# The E&S boom is a capacity problem (https://riska.io/blog/es-growth-is-a-capacity-problem)

_2026-08-07 · Riska Underwriting Team · business-ecosystem_

Surplus lines premium grew 2.8 percent across the stamping office states in the first half of 2026. The number of filings grew 16.9 percent. Growth is reaching underwriters as transactions rather than as revenue, which makes the binding constraint how much of the week goes to submissions that were never going to be written. A declined submission costs the same reading as a bound one, and almost nobody counts it.

## Key takeaways

- Surplus lines premium across the stamping office states rose 2.8 percent in the first half of 2026 while item filings rose 16.9 percent, according to WSIA's 2026 midyear report. Growth arrived as transactions, not as revenue.
- Florida recorded $9.36 billion of surplus lines premium in the first half of 2026, down 6 percent, against 982,627 policies, up 15 percent. Average cost per policy fell from $11,608 to $9,525.
- Kinsale Capital Group's 2025 annual report discloses about 988,000 new business submissions, about 711,000 quotes and 75,000 bound policies. Roughly thirteen submissions were read for every policy that reached the book.
- Hiring will not close the gap. The Bureau of Labor Statistics counted 127,000 insurance underwriters in 2024 and projects the occupation to decline over the decade to 2034.
- A declined submission consumes the same reading, the same clearance and the same loss run review as a bound one, and almost no book records that it happened.

## Premium up 2.8 percent, filings up 16.9 percent

It is Tuesday. Forty-one submissions came in overnight. Eleven of them are coastal property in counties you stopped writing in March, and you will open most of the forty-one anyway, because the only way to know which eleven are the eleven is to read them. A broker you like has been waiting since Thursday on the one you have not got to.

That is not a story about a hard market. It is a story about count.

The Wholesale and Specialty Insurance Association publishes a midyear report of what the stamping offices actually processed. Its [2026 midyear stamping office report](https://www.insurancebusinessmag.com/us/news/excess-surplus/property-softens-liability-hardens-what-eands-midyear-data-means-for-brokers-585004.aspx) put total surplus lines premium at $47.6 billion for the first half of the year, up 2.8 percent on the same period in 2025. Item filings over those six months came to 4.3 million, up 16.9 percent.

A year earlier those two numbers moved together. The 2025 midyear figures were $46.2 billion of premium and 3.7 million items, up 13.2 percent and 12.4 percent. In 2026 they came apart, and the gap runs six to one in favour of the thing that costs you time rather than the thing that pays for it.

Florida is the cleanest single illustration, because its [service office publishes premium and policy count in the same report](https://createsend.com/t/r-D115E9B7597200992540EF23F30FEDED). For the first half of 2026 it recorded $9.36 billion of premium, down 6 percent, against 982,627 policies, up 15 percent. Average cost per policy fell from $11,608 to $9,525. Its executive director, Mark Shealy, put it flatly in the July report: “Policy volume continues to rise, while average cost per policy has moved lower.”

California points the same way. The Surplus Line Association of California posts premium and transaction count month by month on a submitted date basis. Add up January through July and 2026 carries 7.8 percent more premium than the same seven months of 2025, and 24.4 percent more transactions. Those totals are arithmetic on published monthly rows rather than an audited annual figure, and the direction is not in doubt.

Filings are a proxy for how much work arrived. Premium is a proxy for what the work turned out to be worth. When the first grows six times faster than the second, the problem in front of an underwriting team is not pricing.

### Five sources, five different quantities

These bodies are not measuring the same thing, and two of them read commercial property in opposite directions in the same quarter. The Ivans Index, computed from more than 120 million transactions moving between over 38,000 agencies and 700 carriers and MGAs, put commercial property premium renewal change at plus 6.40 percent in the second quarter of 2026. Marsh, whose index measures rate change at renewal and which states that “our rates reflect the segment mix of Marsh's client portfolio”, put US property down 13 percent in the same quarter. Neither is wrong. A rate can fall while a premium rises if insured values are reindexed upward at renewal, and a global broker's property book is not the population an independent agency renews.

| Source | Period | What it measures | Latest reading |
| --- | --- | --- | --- |
| WSIA midyear stamping office report | H1 2026 | Premium and item filings actually lodged in the states that operate a stamping office. A census of transactions, not a survey. | Premium up 2.8 percent, items up 16.9 percent |
| Florida Surplus Lines Service Office | H1 2026 | Premium, policy count and average cost per policy filed in one state. | Premium down 6 percent, policies up 15 percent, average cost per policy down 18 percent |
| Surplus Line Association of California | Jan to Jul 2026 | Premium and transaction count by month, dated when the broker submitted the filing. | Premium up 7.8 percent, transactions up 24.4 percent |
| Ivans Index | Q2 2026 | Year over year premium change on a consistent renewing policy, computed from transactions moving between agencies and carriers. | Commercial property up 6.40 percent |
| Marsh Global Insurance Market Index | Q2 2026 | Rate change at renewal on the programs Marsh places, weighted by the segment mix of its own client portfolio. | US property down 13 percent |

_California figures are sums of the monthly rows the association publishes, on a submitted date basis, rather than an audited annual total._

## Hire more underwriters, or tighten the appetite guide

Two answers to this are already in the room, and both of them are what good shops do.

The first is headcount. It is the honest answer, it works, and it is the slowest lever available. A property underwriter who can read a 34-location coastal schedule and price it is a five-year hire, not a job-board hire. The wider numbers do not help either: the Bureau of Labor Statistics counted [127,000 insurance underwriters in 2024](https://www.onetonline.org/link/summary/13-2053.00) and projects the occupation to decline rather than grow over the decade to 2034. Whatever solves a volume problem in this segment, it is not a bigger hiring plan.

The second is the appetite guide. Tighten it, get it current, push it out to your brokers, and the submissions that arrive are more often the ones you want. This is real discipline and the shops that do it well have better inboxes than the shops that do not.

It has two ordinary failure modes. The guide only works if the broker reads it, and a wholesaler placing across nine carriers is not memorising nine appetite documents. And it is only true on the day it was written. The guide you revised in March describes March, and the property market moved underneath it by a magnitude that the two rate indices in the table above cannot even agree on.

Neither answer is wrong. They are both slow, and the volume is not.

### How volume turns into a worse book

Nothing in this cycle requires anyone to make a bad decision.

1. **Volume rises** more submissions arrive than the week has hours for
2. **Everything opened** a risk is not ruled out until it is read
3. **Response slips** good submissions wait behind ones you were never writing
4. **Brokers re-route** the next one goes to whoever answers first
5. **Mix worsens** you read more to find the same good risks

_Returns to Volume rises._

_Every turn is paid for in reading time on business that was never going to be written._

## A declined submission is not free

Most books treat a decline as a null event. No premium was written, no commission moved, no policy exists, so nothing is recorded except a terminal status on a row nobody queries afterwards.

Here is what happened before that status was set. Someone opened the mail and read the cover note. They opened the application and found the location schedule, which was a spreadsheet with 34 rows and merged cells. They checked whether the account was already on the book under a slightly different name. They opened the five-year loss run, which was a scan, and went looking for the two large ones. They checked construction class and year built, checked distance to coast, and got far enough down the guide to know the answer was no.

Call it twenty minutes. That figure is my arithmetic on the steps above and nothing more, and you should check it against your own week rather than take it from me. What is not arguable is that every one of those steps happened, and that they happened whether or not anything downstream records that they did.

There is no good recent published measure of this, and the most quoted one is worse than useless. The claim that underwriters spend seventy percent of their time not underwriting traces back to a 2021 survey of 434 US underwriters run by Accenture with The Institutes, in which the reported figure was up to forty percent on non-core and administrative work. The seventy percent version gets there by counting negotiation and sales support as non-underwriting, which in surplus lines is the job. Be suspicious of the whole category, including my twenty minutes.

What is measurable is the funnel. Kinsale Capital Group discloses its own in its [2025 annual report](https://www.sec.gov/Archives/edgar/data/1669162/000166916226000015/knsl-20251231.htm): about 988,000 new business submissions processed, about 711,000 quotes issued, and 75,000 policies bound. That is a 7.6 percent policy-to-submission ratio at a carrier that ran a 75.9 percent combined ratio for the year. Whatever the right number of minutes is, it is being spent roughly thirteen times for every policy that reaches the book.

The same filing shows the 2026 shape arriving early. Kinsale's commercial property premium fell 17.9 percent in 2025 while the rest of the book grew 13.3 percent, which it attributes to continued strong submission flow from brokers, and its average premium per policy fell from $15,100 to $13,400. More policies, each one smaller. The reading per dollar went up.

The cost is not only the hour. It is which hour. The submission you did want was in the same inbox and it waited. Quote turnaround is what a broker actually ranks you on, and the broker who waits sends the next one somewhere else first.

## What you give up by scoring appetite before a person reads it

Score a submission before anyone opens it and you will sometimes turn away business you would have written. The 34-location schedule with the one warehouse that pushes total insured value past your threshold is also the account where the other 33 locations are exactly what you want. A rule does not know that. An underwriter who has written that broker for six years does, and that judgment is what you are paying for.

The second limit is quieter and worse. An appetite guide encoded in software is only as current as the day someone last edited it. A PDF that is eighteen months old at least looks eighteen months old. A scoring model that is eighteen months old looks exactly like one that was updated this morning, and it goes on producing confident numbers the whole time.

That is a real cost and it is worth naming rather than arguing away. Two things reduce it and neither removes it. A score that shows its arithmetic can be disagreed with, which is not true of a number on its own. And a result that is recorded rather than enforced leaves the decision with the person, so a model being wrong costs a conversation instead of an account.

## How Riska builds appetite so a decline leaves a record

What we took from all of this is that the expensive part of a growing book is not the decision. It is everything that has to happen before the decision can be made, and the fact that when the answer is no, none of it is written down anywhere you can use.

So [appetite](/platform/appetite) in Riska is a model rather than a document. Categories group dimensions, each dimension turns one value off the record into a score from 0 to 100 by tiers, a curve, a fixed value or straight passthrough, and you decide how hard a single weak input is allowed to drag the category it sits in. Modifier categories multiply the base score inside a floor and ceiling you set. What comes out is one number and the whole arithmetic behind it: for every dimension the raw value, the transformed score, the normalised score and its weight.

Knockouts and referrals attach to the rule that fired them, at the profile, a category or a single dimension, and they never move the score. Recorded, not enforced. That is deliberate. The software's job is to say which rule this hit and what you wrote next to it; deciding is still a person's job. Every run writes a new snapshot with the definition it read pinned to it, so editing the profile tomorrow does not rewrite yesterday's decision.

The first hour of a submission is a separate problem, and [submission triage](/platform/submission-triage) is where it goes. Attachments are registered against the submission and classified, so the queue tells you the application arrived and the loss runs did not. The account is matched against what you already hold. Extracted values land as a proposed change to the record rather than a write to it, and when a run cannot resolve something it stops and asks rather than picking the highest-scoring guess.

And because lifecycle states carry roles, declined among them, a [portfolio metric](/platform/reporting) breaks out by the state a risk is sitting in. The declined bucket is something you can look at, next to the bound one, with the same definition behind both.

None of that makes a decline free. It makes it counted, which is the part missing today. A book that knows what it turned away and why can argue about its appetite. A book that only records what it bound is arguing from memory.

## Questions we get asked

### Is the E&S market still growing?

Yes, but the growth looks very different depending on what you count. WSIA's 2026 midyear report put surplus lines premium across the stamping office states at $47.6 billion for the first half of the year, up 2.8 percent year over year, while item filings reached 4.3 million, up 16.9 percent. Transaction volume grew roughly six times faster than premium.

### What is a stamping office?

A state body that surplus lines brokers file each transaction with. It checks the filing, records it and charges a stamping fee. Fifteen states operated one as of WSIA's 2025 midyear report. Because every transaction has to be filed, stamping office data is a count of what actually happened rather than a survey of what people remember.

### Why did surplus lines premium fall while policy count rose in 2026?

Because property rates softened while transaction volume kept climbing. The Florida Surplus Lines Service Office recorded $9.36 billion of premium in the first half of 2026, down 6 percent, against 982,627 policies, up 15 percent, with average cost per policy falling from $11,608 to $9,525. More risks are being placed and each one carries less premium than it did a year ago.

### How do MGAs measure submission volume?

Most count what arrived and what bound, and nothing in between. The three numbers that actually describe capacity are submissions received, submissions that reached a quote, and policies bound. Kinsale Capital Group publishes all three in its annual report: for 2025, about 988,000 submissions, about 711,000 quotes and 75,000 bound policies.

### What is a hit ratio and what is a good one?

A hit ratio is bound business divided by opportunities, and the number means nothing until you say what the denominator is. Kinsale's 2025 filing shows both ends of the range on one book: 7.6 percent of new business submissions became bound policies, while 72.0 percent of submissions received a quote. There is no universally good figure, because a book that quotes selectively and a book that quotes everything are counting different things.

### Will scoring appetite automatically decline business I would have written?

It would if the score were allowed to decide, which is why in Riska it is not. A knockout or a referral is recorded against the rule that fired it and never changes the score, and nothing is declined by the software. The point of scoring against a defined model is that the reasoning becomes visible and arguable, not that the judgment moves off the underwriter.