First Principles Daily — Episode Brief

Topic: Car Insurance: Risk Pool Priced Through Agents

Segment 1 — The Cold Open

Americans spend a few hundred billion dollars a year on personal auto insurance, and a typical full-coverage policy now commonly lands well over a thousand dollars — in many states closer to two. The actual expected cost of the crashes, the repairs, and the injury payments that policy is supposed to cover is substantially lower. The gap is not mystery physics. It is a risk pool still priced and sold the way it was when the product lived on paper, in an agent's office, one six-month policy at a time.

Segment 2 — Why It Costs What It Costs Today

The modern personal-auto policy is a descendant of a nineteenth-century agency system, and a large share of it is still sold that way. Independent and captive agents explain coverages, shop a handful of carriers, and take a commission that industry figures commonly put in the low-to-mid teens as a percent of premium on new business, somewhat less on renewals. That arrangement was rational when comparing policies meant reading paper and when a driver had no other way to find a solvent insurer. It is harder to justify once the actuarial work is a database query and the premium is paid by card. Stacked on the agent is a fifty-state regulatory machine. The McCarran-Ferguson Act of 1945 left insurance primarily to the states, so a national carrier does not file one product — it files rates and forms, minimum limits, and uninsured-motorist rules in fifty-plus jurisdictions, each with its own clock and its own objections. That compliance apparatus is not optional, and its cost sits inside every declarations page. Because almost every state makes liability coverage mandatory, insurers do not have to convince you that you need the product; they have to convince you to buy theirs, which is why auto insurance is one of the heaviest advertising categories in the country. The gecko and the spokesperson are not decoration. They are the competitive battlefield, and the bill is paid by people who are legally required to buy. Inside the industry this stack feels like weather. Six-month terms, a long application, a credit-based insurance score, a territorial rate, a commission, a filing — these are how the product has always been made. Expense ratios in the neighborhood of a quarter of premium are treated as the cost of doing business, and combined ratios that hover near break-even in hard years are taken as proof the price is already tight. Nobody wakes up thinking the design is the problem. They wake up thinking about the next rate filing and the next ad flight.

Segment 3 — The Magic Wand Number & The Idiot Index

The magic-wand question for car insurance is not what the paper costs, because there is almost no paper left; the irreducible purchase is a statistical promise that if this car hurts someone or gets hurt, the pool will pay. Wave the wand and you know each driver's true expected loss, you pay every honest claim at the commodity cost of parts, labor, and medical care, and you hold only the surplus a real tail event requires — that combined figure is the floor. Recent consumer surveys have put a typical full-coverage policy well above a thousand dollars a year, and in a long list of states nearer two thousand, while personal-auto premiums across the United States add up to a few hundred billion dollars a year. A typical insured car has only a modest chance of a collision claim in any given year, and when the claim happens the repair bill is usually a few thousand dollars, not the size of the annual premium. Multiply a low frequency by a mid-four-figure severity and the expected physical-damage cost is already only a fraction of what a full-coverage driver pays, before liability is added. Liability is the swing factor — medical bills, lost wages, and pain-and-suffering on at-fault crashes — and it is also the line where process inflates the check farthest from the underlying injury. Even after a realistic expected-liability number, industry-wide figures still leave a large residual that is not the crash. Insurers' published loss ratios commonly live around two-thirds of premium, swinging higher when parts and medical inflation outrun filed rates and lower after a hard market. That published ratio is muddy, because some of what gets labeled "loss" is the cost of handling the claim rather than the indemnity check, and a cautious reading of statutory filings is that pure indemnity is perhaps something like six-tenths of the premium, plus or minus a noticeable amount by year and by line. Capital belongs next to that indemnity, not above it: a pool that must survive a nuclear verdict or a statewide hailstorm carries surplus, and a few cents of each premium dollar is a fair allowance for that carrying cost. Add expected indemnity to that thin capital charge and the magic-wand price lands, roughly, at two-thirds to three-quarters of what drivers currently pay. Divide the finished premium by that floor and you get a homemade Idiot Index of about one and a half to a little over two. This is not a laboratory measurement, and a listener who picks a different loss-ratio year will get a different number, but the size of the answer will not flip. The obvious objection from inside the industry is that combined ratios already sit near one hundred percent in hard years, so there cannot be fat left to squeeze. That objection confuses a full cost-plus stack with a physical floor: if you load commissions, ads, filings, and adjustment into the definition of cost, of course the ratio looks tight, the same way a rocket looks fairly priced once every bracket is accepted as a machined forging. The Idiot Index is the tool that refuses that acceptance; a ratio of two is not a rocket-engine scandal, but a quiet one on a mandatory product attached to more than two hundred million vehicles, and half a premium dollar times a few hundred billion is the prize. Walk the stack instead of the combined ratio and the extra dollar has names. The first pocket is distribution — agent commissions in personal auto are widely reported in the low-to-mid teens as a percent of premium on new business and lower on renewals, the carrier still has its own sales overhead, and although direct writers spent decades proving that removing the agent cuts a real slice of load, they then spent a large share of the savings on television, so customer-acquisition cost put on the commission's clothes. The second pocket is the comparison mess that makes that acquisition expensive: a six-month bundled quote, larded with discounts and filed differently in every state, is not a unit price anyone shops the way they shop gasoline, which is why auto insurance stays among the loudest advertising categories in the country and why a meaningful fraction of a mandatory premium is the cost of being persuaded which logo to buy. The third pocket is the filing machine, because fifty-state rate and form regulation turns every new price, endorsement, and telematics discount into a legal project whose per-policy cost looks small but whose industry-wide cost is a permanent headcount, and whose nastier effect is to freeze the product in the shape of a six-month policy a prior-approval statute already knows how to review. The fourth pocket is the process around the claim: paying the body shop and the clinic is the product, while fighting over betterment, rental days, total-loss valuation, and bodily-injury files with attorneys on both sides is the process, and loss-adjustment expense plus lawyer-driven severity sit on top of indemnity even after you grant that some fraud investigation is genuinely irreducible. The fifth pocket is analogical pricing — territory, credit-based insurance score, marital status, education, prior limits — correlates rather than miles, speed, time of day, or closing distance, producing a pool that overcharges the careful mile and undercharges the sloppy one until the good risks shop and the carrier spends still more to replace them. The sensors that would let you price the mile already live in the car and in the phone, so the expensive part is no longer measurement but permission, regulation, and the will to throw away the old factors — adverse selection is what you get when you price the proxy, not a law of nature. Stack those layers and the Index stops being abstract: the raw expected crash is steel, glass, labor, and medicine, and almost everything above it is a process designed when information was scarce and the agent was the computer.

Segment 4 — The First-Principles Opportunity

A from-scratch redesign would not start by inventing a cleverer discount. It would start by asking what has to exist at all. First move: take the commissioned conversation out of a standardized, mandatory product. Coverage attaches at the vehicle — at registration, at the OEM, at a posted per-mile meter — and the driver is in the pool by default, the way they are in the road system by default. For that to work you need a solvent rated balance sheet or a transparent residual mechanism for the tail, a coverage contract simple enough that it does not require a translator, and a regulator willing to accept a product that is not distributed by an appointed agent. The save is the commission plus a large share of the brand spend, because there is much less left to choose. Second move: throw away the analogical rating plan as the primary price and put the mile and the manner of the mile at the center. The car already knows how far it went, how late, how fast, how hard it stopped; phones know a version of the same thing. What has to be true is default-on data with a privacy bargain people will actually sign, enough participation that the program is not just a magnet for careful drivers, and a filing regime that can accept an algorithm without demanding it be restated as a 1994 territory-and-credit table. The save is the adverse-selection tax and a chunk of the war for "preferred" names. Third move: make physical-damage claims look like a parts-and-labor invoice instead of a negotiation. Direct repair with transparent parts prices, instant total-loss when the numbers say so, rental handled as a known unit cost — that attacks loss-adjustment expense on the metal side, while bodily injury is honestly harder because the severity there is as much civil-justice system as medicine. What remains genuinely hard should be named without romance: insolvency law will still demand surplus, fraud does not vanish because the app is prettier, fifty-state statute is a political fact, and the independent-agent channel is organized enough to defend itself. Credit-based scores and territory are genuinely predictive, so discarding them before the driving signal is universal will raise prices for some sympathetic households and start a political fire, and a residual market for drivers whose physics are expensive will still exist and still need funding. People will resist being watched. The prize is not a world with no insurance companies; it is a pool whose load looks like a thin capital charge and a cheap computer, not like a television budget and a commissioned sales force.

Segment 5 — The Lesson

The load on a car-insurance premium is a map of every job that used to require a person and has not yet been retired. When the thing being sold is an expected value that a processor can compute for free, a commissioned agent between the pool and the driver is a leftover from the years when the computer was a human being with a rate manual. A legal mandate wrapped in an opaque six-month quote will also, reliably, generate an advertising arms race, because the customer cannot shop a unit price and must buy something; the clean response is to hang the price on the unit that creates the risk — the mile, the car, the registration — until there is almost nothing left to brand. Neither of those moves requires a new law of probability; they require treating the agent, the ad, and the analogical score as design choices rather than as the product. The question worth carrying is who actually tries this as a system, not as a discount: an automaker, a state that is tired of filing theater, a direct carrier willing to make per-mile the whole rate plan. The first concrete signal would be simple to read — a quote that begins from the car's own miles rather than a forty-question application, and an expense ratio, claims set aside, that has fallen into the single digits.

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.