Who Makes the Shortlist: The New Distribution Penalty

Jul 22, 2026

A lighthouse beam illuminates select ships in a harbor at night, representing how AI shapes which insurance carriers get considered

Distribution Economics — Part 3

By James W. Moore

Key Takeaways

  • Distribution has always run on an informal algorithm: the producer’s accumulated feel for which carriers to trust, built from years of anecdote and correction. AI is starting to formalize that knowledge, making it portable, transferable, and persistent beyond any one person’s memory.
  • What exists today is a real split, not a single capability. Some tools surface carrier options neutrally. At least one already ranks carriers by predicted likelihood of binding. Full automated routing, a system acting without a human choosing, is not here yet, and this piece says so plainly.
  • The first gate this changes is not where a submission ends up. It is whether a carrier gets considered at all. A carrier that never makes an AI-shaped shortlist has effectively lost the business before a human ever weighed in.
  • The deeper risk is not that a carrier’s score is wrong. It is that the score becomes authoritative even when it is measuring the consequence of a carrier’s own past inconsistency, not its current underwriting.
  • A commission that requires five submissions and three phone calls to realize does not behave economically like a commission that binds cleanly on the first try. Agent-side scoring tools are starting to make that difference visible.

Ask an experienced commercial lines producer why she sends a contractor risk to one carrier over another, and you rarely get a spreadsheet. You get a story. That carrier writes contractors well, mostly. Except last spring, when the underwriter changed and three submissions in a row came back declined for reasons nobody could quite explain. So now she calls a different market first, unless it is a habitational risk, in which case the first carrier is still her best bet, appetite change notwithstanding, because the underwriter she actually trusts moved desks but not companies.

None of that lives in a document. It lives in her head, assembled over years, revised by experience, and gone the day she retires or changes agencies.

That is the real subject of this piece. Not whether artificial intelligence can help agents pick carriers, which is a modest and largely settled question by now, but what happens once the informal decision engine every experienced producer carries around starts getting formalized into something that outlives her.

For decades, agencies have run on exactly this kind of tacit knowledge. The best producer knew which underwriter actually liked restaurant risks, which carrier’s published appetite guide overstated what it would really write, and which wholesaler could rescue a submission everyone else had given up on. That knowledge was valuable in large part because it was hard to transfer, locked inside one person’s judgment and unavailable to anyone who hadn’t spent years earning it the same way. What AI is starting to change is not the knowledge itself, but its economics. Once captured, tacit knowledge stops being scarce. It becomes searchable, reusable, and before long, expected.

The other side of the transaction

The Distribution Penalty, an article we published in April, 2026, made the carrier-side version of this argument: inconsistent underwriting quietly costs carriers business they never see leave, because agents route around unreliable markets without ever filing a complaint. The fix that piece proposed was carrier-side observability, submission pattern analysis, quote consistency monitoring, the kind of tooling that lets a carrier see its own reputation from the outside.

This article is about the other side of that same transaction. If a carrier’s inconsistency has always been visible to agents in an informal, word-of-mouth way, what changes when that visibility becomes a formal, shared, algorithmic one?

The short answer is that relationships absorb friction, and algorithms record it. A producer who gets burned by a carrier can still be talked back into the relationship by a good renewal season, a wholesaler’s reminder, or an underwriter who finally returns her calls. A carrier score has no such capacity. It only moves if the data underneath it moves.

That is not, by itself, an argument that the algorithm is better. Human judgment carries its own quiet bias, built on small samples and outdated impressions, one bad account coloring a whole class of business for years after the account itself is forgotten. Algorithmic scoring can institutionalize that same bias, just faster and at greater scale, unless someone is watching how it is built. Both need governance. Neither gets a pass simply for being more consistent, or more human.

Where the technology actually is

It is worth being precise here, because the temptation in a piece like this is to describe a fully automated future as though it already exists, and a skeptical reader will notice the gap immediately.

First Connect’s AI Appetite Finder, launched in March, lets agents describe a risk in plain language and surfaces carriers whose appetite fits, reconciling classification systems an agent would otherwise have to check by hand across multiple documents and portals. First Connect describes the tool explicitly as carrier-neutral guidance that surfaces options without prioritizing specific markets, and not a replacement for underwriting judgment. That is a deliberate design choice. Someone at First Connect decided that ranking carriers by likelihood of success was a line the tool would not cross, at least not yet. That decision is a small, real-world echo of exactly the governance question this piece is raising.

Agentero’s AI Appetite Checker, announced last November, goes a tier further. It draws on historical quoting and binding data, not just appetite guidelines, to identify the carriers most likely to quote or bind a given risk, explicitly optimizing for hit rate. That is not neutral discovery. That is ranking by predicted outcome, a meaningfully different claim.

Neither tool routes a submission outright. Nothing in production today removes the human from the decision entirely. CB Insights now tracks underwriting appetite intelligence platforms as one of eleven distinct technology markets shaping how insurance agencies operate, a useful signal that this is a recognized, growing category rather than two isolated product launches.

It is also worth noting, briefly, that carriers built this kind of scoring for themselves first. Federato’s quadrant scoring system ranks incoming submissions for underwriters on appetite and winnability simultaneously, helping carriers decide where to spend their attention. The agent-side mirror of that capability is younger and thinner by comparison. That gap is itself a small piece of evidence for the argument this series opened with: carriers had both the tools and the incentive to score first, and are only now on the other end of being scored themselves.

So the honest state of the technology is a ladder with two rungs filled and one still open. Some tools advise. At least one ranks. None yet routes. But the ladder is not really the point. Whether a system executes the final decision matters less than whether a carrier makes the shortlist at all. A carrier that never reaches a producer’s attention has effectively lost the submission, whether or not anything ever formally “routed” it elsewhere. Attention, not destination, is the first gate, and that gate is already changing.

The compounding problem

Follow that gate forward and a second, less obvious problem appears.

The original distribution penalty was hidden in relationships: a carrier losing business one quiet decision at a time, invisible because it never showed up as a complaint, only as an absence. The new one is hidden in algorithms, and it moves faster precisely because algorithms do not get talked out of a bad impression the way a person can.

The Distribution Penalty described an adverse selection spiral: an inconsistent carrier loses its best agents first, is left with a thinner and riskier submission pool, and its results worsen as a consequence, reinforcing the very reputation that caused agents to leave in the first place. That spiral has always existed informally. Shape agents’ shortlists algorithmically, even just through ranking rather than routing, and the spiral turns faster, because the human correction layer that used to slow it down, a wholesaler vouching for a carrier, a producer willing to give a market one more try, shrinks with every tool that substitutes a score for that judgment.

But there is a sharper problem underneath the compounding one, and it is the real governance argument of this piece. The danger is not that a carrier’s placement score might be wrong. It is that the score becomes authoritative regardless of what actually produced it. A carrier with a mediocre score for contractor risks might genuinely deserve it. Or the number might be measuring the aftermath of exactly the informal defection described above: good agents quietly left years ago, the submission pool skewed toward harder risks, loss ratios followed, and now a model trained on that history reads the carrier as a poor fit going forward, with no visibility into why. The data has memory, but no context. It cannot distinguish a carrier that is genuinely inconsistent today from a carrier that is still recovering from having been informally routed around years earlier.

AI models do not just learn reality. They learn the consequences of previous decisions, and then they treat those consequences as fact.

That is a different kind of risk than the usual AI bias conversation, and it is one insurance people should recognize immediately, because it mirrors something underwriters already understand about books of business: a portfolio’s current shape is never just a measure of present appetite. It is also a record of every decision that built it, good and bad, often for reasons that have long since stopped mattering.

What this actually costs

Distribution Economics opened this series by asking what a 12% commission actually buys. Agent-side scoring tools are starting to supply an uncomfortable answer: not the number on the contract, but the real, variable cost of earning it. A 12% commission that requires five submissions and three phone calls to close does not behave, economically, like a 12% commission that binds cleanly on the first attempt, even though the contracted rate is identical. That gap, invisible in any commission statement, is precisely what appetite matching and ranking tools are beginning to expose, carrier by carrier, in a way no producer’s private mental model ever could at scale.

Morgan Stanley’s January research projected a 350 basis point operating margin uplift for brokers from AI by 2030, against 180 basis points for carriers. That gap does not by itself explain the mechanism described here, and this article is not claiming it does. But it is consistent with a broker sitting at the exact point in the transaction where information turns into placement behavior, which is precisely where these tools are starting to concentrate.

What carriers can still do about it

The tempting advice at this point is a familiar list: build better appetite guides, open better APIs, feed the models cleaner data. All true, and all a little beside the point.

The more uncomfortable point is this: carriers are entering a world where the market may evaluate them using the same kind of automated intelligence they have spent years applying to everyone else, agents, producers, even their own underwriters. The Distribution Penalty argued that carriers needed observability into their own consistency before their agents lost patience with them informally. That deadline has not gone away. It has just gotten shorter, because part of the audience for that observability is already split between tools that advise and at least one that ranks by outcome, and the tools that still just advise are advising an increasingly well-informed reader.

The 19% of carriers offering agents real-time appetite visibility, the figure this series has cited before, are no longer just serving human curiosity. They are feeding the same systems that are starting to decide who gets a look at all.

The choice underneath the choice

Human relationships can be repaired with time and attention, but they can also be manipulated by relationship capital that has nothing to do with risk quality, a good lunch, a long friendship, a favor owed. Algorithmic scores cannot be charmed. They also cannot be reasoned with, and they carry the specific risk of mistaking yesterday’s damage for today’s truth.

Neither one, on its own, is the honest broker insurance distribution actually needs. Distribution has always run on an algorithm, informal or formal, human or machine. The only real choice carriers have now is whether that algorithm gets built and governed in the open, with someone accountable for what it optimizes and what it forgets, or left to run the way it always has: unexamined, just faster.

Related IIAI Articles

Continue the Distribution Economics series:

Sources

Subscribe to AI Insights Newsletter

AI Disclaimer: This content was created with assistance from artificial intelligence technology. While content is based on factual information from the source material, readers should verify all details directly with the respective sources before making business decisions.

Related Posts

468