Who Actually Wins the Insurance Market
Insurance changes slowly, and then all at once. Study the history of this industry and the pattern repeats for three centuries. Marine underwriters scratching lines in a London coffee house gave way to standardized forms and admitted markets. Judgment rating gave way to actuarial tables, then to catastrophe models, and each transition looked impossible right up until it was mandatory. Every technology wave since the mainframe has promised to disintermediate distribution and mostly ended up feeding it instead. The insurtech era ran that same cycle in fast forward: a direct-to-consumer wave that burned billions of venture dollars learning that customer acquisition cost is an underwriting problem, a quieter MGA and embedded wave that actually worked, and now an AI wave arriving with the same confident pitch decks as the last one. I have spent over a decade working inside this industry through those cycles, on the brokerage side and the technology side, and the reason I keep going back to the history is that it remains the most reliable guide to what happens next.
And what happens next is genuinely unusual, because for the first time in that long record, three structural shifts are hitting the market simultaneously. Risk is migrating out of the admitted market into specialty channels at a pace the standard market has not clawed back. Ownership of distribution is concentrating into fewer hands than at any point in modern history. And AI is resetting the cost of producing insurance work, which is quietly repricing every book of business, every agency, and every underwriting operation in the country.
Most commentary treats these as three separate stories. They are one story. The premium is moving to wherever expertise lives, the expertise is retiring, the buyers are consolidating what remains, and the technology determines who can actually service what they bought. Follow those threads to where they intersect and you get a fairly specific answer to the question of who wins the insurance market over the next decade. It is not the answer most of the market is currently underwriting.
Let me walk through each layer with the numbers, because the numbers are where the argument lives.
Part One: Where the Risk Is Going
The E&S Decade Is Not Ending. It Is Maturing.
Start with the flow of premium itself. Per WSIA’s 2025 annual stamping office report, surplus lines premium across the 15 stamping office states reached $90.3 billion for calendar year 2025, a 7.8 percent increase over the $83.8 billion reported through 2024. Those 15 states represent roughly 63 percent of all US surplus lines premium, so the true national figure is well north of that. And here is the number that matters more than the premium: item counts rose 14.1 percent. Transactions grew nearly twice as fast as premium.
Read that gap carefully, because it is the entire E&S story in one line. When items grow faster than dollars, rate is softening while submission flow keeps accelerating. Risks are still pouring out of the admitted market, they are just pouring in at lower price points. AM Best confirmed the same read in November 2025 when it revised its outlook for the US E&S segment from positive to stable, citing moderating premium growth and early signs of rate softening. The wholesale channel is not shrinking. It is transitioning from a market where anyone could grow on rate to a market where you grow on volume, speed, and placement expertise.
The geography tells the same story. Texas overtook Florida in 2025 to become the second largest stamping office state, with premium up 9.8 percent and commercial property and liability accounting for over three quarters of it. The E&S market has stopped being a coastal catastrophe phenomenon and become the permanent home for complex commercial risk nationally. Once a risk learns to live on nonadmitted paper, with its freedom of rate and form, it rarely goes home. Anyone modeling a great reversion of E&S business back to the standard market is modeling something the data has never shown.
The MGA Channel Stopped Being a Workaround and Became Infrastructure
The second migration is even more consequential. According to AM Best’s June 2026 market segment report, drawing on NAIC data, direct premiums written through delegated underwriting authority enterprises reached $108.7 billion in 2025, up from $92.3 billion in 2024, the fifth consecutive year of growth. Conning’s estimate runs even higher, at $114.1 billion for 2024, growth of 16 percent that outpaced the broader P&C market, with 91 percent of surveyed insurers reporting increased use of MGA partnerships.
Think about what that number means structurally. Well over $100 billion of premium, more than a tenth of the US P&C market, is now underwritten by entities that do not hold the risk. Carriers are renting underwriting expertise they no longer maintain in-house. And the delegation is deepening, not just widening: AM Best found underwriting authority granted in more than 75 percent of MGA contracts in 2025, with a growing share of MGAs also empowered to handle claims.
But the same report carries a warning that most of the MGA world has not fully absorbed. Capacity is becoming selective. AM Best revised its delegated authority outlook from positive to stable, and the fronting carriers and reinsurers behind the channel are now underwriting their MGA partners on loss ratio stability and operational quality, not premium growth. For a decade, an MGA could raise capacity on a niche and a narrative. In the market forming now, capacity follows demonstrable underwriting discipline and clean data. That distinction will decide which MGAs survive the soft market, and it is worth holding onto, because it comes back at the end of this piece.
Part Two: Who Owns Distribution
The Consolidation Machine Never Stopped Running
Now layer the ownership question on top of the premium flows. Per MarshBerry, there were 649 announced US insurance brokerage transactions through November 2025, tracking ahead of 2024, a year that closed at 847 deals. And MarshBerry estimates only 50 to 60 percent of transactions are ever made public, so the true count is materially higher. Private capital backed buyers accounted for 72.6 percent of announced activity, up from 59.3 percent in 2019. Just three buyers, BroadStreet Partners, World Insurance, and Hub, accounted for one in every five deals.
The top of the market is consolidating just as aggressively as the bottom. Brown & Brown agreed to acquire Accession Risk Management Group, the parent of Risk Strategies and One80 Intermediaries, for roughly $9.8 billion, a deal that folded one of the largest remaining independent specialty platforms into a public broker. The pattern of 2024 and 2025 was large buyers acquiring other large buyers: Aon closing NFP, Marsh McLennan taking McGriff, Blackstone taking a stake in Higginbotham. The inventory of mid-sized independent firms is not merely being bought. It is running out.
The Retirement Cliff Supplies the Sellers and Removes the Product
Feeding that machine is the demographic event everyone cites and almost no one prices correctly. The Bureau of Labor Statistics found the number of insurance professionals aged 55 and older grew 74 percent over a decade, and NAMIC, in the US Chamber of Commerce’s America Works Report, projected that half the current insurance workforce retires within 15 years, leaving more than 400,000 positions unfilled. Roughly a quarter of the industry is already 55 or older, against fewer than a quarter under 35.
The conventional read is that this is a seller supply story, and it is. Most independent agencies have no perpetuation plan that survives contact with a twelve times EBITDA offer. But the cliff has a second edge that the roll-up models consistently underweight. The people retiring are not overhead. They are the product. In insurance distribution, the client relationship and the placement expertise are the revenue, and both walk out the door with the producer. Every acquisition underwritten on a retiring principal’s book is an acquisition underwritten on an asset that is actively depreciating on the day the LOI is signed. The consolidators that win are the ones that learned, expensively, to structure earnouts, producer equity, and knowledge transfer around that fact. Which brings us to the new buyer who has not learned it yet.
The New Buyer at the Door
Nikola Lazarov, the founder of Eilla AI, recently published a piece asking who wins the $1 trillion professional services market, and it deserves the attention it is getting. His map: the largest firms hold the ceiling by quietly becoming the biggest AI-native service providers on earth, while the middle of every professional services market stops being a contender and becomes inventory. The buyer he describes is genuinely new: permanent capital paired with an embedded AI lab, acquiring firm after firm and rebuilding each around AI built in-house. His three legs are that the retirement cliff supplies the sellers, revenue renews by law, and the production data finally exists. And his closing call, the one he asked readers to hold him to, is that within twelve months this template gets pointed at legal and insurance brokerage.
The template is not hypothetical. In June 2025, Reuters reported that Crete Professionals Alliance, backed by Thrive Capital alongside ZBS Partners and Bessemer Venture Partners, planned to deploy more than $500 million over two years acquiring US accounting firms and equipping them with OpenAI-powered tooling built by Thrive’s in-house tech team. Founded in 2023, Crete had already passed $300 million in annual revenue across more than 20 firms. In accounting, the logic mostly holds: audits and tax filings are statutory, the work recurs whether clients love you or not, and AI collapses the cost of producing it.
Lazarov is right that this buyer is coming for insurance distribution, and probably on roughly his timeline. Where I part ways with the thesis is on what that buyer finds when it arrives.
Why Insurance Brokerage Is Not Accounting
First, the inventory problem. Lazarov’s framework assumes a fragmented middle waiting for a disciplined buyer. Insurance distribution had that moment, and it was fifteen years ago. An AI-native entrant buying agencies today is not picking up unwanted inventory at a discount. It is walking into the most competitive auction in professional services, bidding against acquirers who have closed hundreds of deals each, price integration synergies into every letter of intent, and pay with equity that sellers actually want. The venture roll-ups reached accounting before consolidation matured. In brokerage, they arrive several thousand transactions late.
Second, the revenue problem. Revenue renews by law is the strongest leg of the thesis in audit and the weakest in distribution. Policies renew annually, but nothing about the renewal belongs to the broker. A broker of record letter can move an account in the time it takes a client to sign one. Retention is earned every cycle through remarketing, claims advocacy, and service, and commission income rides the rate environment, which means in a softening market the same book pays less every year at even perfect retention. Remember the WSIA gap: items up 14.1 percent, premium up 7.8 percent. That spread flows directly through to commission income across the specialty market.
The preview of what softening does to a leveraged distribution thesis is already running in the UK. MarshBerry counted just 99 UK insurance distribution deals in 2025, the quietest year since 2017, with more than half involving targets of fewer than ten employees, and pointed directly at softening rates removing the organic growth tailwind while PE owners struggle to exit at prior multiples. The US public markets are pricing the same concern: MarshBerry’s Broker Composite Index fell 21 percent from its March 2025 peak even as the S&P 500 rose 12.3 percent over the comparable window. The market has already begun repricing the assumption that distribution revenue compounds on autopilot. Anyone raising permanent capital on that assumption today is buying the top of a belief cycle, not the bottom of an opportunity.
Part Three: Who Runs the Machinery
The Production Data Finally Exists, and That Changes Everything
Here is where Lazarov is not just right but more right than most of the insurance industry realizes. The genuinely new condition in this market is that the production data finally exists in a usable form. Submissions, loss runs, statements of values, quotes, binders, endorsements, subjectivities, bordereaux: the raw material of this industry has been sitting in agency management systems, carrier portals, and email inboxes for decades, unstructured and effectively dead. Modern document AI can now read it. That single fact changes the unit economics of servicing a book, quoting a submission, and auditing a delegated authority relationship more than anything since the agency management system itself.
Every layer of the market described above runs on exactly this kind of paper. The E&S transaction growth that outpaced premium by nearly two to one is millions of additional submissions, affidavits, and filings flowing through wholesale desks. The $108.7 billion MGA channel runs on bordereaux and delegated authority reporting that capacity providers now scrutinize line by line. The consolidators own thousands of acquired books whose data sits in dozens of incompatible systems. Whoever can actually read, structure, and act on that paper at production scale holds an advantage that compounds across every one of these trends simultaneously.
The 95 Percent Problem
Which makes the next number the most important one in this entire piece. MIT’s Project NANDA, in its widely discussed State of AI in Business 2025 report, found that roughly 95 percent of enterprise generative AI pilots delivered no measurable P&L impact, despite an estimated $30 to $40 billion in enterprise spending. The failure was not model quality. The researchers pointed to flawed integration and what they called a learning gap: tools that do not adapt to workflows, deployed by organizations that do not adapt to the tools.
Buried in that same research is the finding that should reorganize every insurance executive’s technology strategy. In MIT’s interview sample, externally partnered, customized tools reached successful deployment about 67 percent of the time. Internally built tools reached deployment about 33 percent of the time. The build side of build versus buy did not just lose. It lost by half.
I watch this play out in real time across the industry. The demo is always spectacular. Getting AI to read an ACORD 125 in a proof of concept takes an afternoon. Keeping it accurate in production across fifty carrier portals, appetite guides that shift quarterly, E&S paper that follows no standard whatsoever, and forms that change every renewal season is a different discipline entirely. It is not a build project with an end date. It is a maintenance function with a permanent budget line. Most in-house builds die in year two, when the novelty budget runs out, the forms change again, and the engineers who built the pipeline have moved on to something more interesting than endorsement processing. The carriers with the deepest actuarial and data science benches are not exempt from this, because pricing talent and production LLM operations are categorically different skillsets. Being excellent at one tells you nothing about your capacity for the other.
The Regulator Is Already in the Room
And this maintenance burden is no longer optional, because the regulatory apparatus around insurance AI is now real and actively developing. Since the NAIC adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers in December 2023, 24 states and the District of Columbia have adopted it, per Mayer Brown’s summary of the NAIC’s Spring 2026 national meeting, with four additional states adopting their own insurance-specific AI regulation or guidance. The bulletin expects a written AI systems program with board-level accountability, model validation, bias testing, and, critically, oversight of third-party AI where the insurer remains fully responsible for the vendor’s behavior.
This is moving from principle to examination. The NAIC is running a pilot of its AI System Evaluation Tool from March through September 2026, explicitly designed to help examiners assess how companies govern AI in production. Regulators have signaled they will look through vendor relationships during examinations. In plain terms: the era when an insurance organization could run an ungoverned AI experiment on live submissions is closing, state by state. AI governance in this industry now requires the same institutional maturity as solvency monitoring or cybersecurity. That raises the fixed cost of doing AI properly, and fixed costs favor whoever can amortize them across the most volume. Which is precisely why maintenance, not modeling, is where the moat forms.
So Who Actually Wins
Pull the three layers together and the answer stops being abstract.
In distribution, the middle does not go to a de novo AI roll-up. Lazarov’s template arrives on schedule, and it will do real deals in the sub $2 million revenue tier where the big consolidators do not reach and licensing plus carrier appointments are the only real moat. But the bulk of the middle goes to the incumbent consolidators who already own the inventory, already solved producer retention the expensive way, and can bolt the embedded AI lab onto permanent capital they already control. The interesting race is which of them treats AI as an operating discipline with its own headcount and budget rather than a diligence checkbox and a press release.
In the specialty market, the winners are the wholesalers and MGAs that treat the WSIA transaction data as the assignment it is: volume is growing twice as fast as premium, which means margin survives only where the cost per transaction falls. The MGAs that pair genuine underwriting discipline with production-grade document AI will be the ones capacity selects in the market AM Best is describing, because clean bordereaux and defensible data are becoming the price of admission to capacity itself.
On the carrier side, the winners are the organizations honest enough to apply MIT’s finding to themselves. A 33 percent deployment rate on internal builds is not a challenge to be managed. It is a verdict on a strategy. The carriers that treat production AI the way they treat reinsurance, as a specialized capability you source from parties whose entire existence depends on being good at it, will spend less, deploy faster, and pass their NAIC examinations with documentation their in-house peers cannot produce.
And two predictions to hold me to, in Lazarov’s spirit. First, within twenty-four months, at least one top ten brokerage consolidator stands up a dedicated AI operations unit with its own P&L and begins pricing acquisitions partly on the quality and volume of unstructured production data that comes with the book. When data room checklists start asking for loss run archives and submission histories alongside the commission statements, the shift has happened. Second, within the same window, at least one major capacity provider makes production-grade AI governance an explicit condition of a delegated authority agreement. The moment underwriting capacity starts underwriting your operations stack, the technology conversation stops being a technology conversation.
The Takeaway
Three migrations are converging on one industry at once. Premium is moving to specialty channels where expertise clears the market, and the transaction data shows that shift accelerating even as rate softens. Distribution ownership is concentrating into the hands of buyers who have been running the consolidation playbook for fifteen years, just as a new AI-native buyer template, proven in accounting, arrives at the door. And the production data that has been buried in this industry’s paper for decades is finally readable by machines, at the exact moment regulators are building the examination apparatus to govern how it gets used.
The winners of the next decade will not be determined by who bought the most agencies, raised the most permanent capital, or built the flashiest demo. Using AI is not a value proposition. Ninety-five percent of the market has now proven that at a cost of $30 to $40 billion. The winners will be the firms, at every layer of this market, that treat AI as an operating discipline: staffed permanently, governed to examination standard, and maintained through every carrier appetite shift, form revision, and rate cycle that this industry will absolutely continue to produce. In insurance, the disruption was never the technology. It is the unglamorous, compounding work of keeping it accurate. That has always been how this industry separates its winners. It is not going to stop now.
Fabio Faschi is an Enterprise AI Solutions and Sales leader helping carriers, MGAs, and brokerages put artificial intelligence to work across underwriting, claims, and distribution. A National Producer, Board Member of the Young Risk Professionals New York City chapter, and Committee Chair at RISE, he brings over a decade of insurance industry experience and has built and scaled more than a dozen national brokerages and SaaS-driven insurance platforms. He is the founder of ScholarusAI.com and Hogglet.com for Enterprise AI transformation and risk management. Fabio’s expertise has been featured in publications like Forbes, Consumer Affairs, Realtor.com, Apartment Therapy, SFGATE, Bankrate and Lifehacker.
