Every Kinsale Employee Has an AI License. The Moat Was Built 15 Years Before.

THE CRAFT

There is a stretch of the excess-and-surplus insurance market — the carriers who write the policies your standard insurance company refuses to write — where the average commercial customer is paying around fourteen thousand dollars a year for coverage on a risk that the rest of the market said no to. A small contractor in a high-litigation state. A specialty manufacturer with one fire-loss in its history. A franchise operator three counties away from a hurricane line. These are the policies almost nobody wants. They are written, when they are written at all, by a small group of carriers willing to underwrite risks individually rather than by class.

Kinsale Capital Group writes a lot of those policies. About a million of them a year, give or take. In 2025, Kinsale wrote roughly $1.6 billion in gross premiums on policies the rest of the insurance market mostly didn’t want — and in Q1 2026, the company reported a combined ratio of 77.4 percent.

If you have not spent time with insurance financials, that number deserves a sentence of context. A combined ratio is the sum of an insurer’s losses and operating expenses, divided by the premiums it earned. At 100, the insurer breaks even on underwriting. The industry average runs in the high 90s, sometimes north of 100. At 77.4, Kinsale is keeping more than twenty-two cents of every premium dollar after paying claims and running the business. The S&P insurance index at the same moment was hovering around 96. Kinsale’s number is not in the same conversation as its competitors. It is in a different sport.

The popular explanation for that number, the one Kinsale tells in its own investor calls and the one that is now starting to show up in trade press, is that Kinsale has built an AI-saturated underwriting machine. Every Kinsale employee — every underwriter, every claims handler, every actuary, every developer — has an enterprise AI license that the company funded twelve months ago. Dozens of AI bots and agents run daily inside the underwriting and claims workflows. The company’s proprietary core system — the homegrown enterprise software platform Kinsale has been building since the day Michael Kehoe founded the company in 2009 — now has AI agents embedded inside it. Most of the analytics and IT work has been industrialized: bots writing and testing code, bots converting unstructured policy data into structured records, bots flagging high-severity claims early, bots running risk segmentation against the historical book. The implication is that the technology produced the 77.4.

I think that explanation is half right, and the other half is the more important half.

The longer I sat with Kinsale’s filings, earnings calls, and the bear case Edwin Dorsey wrote against the company at The Bear Cave — and the longer I looked at the leadership reshuffle Kehoe announced on April 29 of this year, in which he collapsed the Chief Information Officer role and the Chief Actuary role into a single new C-suite seat called the Chief Analytics and Technology Officer — the more I came to think Kinsale’s real moat is not the AI deployment. It is the underwriting philosophy Kehoe set in 2009, before the company had a single line of code, much less a single bot. The custom enterprise core was built to enforce that philosophy at the volume it required. The AI agents were built to enforce it faster. The April 29 restructure was an organizational acknowledgment that, inside Kinsale, the actuary and the technologist are now the same job.

The story most operators reading this on a Sunday night want to hear is “Kinsale used AI to crush the combined ratio” — meaning the company’s losses and operating expenses are far below the premiums coming in, so every dollar of policy written is more profitable than the rest of the industry’s. The story Kinsale is actually telling, when you read it carefully, is “Kinsale crushes the combined ratio because the underwriting philosophy is contrarian — and the AI is the enforcement mechanism for a philosophy that almost no other carrier has the discipline to run.” That is a much harder story to copy. It is also the only Brief I think is worth writing about this company.

This is the AI deployment story I think a small-cap or mid-cap leader needs to read this week. Not because you are going to write specialty surplus-lines insurance. But because Kehoe’s order of operations — philosophy first, custom system second, AI third — is the inverse of how almost every operator I talk to is funding their AI program right now. And the inversion is why most of those programs are not going to produce a Kinsale-shaped result.

THE OPERATOR

The situation

Kinsale Capital Group writes excess and surplus lines insurance — the part of the U.S. commercial insurance market where standard carriers say no and a smaller group of specialty carriers say maybe. The surplus-lines market exists because some risks are too unusual, too concentrated, or too small to fit the rate-and-form filings of admitted insurance. A contractor doing a particular kind of demolition work in a particular state. A small medical practice with a litigation history. A trucking firm with five units and a customer base in two counties.

The carriers who write those policies are not bound by state rate filings. They are free to price risk individually, exclude what they choose to exclude, and decline what they choose to decline. The classic problem with the surplus-lines market is that the freedom to price individually is also the freedom to mis-price individually — and a single bad season, or a single underwriter with a soft touch, can turn a profitable book into a loss-ratio nightmare. The standard surplus-lines carrier is fighting that risk by hiring more underwriters, writing more policies, and hoping the law of large numbers smooths the variance. Most of them run combined ratios in the mid-90s.

Michael Kehoe founded Kinsale in 2009 inside a marketplace that had spent the prior decade consolidating. The big specialty carriers were getting bigger; the smaller carriers were either getting bought or being squeezed out of the most profitable corners. The conventional industry view at that moment was that the specialty surplus-lines market was a scale game — that you needed a broad national footprint, a deep distribution channel, and a large book of business to compete.

Kehoe took a different view, on the record from the beginning. Instead of competing on scale, Kinsale would compete on selectivity — the willingness to write smaller policies, smaller customers, smaller risks, but with tighter exclusions and more rigorous individual pricing than any competitor was willing to do at that policy size. The average Kinsale policy today writes for roughly $14,000 to $15,000 in premium. That is small. By design. A standard specialty carrier would not bother to underwrite a $14,000 policy individually; the underwriting cost would exceed the margin. Kinsale’s bet, from the founding, was that they would underwrite those policies individually anyway — and the way they would do it economically was to build the technology to make individual underwriting cost less than the competitors’ batch underwriting cost.

That is the founding sentence of Kinsale, and the one that I want every Brief reader running a selective business — one that deliberately turns away customers, deals, or projects that don’t fit a tight definition (premium pet care, specialty medical clinics, niche industrial services, custom manufacturing, anywhere you survive by saying no to most of the market) — to read twice. Kehoe did not start with technology. He started with a position on what kind of policies the company would write, who the customers would be, what the exclusions would look like, and what the underwriting standard would be. The technology came after, and the technology’s job was always to enforce the position at every individual policy.

The first thing Kehoe did after raising capital was not buy off-the-shelf insurance software. The off-the-shelf insurance core systems — Guidewire, Duck Creek, the big enterprise platforms most carriers run on — are built for the standard market. They support standard product filings, standard policy administration, standard claims workflows. They are not built for an underwriter who needs to write a policy that nobody else has written, with exclusions nobody else has written, at a price point nobody else has bothered to underwrite individually. So Kehoe built the core system himself, in-house. Kinsale Capital has been writing its own underwriting platform, its own policy administration system, its own claims management workflow, and its own analytics engine for sixteen years. The company refers to it now as its proprietary core. It is the single biggest piece of capital expenditure Kinsale has ever made, and it is the thing that allowed the selective philosophy to operate at the policy economics the philosophy required.

That was the situation when AI showed up at Kinsale’s door in 2023.

The move

Kehoe’s response to the wave of generative-AI vendor pitches that hit the insurance industry in 2023 was the opposite of what almost every other insurance CEO did. Most of his peer-set ran the playbook everyone in this newsletter has now seen play out: hire a Chief AI Officer (or rebrand an existing CIO to that title), build an AI Center of Excellence, run four use cases in parallel (a claims chatbot, an underwriting copilot, a marketing personalization engine, a developer productivity tool), spend twelve to twenty-four million dollars, and produce two abandoned pilots and a slide deck.

Kehoe did three things in a specific order instead.

First, he put an enterprise AI license on every employee’s desk. Not a curated rollout. Not a “pilot group.” Not “key roles in claims and underwriting.” Every single Kinsale employee. The reasoning that comes through Kehoe’s earnings calls is that he did not yet know which roles inside Kinsale were going to find the most yield in AI — and his bet was that the underwriters, claims handlers, actuaries, and engineers inside his building were better positioned to discover that than any consultant he could pay to tell him. The license was a research tool first and a productivity tool second. The company spent the first six months tracking which functions used it, what they used it for, and what kind of work-product changes were starting to show up. Every Brief reader running a selective business should hold that decision in mind: the question of which AI use case to fund first was answered by watching employees use the tool, not by deciding from the executive suite.

Second, he made the AI deployment a job for the custom core system, not a standalone initiative. This is the move that most insurance carriers cannot copy — because they do not have a custom core. Kinsale spent the first year of AI deployment writing agents and bots that lived inside the proprietary core, not in some separate AI platform. The agents had access to Kinsale’s underwriting history, claims history, policy exclusion library, and risk segmentation models because the core was the source of all of it. The agents could be wired into the existing underwriting workflow because the workflow was already inside the core. A carrier running on Guidewire or Duck Creek would have had to negotiate API access, build middleware, run integration testing, and probably hire a systems integrator. Kinsale’s engineers wrote the agents directly into the system they already maintained. The 2009 decision to build the custom core was the decision that made the 2024 AI deployment cheap.

Third, he reorganized the leadership to match the operating reality. On April 29 of this year, Kinsale announced that Diane Schnupp, the company’s longtime Executive Vice President and Chief Information Officer, would retire and stay on as a consultant during the transition. The role Kinsale created to replace hers was not another CIO. It was a new C-suite seat called Executive Vice President, Chief Analytics and Technology Officer. Salmaan K. Allibhai — Kinsale’s Chief Actuary since March 2025, with nine years inside the company — was promoted into that seat. The actuary and the technologist became the same job.

That is a sentence I want you to read twice as well, because it is the kind of organizational decision almost nobody else in the insurance industry has made yet, and the implication for the rest of the market is not subtle. Kehoe is saying, in the most public way an organizational chart can say anything, that he does not believe analytics and technology are separate disciplines anymore. The work of pricing risk, segmenting customers, identifying claim severity, and writing the software that performs all three is now the same work — and the person responsible for it is one person, not two. Most carriers still have a Chief Actuary who reports through a financial line and a Chief Information Officer who reports through a technology line. Those two people meet on Wednesdays. Kinsale just abolished the meeting.

The fourth thing Kehoe did, which is harder to point at because it has not been announced as a single initiative, is to keep the AI deployment relentlessly focused on the selective philosophy. The bots are doing the work the philosophy requires. They are converting unstructured claim documents into structured records so the underwriting team can see patterns at the individual-policy level instead of the class level. They are writing test code so the engineering team can ship core-system updates faster. They are flagging high-severity claims early so the claims handlers can intervene before litigation escalates. They are running pricing accuracy checks across the book in near-real-time. None of these are growth-side AI deployments. None of them are about remaking the customer experience. Every one of them is about pricing risk more accurately and managing claim losses more aggressively — which is what the 2009 philosophy was about in the first place.

The result

The Q1 2026 financial result is what Wall Street is now reacting to, so let me say it again clearly. Kinsale reported Q1 2026 net income of $112.6 million, up from $89.2 million in the year-ago quarter — a 26.2 percent year-over-year increase. Diluted operating earnings per share rose 26 percent to $5.11. Net earned premiums grew 9.3 percent. The combined ratio was 77.4 percent. For the full year 2025, Kinsale had already posted an operating return on equity of 26 percent — roughly double the long-term industry average for property and casualty carriers.

The bears’ reading of that result, and the reason this Brief needs to engage Edwin Dorsey’s critique honestly, is that Kinsale’s numbers are not actually a function of underwriting discipline or AI deployment — they are a function of an underwriting strategy that loads policies with exclusions, targets small businesses that may not fully understand what they are buying, and denies claims aggressively when they are filed. The Bear Cave essay points to a 60 percent customer retention rate (versus an industry standard around 90 percent), an unusual volume of regulatory complaints, and a litigation pattern around claim denials. The implication is that Kinsale’s combined ratio is artificially low because the company is shedding customers who have bad claims experiences and underpaying the customers who stay.

I do not think the bear case kills the AI story. I think it sharpens it. Here is why. A selective carrier — one that prices risk individually, writes policies with tight exclusions, and is willing to deny claims that fall outside coverage — will always look bad on retention and complaint metrics relative to a standard-market carrier. That is not a Kinsale-specific scandal. That is the unavoidable byproduct of running the business the way Kehoe set it up to run in 2009. The question for an investor — or for a Brief reader trying to learn from this — is not whether Kinsale’s retention is below the industry average. It is whether Kinsale’s combined ratio holds because of the underwriting philosophy, or despite it.

The AI deployment, the way I read it, is what makes the underwriting philosophy hold at the volume Kinsale is now running. Without the AI, an underwriter writing a small specialty policy at $14,000 in premium has to make individual pricing and exclusion decisions at a cadence that no human underwriter can sustain without quality degradation. The risk-segmentation accuracy slips. The exclusion library gets stale because nobody has time to update it. Claim handlers cannot triage severity early enough. Code in the core system ships with bugs because the engineering team is overloaded. The AI is the only way to run selective underwriting at the volume Kinsale is now running without the philosophy collapsing under its own weight.

That is the load-bearing point of the result, and the only reason I think the Bear Cave bear case strengthens, rather than weakens, the AI story. If Kinsale were running a soft-touch underwriting strategy and trying to use AI to look disciplined, the Bear Cave critique would land hard. But Kinsale’s underwriting has been disciplined since 2009. The AI is the enforcement mechanism that lets the discipline scale. Without it, the 77.4 percent combined ratio would not hold at $1.6 billion in premium.

The Craft of AI read

Here is what I think is going on, and why I think it matters for an operator running a small-cap or mid-cap business that is not in insurance.

The dominant AI deployment story in mid-market financial services right now is the one I described earlier — a Chief AI Officer, four use cases, two abandoned pilots, twelve to twenty-four million dollars. That story is failing because it gets the order of operations wrong. It starts at the technology layer and tries to retrofit an operating philosophy around it. The philosophy ends up being whatever the four use cases happen to suggest — which is usually some combination of “improve productivity,” “automate customer service,” and “use data to make better decisions.” Those are not philosophies. They are the absence of a philosophy.

Kehoe is doing the opposite. The Kinsale operating philosophy is one sentence long: We will write the policies the rest of the market won’t, and we will price the risk individually with rigorous exclusion discipline at small premium sizes. Every operational decision since 2009 has been a derivative of that sentence. The custom core system was built to make individual underwriting at $14,000-per-policy economically viable. The 2024 AI license rollout was built to scale individual underwriting at that volume without quality degradation. The April 29 restructure was built to put the actuary and the technologist in the same chair because the philosophy requires them to be the same person. Each move follows from the philosophy. The philosophy did not follow from the moves.

Here is the load-bearing sentence of this issue, and the only one I want you to memorize: Kinsale’s AI deployment looks like discipline because the underwriting philosophy was already disciplined. Most other carriers’ AI deployments will look like exactly the philosophy they have. If your philosophy is the absence of one — if you are trying to be all things to all customers, write every line of business, hire every kind of underwriter — the AI will scale the absence. Discipline does not come from the tool. It comes from the position the operator took before the tool arrived.

There is a second-order point that is worth lifting out for any operator running a selective business — premium specialty manufacturing, niche professional services, contrarian financial services, specialty industrial distribution, anywhere that the business model is to do something the bulk of the market refuses to do. The custom-built system is the moat, not the AI on top of it. Kinsale spent fifteen years building a core platform that almost no other insurance carrier was willing to build, because most CEOs in the industry could not justify the capital expenditure when off-the-shelf systems were available. That fifteen-year build is now what makes Kinsale’s AI deployment cheap and the competition’s expensive. A standard specialty carrier looking at Kinsale’s 77.4 combined ratio and trying to copy it with Guidewire-plus-Microsoft-Copilot is going to discover that the AI deployment costs five to ten times more than Kinsale’s did, and the resulting discipline is two to three notches less rigorous.

The same dynamic applies in other industries. The reason WD-40’s planning AI (covered in #009 of this newsletter) produced 100 basis points of gross margin expansion in one year was that Steve Brass deployed AI inside an existing operating philosophy — the customer position (the can on the shelf) and the employee position (planners doing exception work, not manual reforecasting). The reason Lindsay Corporation’s Smart Pivot platform (covered in #008) produced ARR growth in ag-tech was that Randy Wood deployed AI into an existing distribution philosophy of dealer-network proximity. In every Brief operator story I have profiled, the AI deployment was wrapped around a pre-existing operating philosophy. Kinsale is the cleanest version of that pattern I have seen yet, because the philosophy is the only thing the company sells.

The thing I would have done differently if I were Kehoe is this: I would have separated, in disclosure, the cost of the enterprise AI license rollout from the cost of the AI agents embedded in the core system. Right now both are bundled inside Kinsale’s general technology spend, and the analyst community has no way to model the productivity yield of the license deployment versus the productivity yield of the core-embedded agents. Those are two different deployments with two different return profiles. Splitting them out would help Wall Street price the work more accurately and would make the Kinsale story more durable in the long run. That is a cosmetic disclosure complaint against a real piece of operating work. But cosmetics and disclosure shape multiples — and Kinsale, at the current multiple, is trading at a premium that the bears think is unjustified. Cleaner disclosure would settle the debate one way or the other.

Things to consider

• The philosophy precedes the technology. Always. Kehoe set the underwriting philosophy in 2009 and built the technology around it for sixteen years. Every Brief reader funding an AI program right now should be able to write their operating philosophy on a single index card before they take a vendor call. If you cannot — if your philosophy is some combination of “grow the top line,” “serve our customers better,” and “be more efficient” — the AI will scale that absence rather than fix it. The carriers losing to Kinsale right now are carriers whose underwriting philosophy is “be like the rest of the market, only slightly cheaper.” AI is making that strategy fail faster, not slower.

• A custom system is the moat. The AI runs on top of it. Kinsale’s 2009-2024 capital expenditure on the proprietary core is what makes the 2024 AI deployment defensible. A small or mid-cap operator looking at the Kinsale story should not conclude “I need AI.” They should conclude “I need to identify the one or two operational systems in my business that the standard off-the-shelf vendors cannot serve well, and I need to build or commission those custom — because that is where the AI will land cheaply when it gets here.” For a specialty distributor, that might be the inventory-and-quoting system. For a specialty manufacturer, the production-scheduling system. For a niche professional services firm, the engagement-pricing system. Identify the system the standard vendors mis-serve. Build the custom version. The AI investment is downstream of that build, not upstream of it.

• Put an AI license on every employee’s desk before you decide which use case to fund. Kehoe did this on purpose. The license is a research tool first. The point is to watch which functions inside the company find the most yield, what work-product changes show up, and which roles begin to industrialize their own work. The wrong move is to pick four use cases from the executive suite and fund them. The right move is to give every employee the tool and let the yield discover itself. Eighteen months later, the patterns of use are the basis for which workflows to industrialize, which agents to write into the core system, and which roles to redesign. The information cost of this discovery is the cost of the license. The information cost of getting the use case wrong is twelve to twenty-four million dollars.

• Collapse the analytics and technology leadership into one seat. This is the April 29 move, and it is the organizational decision that most insurance carriers — and most operators in other industries — are going to find it hardest to copy. The actuary and the technologist are the same job now. The risk analyst and the data engineer are the same job now. The pricing strategist and the software developer are the same job now. If you have two C-suite seats running those functions inside your business, you have two people who meet on Wednesdays to argue about whose model is right. Kehoe abolished the meeting. The implication for a small or mid-cap operator is not necessarily a single C-suite seat — most operators at that size do not have a CIO and a Chief Actuary to begin with. But the organizational principle holds. The analytics function and the technology function are no longer two functions. Whoever runs them should be one person, reporting directly to the operating CEO.

• The bears sharpen the story. They do not kill it. A selective business will always look worse on retention, customer satisfaction, and complaint metrics than a standard-market business. That is the price of running the contrarian strategy. The question is whether the operating margin and the loss ratio justify the cost. For Kinsale, the answer right now is yes — 77.4 percent combined ratio and 26 percent operating ROE are not numbers you produce by accident. But the Bear Cave critique is also not wrong about what those numbers cost in customer trust. An operator running a similar selective strategy needs to know, going in, that the customer-satisfaction and retention metrics will be permanently below the standard-market peer set. That is not a bug. It is the strategic price of the position. If you cannot defend it in front of your board, your investors, and your regulators, you should not run the strategy.

THE WORKBENCH

Do this on Tuesday

Block sixty minutes on your calendar. Take three sheets of paper. (Yes, paper. The point is not to draft a memo. The point is to find the position.)

On sheet one, write the underwriting equivalent for your business. Every business has one — the moment where you accept or reject a deal, a customer, a project, an account. For a specialty manufacturer it is the moment a new specification crosses the engineering team’s desk. For a niche professional-services firm it is the moment a prospect calls asking for a proposal. For a specialty distributor it is the moment a new SKU is considered for the warehouse. For a regional service business it is the moment a new geography is considered for expansion. Whatever your underwriting equivalent is, write down the philosophy you use to make those decisions today. Not the procedure. The philosophy. The sentence that, if you wrote it down and gave it to every operator on your team, would let them make the same decisions you would make.

If you cannot finish that sentence in five minutes, you have found the work. The reason your AI investments are not yielding what Kinsale’s are yielding is that the underwriting philosophy was never explicit — and the AI is scaling the absence of one rather than the presence of one.

On sheet two, write the custom system equivalent for your business. Identify the one or two operational systems where the off-the-shelf vendors do not serve your business well. The system the rest of the market cannot copy because they would not justify the capital expenditure. For Kinsale it is the proprietary core. For WD-40 it is the planning layer that sits between ToolsGroup ML and Dynamics 365. For Lindsay it is the FieldNET subscription stack. Be specific. What is the system? What does it do that no off-the-shelf vendor can do as well? What would it cost to build or commission? What does the build buy you that buying the vendor product would not?

If you do not have a system on that sheet, you may not have a defensible AI position at all. The AI investment is downstream of a custom system that enforces the philosophy. If both the philosophy and the system are absent, the AI is going to look like everyone else’s AI — and the financial result is going to look like everyone else’s financial result.

On sheet three, write the research deployment. Take your most current AI license budget and divide it across every employee in the company. Stop curating which roles get access. Stop trying to predict where the yield will be. Give every employee the tool and instrument the use patterns. Write down what you would watch for, what you would measure, what work-product changes you would expect to see, what role redesigns you would consider funding if the patterns suggested them. The Tuesday version of this is the cheapest research the operator can run on their own business. The patterns of license use over the first six months tell you which workflows belong in the custom system, which roles should be industrialized, and where the next dollar of AI investment should go.

What you will learn: the strategic work is on sheets one and two. Sheet three is the cheap research. Most operators are spending the strategic effort on sheet three (picking the right use cases from above) and the research budget on sheets one and two (hiring consultants to write a philosophy and an architecture). The order is wrong.

What it costs you if you do not do this: the same thing it costs the standard specialty carrier trying to copy Kinsale right now. Five to ten times more in technology spend, and a combined ratio that lands somewhere in the high 90s instead of the high 70s. The technology is not the gap. The philosophy and the system are the gap, and the technology you bought is enforcing whichever one of them you happen to have.

The rigorous version

The Tuesday exercise gets you three sheets of paper and a sharper understanding of where the gaps are. It is the shape of the work, not the work.

The rigorous version is The Ground-Up Workshop. It is how you and a small group of your subject-matter experts get on the same page — for two days in person — about what AI is actually for inside your business. Not a strategy deck. Not a vendor selection. A worked-through underwriting-equivalent philosophy for your business, a worked-through custom-system inventory, and the operating-model sketch that makes both possible.

Every AI strategy you’ve been sold starts with vendors. This one starts with the small group of people in your business who already know where the operational truth lives. Two days in person. A target operating model that has AI built in from the ground up. A 90-day starting plan. The alignment step that gets everyone on the same page about what AI can actually do for your business — before the deeper work begins.

What you walk out with: a position on your business’s selective strategy you can defend in writing, an inventory of the one or two custom systems your business needs to build or commission, a target operating model with AI embedded at the work level (not at the org chart), and a 90-day starting plan that aligns the team on what to do first. The deeper operational work — the build, the procurement, the rollout — comes after, if and when you’re ready. The workshop is how we start.

Price: $10,000. I run a small number of these each quarter. What you walk out with is the document Kehoe would have wanted before he founded Kinsale in 2009 — the underwriting philosophy the AI is built to enforce, the custom system the AI is built to run inside, and the operating-model sketch that makes the technology procurement obvious.

THE QUESTION

What is the underwriting equivalent inside your business? What is the philosophy you use to decide which deals, which customers, which projects, which accounts you take — and which ones you turn away?

Most CEOs cannot answer that question, because they have not been forced to. The standard-market carrier writes the policies that fit the rate filings. The standard-market operator takes the customers that the marketing team brought in. The philosophy is implicit, and implicit philosophies do not scale. AI is making implicit philosophies fail faster, because the bots will scale whatever they find — and what they usually find inside a typical mid-cap operator is the absence of an explicit position rather than the presence of one.

Michael Kehoe took an explicit position in 2009. He has spent sixteen years building the systems to enforce it. The combined ratio is the receipt — and the bears who think the receipt is artificially flattering are reading the operating model correctly. They are just disagreeing about whether the strategic price of running the position is worth the financial result. That is a fair debate. It is also the debate every operator running a selective business will eventually need to have with their board.

You either know your position, or you don’t. If you don’t, the bots will run whatever you do have — and what most operators have is some combination of last quarter’s growth targets, this quarter’s customer-satisfaction goal, and next quarter’s board presentation. That is not a philosophy. It is a calendar.

If you want to talk through what the underwriting-equivalent looks like for your company — the deals you refuse to take, and the work you refuse to keep imposing on the team that’s supposed to be making judgment calls instead of running mechanical workflows — hit reply, or send a note to grant@thecraftofai.com. I run a small number of these conversations each quarter, and I start each one by asking you the same question I just asked above. If you can answer it cleanly, we can probably do real work together. If you can’t, the workshop is designed to get you to the answer.

— Grant grant@thecraftofai.com