Will AI Replace Insurance Underwriters? What the Data Actually Shows

Insurance underwriting is one of those professions that most people can't explain at a dinner party. Which means when i tell you it's being significantly affected by AI, the general public shrugs. But if you're an underwriter, you're not shrugging. You're watching it happen in real time.

I was a data scientist before AI restructured my career. Now i consult on AI strategy and sit in restructuring meetings. Insurance companies come to me quite often, because their entire business model is built on predicting risk. Which is exactly what AI is best at.

Here's what i'm seeing.

What AI can already do in underwriting

Risk assessment. This is the core of underwriting, and AI is already doing a significant chunk of it. Machine learning models analyse historical claims data, applicant information, third-party data sources, and market conditions to produce risk scores that are, in many cases, more accurate than human underwriters.

Personal lines underwriting for standard risks is essentially automated at most major insurers now. Motor, home, travel, pet insurance. If you're a straightforward applicant with no unusual risk factors, an AI makes the decision. No human involved. The policy is quoted, bound, and issued without anyone looking at it.

Commercial lines are following. Small business insurance, standard commercial property, fleet... AI handles the data gathering, risk scoring, pricing, and even generates the quote. More complex commercial risks still need human involvement, but the line is moving.

Claims data analysis and fraud detection. AI spots patterns that humans miss. Suspicious claims, unusual patterns, potential fraud rings. It processes volumes of data that no team of underwriters could manage manually.

Portfolio analysis and pricing optimisation. AI can model the profitability of entire portfolios, suggest pricing adjustments, and identify segments where the company is under or over-charging. What used to require an actuary and an underwriter working together for weeks now takes a model a few hours.

Document intake and processing. Submission data from brokers, financial statements, loss runs, property surveys. AI reads, extracts, and organises this information automatically. The admin side of underwriting is largely automated.

What AI still can't do

This is where it gets more nuanced. And if you're a senior underwriter, more reassuring.

AI cannot underwrite novel risks. When something genuinely new comes along, something there's no historical data for, AI has nothing to learn from. A new type of cyber liability. An emerging technology. A geopolitical risk that's never materialised before. These require human judgement, creativity, and the willingness to make a decision with incomplete information. That's what experienced underwriters do.

Relationship-based underwriting. The broker who calls you because they've got a client with a messy risk and they need someone who'll actually look at it rather than auto-decline. The insured who's been with you for twenty years and is going through a bad patch. The judgement call about whether to renew a policy that the model says to drop but your instincts say is worth keeping. These are human decisions.

Complex and specialty risks. Marine cargo through a conflict zone. A pharmaceutical company's product liability. A fine art collection. Environmental liability for a decommissioned industrial site. These require deep domain expertise, negotiation with brokers, and the ability to structure coverage creatively. AI can provide data. It can't make these calls.

Regulatory interpretation. Insurance regulation changes constantly and varies by jurisdiction. Understanding how new regulations affect underwriting decisions, and explaining those decisions to regulators when challenged, requires legal awareness and professional judgement.

And reading brokers. A good underwriter knows when a broker is telling them everything and when they're hiding something. They know which brokers present risks honestly and which ones... don't. That intelligence comes from years of relationships. AI doesn't have years.

The honest assessment

Insurance underwriting is being significantly restructured. i see it in meeting after meeting.

The pattern is clear: personal lines underwriting has been largely automated. Standard commercial lines are following. The underwriting teams that remain are smaller, more senior, and focused on complex and specialty risks.

One large insurer i worked with reduced their underwriting team from forty-five to twenty over two years. The twenty who remained are experienced specialty underwriters. The twenty-five who left were primarily processing standard risks that AI now handles. They didn't all lose their jobs, some moved into other roles, but the underwriting positions themselves ceased to exist.

Lloyd's market is an interesting case. The specialty and complex nature of Lloyd's business means human underwriters are still essential. But even there, AI is handling more of the data analysis, pricing, and administrative work. Underwriters are spending less time on spreadsheets and more time on decisions. Which is, arguably, what they should have been doing all along.

The Chartered Insurance Institute qualifications still matter. ACII is a signal of the kind of professional judgement that AI can't replicate. If you don't have it and you're in underwriting, get it. It's a differentiator that's becoming more important, not less.

The entry-level pipeline is shrinking. Junior underwriter roles, where you learn the trade by processing straightforward risks, are disappearing. Which creates the same problem as everywhere else: how do you develop senior underwriters if there are no junior roles to learn in?

What to do this week

1. Identify which of your decisions AI could make. Be honest. If a model could reach the same conclusion you did on a particular risk, that's a decision that's going to be automated. Focus your energy on the ones it can't.

2. Deepen your specialty knowledge. Pick the most complex type of risk you handle and become the expert. Read the technical literature. Talk to loss adjusters about claims experience. Visit the types of businesses you insure. The deeper your expertise, the safer your position.

3. Strengthen your broker relationships. Call a broker you haven't spoken to in a while. Have a conversation about their book. The underwriters who survive are the ones brokers specifically want to deal with. If you're interchangeable with an AI portal, you're at risk. Financial analysts face similar pressures around relationship versus technical value.

4. Learn to use the AI tools your company is deploying. Not as a threat, but as an augmentation. If you can use AI to process the data faster and spend more time on the judgement calls, you're more productive. More productive underwriters keep their jobs.

5. Start thinking about adjacent roles. Risk management, loss prevention, claims strategy, product development. Underwriting expertise is valuable beyond the underwriting desk. Having options is always sensible.

If the restructuring anxiety is real, AI replacement dysfunction is probably what you're experiencing. And recognising the signs that restructuring is coming at your company is better than being caught off guard.

The one thing to do today: look at the last complex risk you underwrote and ask yourself what made your decision different from what a model would have produced. If you can articulate that difference, that's your value. If you can't... it's time to find risks where you can.