The retirement forecast the industry has quoted for years named 2026, and it is 2026. The harder problem is that the junior roles which used to replace those people have gone, and the industry may be blaming the wrong cause.
A specialty underwriter retires on a Friday after twenty-eight years. There is a cake, and somebody makes a speech about how she could smell a bad submission through a fax machine.
Four months later a broker sends in a risk that looks a great deal like one she declined in 2019.
It gets written. Not because anyone overrode her, but because the reasoning was never written down anywhere. It lived in her head, and in the habit of the two people who sat near her, and one of those has since moved to a competitor.
This is the part of the talent problem that does not show up in a headcount report. The industry talks about a shortage of people. What it is actually running out of is transferred judgement.
For most of the last five years the trade press has repeated a Bureau of Labor Statistics projection: the US insurance workforce would shed close to 400,000 people to retirement by the end of 2026. It became the standard opening line for conference panels on talent.
It is now 2026.
Worth saying plainly that the primary citation for that number is hard to pin down, and it has been passed between articles for years without one. Treat it as the industry's shared assumption rather than a fresh measurement. The demographics underneath it are easier to check and point the same way: roughly one in four underwriters is over 50, the average insurance professional is in their mid-fifties, and under 25% of the workforce is below 35.
Jullie Hands, a partner at EY, described the gap as “both urgent and potentially destabilising - a demographic cliff that could leave insurers without the institutional knowledge”.
The operational cost is already measurable. Convr's Insurance Talent and Technology Trends survey found 82% of insurance leaders saying staffing limits were hurting growth, expense ratio or both, with underwriting the single hardest area to staff with quality candidates. More pointed: 72% said understaffing had already put inaccurate information into their quotes.
That last figure is not a hiring problem. It is a pricing problem wearing a hiring problem's clothes.
Every previous generation of underwriters was made the same way. You came in to do administrative work and junior analysis, you sat near people who had been doing it for decades, and over several years you absorbed why the answer was no.
That rung is being removed.
Research by Peter John Lambert of Warwick and the London School of Economics, and Yannick Schindler of the Ellison Institute of Technology at Oxford, tracked 243 million hiring records and 407 million job postings across four countries from 2017 to 2025. By 2025 junior hiring sat 8 to 11 percentage points below its 2019 level.
Refilling the bottom of the ladder was never going to be easy either. A Cake & Arrow study of Gen Z attitudes found 79% had never considered a career in insurance and 49% had no interest in one, with 67% calling it boring and only 16% interested in working for a large organisation. The awkward detail is that 55% still viewed the industry positively. They do not dislike insurance. They have simply never thought of it as somewhere to work.
Ask why junior hiring collapsed and the industry has a ready answer, increasingly said out loud: automation took those jobs. Job openings at a decade low, AI doing the document handling and first-pass analysis that used to occupy a graduate's first two years. The graduate hiring model that built insurance is visibly breaking down, and AI is the obvious suspect standing next to the body.
The Lambert and Schindler data does not support the charge.
When the two forces are separated in the same dataset, their finding is blunt: “when WFH exposure and AI exposure are properly disentangled, the WFH effect is robust and the AI effect collapses”. The decline tracks the adoption of hybrid work, not the deployment of AI.
Two things corroborate it. The first is the occupational pattern. If automation were eating this work, you would expect demand for judgement to soften along with it. Instead demand for experienced underwriters, compliance specialists and analytics people is growing while entry-level postings disappear. That is not the signature of a machine doing the job. It is the signature of firms wanting people who already know how, and no longer running the process that produces them.
The second is mundane and probably decisive: more than half of hybrid workers say mentoring is better done face to face. The apprenticeship was never a training programme. It was proximity. Remove the proximity and you have removed the mechanism, without anyone deciding to.
None of which makes AI innocent of everything. It makes it the wrong defendant for this particular charge.
Here is where it gets uncomfortable. APQC surveyed a thousand professionals for its Great Retirement study, sponsored by eGain, and found 93% of insurance CxOs genuinely worried about knowledge walking out of the door.
The same 93% are not consistently capturing that knowledge before it goes.
41% said they rarely or never even try to collect know-how from retiring employees. And 83% of insurance organisations still rely on manual, person-to-person methods to transfer it.
Read that last number against the WFH finding and the two lock together unpleasantly. The dominant knowledge transfer method in insurance is people sitting near people. The best available evidence says the reason the junior pipeline broke is that people stopped sitting near people. The industry has kept the mechanism and removed its precondition.
Diagnosis matters here in a way that is not academic. If you believe automation ate the junior rung, then more automation is a reasonable response: the work is going away anyway, so lean into it. If the mentoring channel is what broke, more automation removes what little of the rung is left and accelerates the thing you are worried about.
Neither reading gets you a replacement for a twenty-eight-year underwriter. What is available is narrower and duller: stop letting judgement live only in people's heads.
A declination reason typed into a note field is a memory. The same reasoning expressed as a referral trigger, a rating factor or an underwriting rule is a thing a new hire can read, question and be taught from, and it does not resign. That is the honest case for treating configuration as institutional memory rather than as plumbing. When appetite lives in rate tables and rules a business user can open, and referrals arrive in an underwriting queue with the evidence attached, the answer to “why was this declined” survives the person who decided it. An audit trail is a poor substitute for a mentor and an excellent substitute for a memory.
Two cautions, though, given what this piece has just argued.
There is also a reason none of this is being funded generously right now. Rates have been falling for eight consecutive quarters in the soft half of a two-speed market, which is not a climate in which anyone approves a graduate intake. The pressure to run the same book with fewer people is exactly the pressure that removes the apprenticeship, and it is strongest at the moment the demographics can least afford it.
Some of this is not a technology question at all. Australian insurers ranked talent their third-biggest challenge for 2026, and 68% said they were increasing investment in training and development, up sharply from 48% a year earlier. That is a local figure rather than a global one, but it is at least a response aimed at the actual mechanism.
The rest is a question about where reasoning lives. Every firm has some number of people whose judgement is undocumented and irreplaceable, and a date in the future when each of them leaves.
Those dates are knowable. It is worth writing them down, and then asking what would have to be captured, in a form somebody else can read, before each one arrives. The rules you can express today are the ones you will still have on the Monday.
In the same quarter, US property rates fell 13% and US casualty rates rose 7%. Calling that a soft market is costing people money. Here is what the two halves are actually doing, and why the winning response is the same in both.
Insurers spent 2025 and 2026 writing AI out of standard policies and building a new market to replace it. Here is what Armilla, Munich Re, AIUC, Klaimee, HSB and Testudo actually sell, and what it takes to launch a product like theirs.
A small commercial MGA integrates a new distribution partner in three weeks. Fortune 100 carriers have been working on the same problem for two years. Here is why the gap exists, what a working insurance API actually has to do, and why the bar is about to rise again.
Book a live, personalized demo with our product team - tell us your use case and see the platform work with your data. No commitment.