlike most of the machinery in this story, that particular mechanism requires no software update. Unless it’s actually your phone. Which is probably listening to you.
Open the pod bay door HAL.
