W. Robert Berkley Jr. did something on Monday's third-quarter earnings call that most carrier CEOs have carefully avoided: he put numbers on AI.
Insurance intake efficiency has improved "to the tune of 20%-plus," he told analysts. More than half of the company's claims settle for $5,000 or less — which, he said, "would invite the idea that we should be able to have a lower and lower human touch when it comes to this." And the line that matters most: "The next mountain that is before us is not just the benefits of speed, but the insights that we'll be enjoying as it relates to decision-making." (Insurance Journal, Oct 21, 2026)
Why this clears the bar
Three things make this more than an earnings-call sound bite.
First, it reframes how AI gets measured. Most AI business cases live and die on hours saved — the productivity metric that makes CFOs nod and then forget. Berkley is describing a two-mountain arc: mountain one is throughput, mountain two is judgment. Throughput is auditable and boring. Judgment is where combined ratios live. If your AI program is still reporting "time saved per intake," you're climbing the wrong mountain.
Throughput is auditable and boring. Judgment is where combined ratios live.
Second, the quiet operating-model admission underneath. More than half the claims book settling at $5,000 or less is a structural fact about claims frequency curves — and the single best argument for the model I've been writing about: agents pre-qualify, a licensed person confirms. The volume lives at the bottom. The money and the regulation live at the top. Berkley said the quiet part out loud — a shrinking human footprint on small claims is no longer a technology question, it's an organizational courage question.
Third — and this is the one most commentators will miss — Berkley treated AI as an exposure, not just a tool. He told analysts the company has worked "very diligently to make sure the exposure is identified, understood, controlled — and we're getting paid for it," with exclusionary language in policies where appropriate. That is a carrier CEO describing AI liability the way carriers describe cyber or cat risk: identified, priced, and ring-fenced. Morningstar DBRS's Nadja Dreff said the same week that AI is "no longer a source of competitive advantage, but can become a disadvantage for companies that adopt it too late." The message: AI is now simultaneously a cost of staying in business and a line of business to underwrite. Those are two very different governance problems, and most carriers I know are running them through the same committee.
What I'd do from the operator's chair
If I were sitting in the CIO or COO seat Monday morning, three moves:
One: retire "hours saved" as your AI headline metric. It got the budget approved; it won't tell you whether underwriting got better. Replace it with decision-quality measures: referral rates on AI-triaged submissions, loss-ratio movement in AI-assisted segments, rework rates on automated intake. The 20 percent Berkley cited is management's own account — fair enough — but your board deserves the underwriting answer, not the productivity answer.
Two: run the exposure exercise before the regulator asks you to. Every carrier should be able to answer three questions this quarter: where does AI sit in our insured portfolio, are we being paid for it, and where have we chosen to exclude it. Berkley just gave you the script. The carriers that can't answer those questions yet are not behind on technology — they're behind on governance.
Three: draw the line at $5,000 — and staff it. The triage logic is obvious: push the small, frequent, low-severity claims into straight-through processing with machine triage and licensed human oversight, and concentrate your experienced adjusters where judgment compounds. But "lower and lower human touch" is a sentence that works on an earnings call and terrifies a regulator. The operating discipline is the control plane around the automation: which claims are eligible, what the audit trail looks like, who owns the exception. Berkley didn't say eliminate the touch. He said lower it. The difference between those two sentences is the entire governance program.
The real signal
The most important thing isn't any single number. It's that a carrier CEO quantified AI on an earnings call at all — and tied it to operating income discipline, not innovation theater. The quarter's numbers give him cover: $456.5 million in operating income and an 18.8% annualized operating return on equity, per the company's earnings release. AI isn't the story of that quarter. It's infrastructure underneath it.
That's the posture the rest of the industry should copy: AI discussed the way carriers discuss reinsurance — a capability you manage, a cost you measure, an exposure you price. The carriers still presenting AI as a transformation narrative are telling on themselves. The mountain has moved.
— Krishna