A few years back, I was sitting across the desk from an underwriter at a mid-sized commercial insurer. We were going through a renewal for a manufacturing client — a company with about 300 employees, some heavy machinery, a fleet of delivery trucks, and a complicated supply chain that touched three different countries. The underwriter had a thick paper file, a spreadsheet that looked like it hadn’t been redesigned since 2004, and about 45 minutes to make a decision on a policy worth several million dollars in coverage.
He circled a few numbers, scratched his head, asked me if the client had any prior losses, and then made a judgment call based largely on experience and gut instinct.
That scene stuck with me — not because anything went wrong, but because I kept thinking: there has to be a better way to handle this much complexity with this little information.
Turns out, there is. And it’s been quietly reshaping the insurance industry for the past several years.
The Old Way Was Educated Guessing
Here’s the honest truth about traditional commercial risk management that nobody in the industry loves to admit: a lot of it was educated guessing dressed up in actuarial language.
An underwriter would look at a business, review its loss history, check some industry benchmarks, and price a policy based on patterns they’d observed over a career. That experience was genuinely valuable. But it was also limited to what one person (or one team) could hold in their head.
For straightforward risks — a small retail shop, a single-location restaurant — that approach worked fine. But commercial risks today are complicated. A logistics company might have cyber exposure, cargo liability, fleet risk, workers’ comp, and product liability all tangled together. The traditional model wasn’t really built for that kind of multi-dimensional assessment.
AI didn’t just speed things up. It changed what’s actually possible to evaluate.
What AI Actually Does in Commercial Insurance (Not the Hype Version)
I’ve seen a lot of breathless coverage of AI in insurance that makes it sound like insurers have basically become tech companies. That’s overstating it. But there are a few specific areas where AI is genuinely making a meaningful difference right now.
Underwriting — faster and more consistent
The most immediate impact has been in commercial underwriting. Platforms like Cytora, Planck, and Coalition’s underwriting engine pull in data from dozens of external sources — web presence, business filings, safety records, court judgments, news mentions, even a company’s digital footprint — and generate a risk profile in minutes rather than days.
When I watched a demo of Planck’s platform for a commercial auto account, it pulled in the company’s DOT safety rating, their fleet size inferred from public records, their accident history, and even flagged that their primary operating territory had seen a spike in weather-related claims. All of this before the underwriter even opened the submission.
That’s not replacing the underwriter’s judgment. It’s giving them a much fuller picture before they start forming an opinion.
Claims — catching fraud earlier
Commercial claims fraud is a significant problem that most insurers underreport because they don’t want to advertise how much they’re losing. AI fraud detection tools, like those built into platforms such as Shift Technology or FRISS, analyze claim patterns and flag anomalies that human adjusters would almost certainly miss.
One example that comes up frequently in industry conversations: a contractor files multiple small equipment theft claims across different policies over 18 months, all just under the threshold that would trigger a deeper investigation. A human reviewer looking at any single claim would see nothing unusual. An AI system looking across the portfolio sees the pattern immediately.
Pricing — getting more precise about risk
Usage-based and behavior-based pricing has been around in personal auto for a while, but it’s now making real inroads in commercial lines. Telematics data from commercial fleets is being used to price policies based on actual driving behavior rather than just the type of vehicle and general territory.
A construction company with a fleet of 40 trucks that drives mostly short-haul, low-speed routes on well-maintained roads is a genuinely different risk from a company running the same trucks on long-haul highway routes overnight. AI can model that distinction. Old actuarial tables mostly couldn’t.
Where It Gets Complicated — The Lessons Learned
Here’s where I want to be honest, because a lot of coverage on this topic glosses over the friction.
Data quality is still a massive problem. AI models are only as good as the data they’re trained on. Commercial lines data has historically been inconsistently collected, siloed across legacy systems, and often just… wrong. I’ve talked to underwriters who’ve found AI-generated risk profiles flagging completely irrelevant information — a negative news mention that turned out to be about a different company with a similar name, for instance.
The tools have gotten better at disambiguation and source quality, but it’s still a live issue. Any insurer (or broker) implementing AI tools needs to build in human review checkpoints, especially for non-standard or complex risks.
Explainability matters — legally and practically. If an AI model declines to offer coverage or prices a policy significantly higher than a competitor, the insured has a reasonable right to know why. In many jurisdictions, there are legal requirements around this. But many AI models — particularly complex neural network-based systems — struggle to produce explanations that a human can actually understand and communicate.
This isn’t just a regulatory issue. It’s a relationship issue. Commercial insurance clients often have long-standing relationships with their brokers. If a renewal comes back with a 40% rate increase driven by an algorithm that nobody can explain clearly, that’s a trust problem.
The talent gap is real. The insurers that are getting the most out of AI aren’t necessarily the ones with the biggest technology budgets. They’re the ones with people who understand both insurance and data science. That combination is genuinely rare, and building it takes time. I’ve seen well-funded AI projects stall out because the insurer couldn’t bridge the gap between what the data scientists built and what the underwriters actually needed.
A Practical Look at How This Actually Gets Implemented
If you’re working at an insurer, a large brokerage, or a risk management department at a corporation, here’s roughly how AI adoption tends to unfold — based on what I’ve observed across a number of implementations.
Step 1: Start with data infrastructure, not AI tools. The instinct is to buy a flashy AI platform. The smarter move is to spend the first phase cleaning up your data — standardizing formats, filling gaps, building connections between systems that currently don’t talk to each other. AI running on bad data produces confident-sounding wrong answers, which is arguably worse than no AI at all.
Step 2: Pick a high-volume, lower-complexity use case to start. Commercial auto is a popular entry point because it has relatively standardized data, high submission volume, and clear performance metrics. Getting early wins in a bounded area builds both technical capability and organizational trust before tackling more complex lines.
Step 3: Build the feedback loop. AI models improve when they learn from outcomes. This means connecting your underwriting model to actual loss data over time, so the system can see whether the risks it priced favorably actually performed favorably. Most insurers have this data — they just haven’t built the pipeline to close the loop.
Step 4: Invest in change management. The underwriters and adjusters who’ve been doing this for 20 years have real expertise. The goal isn’t to make them feel like they’re being replaced — it’s to show them that AI handles the repetitive, data-intensive parts so they can focus on the genuinely complex cases where their judgment matters most. The implementations that fail are usually the ones that skip this conversation.
Real Use Cases Worth Knowing About
Coalition has built its entire commercial cyber insurance offering around AI. They monitor the actual digital infrastructure of their policyholders in real time — scanning for exposed ports, unpatched software, phishing vulnerabilities — and use that data both to price coverage and to proactively alert clients to risks before a claim happens. It’s a fundamentally different model from traditional cyber insurance, and it’s working.
Zurich Insurance has been using AI to analyze large commercial property risks by processing satellite imagery and publicly available data to assess flood, wildfire, and climate-related exposures. Their underwriters are now walking into large account conversations with risk assessments that would have taken weeks to produce manually.
Munich Re has invested heavily in AI-driven scenario modeling for complex catastrophe risks — helping large corporations understand their aggregate exposure across multiple peril types in ways that traditional modeling couldn’t approach.
Common Mistakes Risk Managers Make When Engaging with AI-Powered Insurers
If you’re on the corporate risk management side of this equation, a few things worth keeping in mind:
Don’t assume that because an insurer uses AI they’re making better decisions. Some are using it well; some are using it poorly. Ask questions about how their models work, what data they’re using, and how they handle edge cases.
Don’t let the speed of AI underwriting rush you into decisions. Faster doesn’t mean more thorough. If an insurer can turn around a complex commercial submission in two hours, that’s impressive — but make sure the coverage terms actually reflect your risk profile, not just a template.
Be thoughtful about what data you share. AI-powered insurers often want more data than traditional ones — telematics, operational data, cybersecurity assessments. That data can help you get better pricing, but it can also be used against you at renewal if your performance data isn’t what they hoped.
Where This Is All Heading
The honest answer is that AI in commercial insurance is still early. The tools have matured significantly in the last three or four years, but the industry is still working through significant challenges around data, regulation, explainability, and talent.
What I’m most interested in watching is the shift from AI as an underwriting tool to AI as a risk prevention tool. The Coalition model — where the insurer is actively monitoring and helping manage the risk they’re covering — points toward something genuinely different from the traditional insurance model. Instead of waiting for a loss to happen and then paying a claim, AI enables insurers to be active partners in preventing losses in the first place.
That’s a better outcome for everyone. And it’s the direction the most thoughtful players in the market are heading.
The underwriter with the paper file and the 2004 spreadsheet — I ran into him recently at an industry event. He’s now working with an AI-assisted underwriting platform. He told me he was skeptical at first, thought it was going to deskill the job. Instead, he said, it got rid of the parts he found tedious and let him spend more time on the cases that actually required thinking.
That, more than any particular technology claim, is probably the most honest summary of what good AI implementation in commercial insurance actually looks like.