On May 20, 2026, Acrisure - the world's eighth-largest insurance brokerage by revenue, valued at approximately $32 billion and headquartered in Grand Rapids, Michigan - announced it was cutting 2,250 jobs, approximately 11% of its global workforce, citing advances in artificial intelligence and automation.
In a letter to employees, Acrisure co-founder and CEO Greg Williams was direct about the driver: "Advances in technology, AI, and digital platforms are fundamentally changing how businesses operate, how clients expect to be served, and how value is created." The cuts will roll out in phases through 2027, with the impact concentrated on US operations. It was Acrisure's second AI-attributed workforce reduction in seven months - in October 2025, the company had already cut 400 accounting and back-office roles, citing automation of accounting functions.
Jerry Theodorou, policy director for insurance, finance and trade at the R Street Institute, frames what Acrisure has made public in terms of the financial logic driving it across the industry. "Insurers are looking for areas of the underwriting process to make savings," he said. "To the extent that they do, margins are increased. And for publicly owned companies, the stock goes up. So that's where they're focusing now."
The functions being automated first are not underwriting. They are the back office: claims processing, statutory data production for state regulators, accounting, and reporting.
Theodorou identified statutory NAIC filings - the yellow books - as a specific target. These are detailed premium and loss data submissions in standardized formats, compiled on a mandated schedule. The work is laborious, rule-based, and high-volume. "A lot of the processes there are very time consuming," Theodorou said. "Producing the blanks, you know, the yellow books, statutory data for the NAIC - so there could be an impact on automating a lot of the administrative work that's done."
Claims processing follows the same pattern. AI tools triage incoming claims, route straightforward cases through automated settlement workflows, and flag those requiring human adjudication. Theodorou notes that straight-through processing rates - the share of claims settled without any human touch - have risen as these tools matured. Williams' letter to Acrisure employees named this dynamic explicitly: "We have seen client-oriented work that took days or weeks reduced to minutes."
Chubb, Travelers, Hartford, AIG, and CNA have all disclosed AI adoption strategies in recent quarterly earnings transcripts and annual 10-K filings. Theodorou notes this disclosure is not incidental. "They need to disclose what they're doing to the investors. Because there is - it goes back to governance issues - you've got a board of directors," he said. The disclosure exists because AI investment strategy is material information for shareholders, and because the companies are reporting efficiency gains to analysts.
The 10-K and earnings call disclosures from large publicly traded carriers are the most accessible window into where the industry is actually deploying AI. Theodorou recommends them as a primary research source: "If you want to dig into what the large companies are doing, there's a lot more disclosure there. Just look at the latest transcript - the quarterly earnings reports."
Approximately 4,000 property and casualty carriers operate in the US market, the vast majority of them far smaller than Acrisure or Chubb. Their posture on AI adoption is materially different. "The smaller companies are sitting on the fence," Theodorou said. "They're slower to move because this is a new technology and they don't have the comfort level. They haven't tested it a lot."
The governance exposure shapes that caution. "There's a duty of care" for corporate directors and officers, Theodorou said - and a claims automation failure at a small mutual carrier carries proportionally larger consequences than at a large diversified insurer with a deep legal team. The posture for many smaller carriers is deliberate fast-following: waiting for larger players to validate the technology before committing resources.
"Insurers recognize that this is still a very young tool. It's not fully mature. You've got to monitor it - you can't let it run loose by itself."
The workforce reduction story has a technical limit that Theodorou illustrated with a specific example. He asked an AI system to identify US states with the lowest natural catastrophe exposure. Wisconsin and Michigan appeared near the top of the output. Both states have experienced severe hail, tornadoes, and winter storms significant enough to impair or eliminate regional reinsurers - Wisconsin Re among them.
"Here AI was wrong," Theodorou said. "You can't trust it completely. You have a phenomenon of AI hallucinations which comes up with something which is completely out of order. So you've got to monitor it. You can't abdicate your responsibility as an underwriter to do risk selection and risk pricing."
That monitoring requirement does not eliminate the workforce impact. It shapes it. The carriers reducing headcount are replacing process workers with supervisory roles - a smaller number of people reviewing AI decisions on hard cases rather than a larger number handling routine volume. The net is still a reduction. But it is a reduction toward judgment, not away from it.
Acrisure's May 2026 announcement is a data point, not an outlier. The insurtech conference circuit offers context: Theodorou described events in Las Vegas, Philadelphia, and New York where hundreds of AI-driven insurance startups pitch carriers and investors. "A year or two afterwards, you'll see that a lot of them are not there," he said. "Traditionally only a small%age of the tools that are presented actually work."
The applications that do survive - fraud detection, claims triage, document processing, regulatory reporting - are delivering measurable returns for the carriers that deployed them carefully. For HR and operations leaders at insurance organizations, the Acrisure announcement is a benchmark, not a ceiling. The question is which functions to automate, in what sequence, and with what governance framework - before the gap between early movers and fast followers becomes structural.
