How Mid-Size MGAs Scale Underwriting Operations Without Adding Headcount

US managing general agents wrote 12% more direct premium in 2025 than the year before. Asked about hiring in March 2026, carriers gave the Jacobson Group and Aon plans amounting to 0.91% employment growth over the following twelve months. Different populations and different windows, so not a clean ratio, but the direction holds: premium is compounding roughly thirteen times faster than the industry intends to hire.

Most MGAs see that as a broker who stopped sending the good accounts, and nobody can say quite when it started. Quote turnaround slipped from two days to four over a couple of quarters, survivable each time it happened, and then the submission flow that used to arrive steadily started arriving somewhere else.

The slippage is hard to catch early for a reason that has nothing to do with effort. A desk running at 85% of capacity and one running at 95% look identical on a headcount report and feel much the same to the people sitting at them, but the second makes brokers wait more than three times as long.

This is a capacity model for mid-size MGAs: the five numbers governing how much your existing desk can carry, where each one stops holding, and which function to automate first at each volume band. Whether to build that capability or buy it runs on different ground, covered in our build-versus-buy analysis for MGAs.

Key takeaways

• Premium is outrunning people. MGA direct premium written grew 12% in 2025 while planned industry employment growth sits at 0.91%. The gap has to close through throughput per underwriter, because it will not close through hiring.
• There is no published submissions-per-underwriter benchmark. Private benchmarking services exist for carriers, but nothing citable covers MGAs, so the only defensible number is the one you calculate from your own desk.
• Utilization, not volume, is what breaks. Queue delay scales with ρ/(1−ρ). At 85% utilization a 10% rise in submissions makes brokers wait about 2.5 times longer, which is why growth feels manageable right up until it isn't.
• The failure mode changes with size. Below roughly 50 submissions a month you have a data-entry problem. Around 200 you have a queueing problem. Past 500 you have a divergence problem, and each one calls for a different first move.
• Speed compounds into growth. Lynx Specialty is growing about 35% this year not by adding brokers but by answering the ones it already had more quickly.

Premium is growing about thirteen times faster than headcount

Conning put the total US MGA premium at approximately $128 billion for 2025. The portion reported through statutory Note 19 disclosures came to $102.6 billion, up 12% from 2024 and, in Conning's framing, "more than double the broader property-casualty market's approximately 5% growth rate."

The labor side runs on a different clock. In the Q1 2026 Insurance Labor Market Study, 50% of carriers told the Jacobson Group and Aon they plan to increase staff and 43% to hold flat, the latter a 15-year peak. From those plans the analysts project "a 0.91% increase in employment during the next 12 months."

Jacobson and Aon describe a stable market rather than a scarce one. Underwriting sits among "the industry's greatest need" alongside technology and claims, while the roles they name as hardest to fill are actuarial, executive and analytics. So the constraint is not that underwriters cannot be found, but that a segment growing at 12% cannot staff its way forward on a 0.91% hiring plan, and headcount is a slow instrument for a fast problem.

Capgemini arrives from the other direction. In its World Property and Casualty Insurance Report 2026, 57% of the 209 underwriters surveyed said they spend the majority of their time on routine tasks: data gathering, document review, basic eligibility checks. Capgemini's own summary of that finding: "Underwriting is where the capacity trap is most costly." The consequences it records are commercial: among those underwriters, 61% struggle to increase quote-to-bind conversion and 57% to maintain underwriting accuracy and risk quality.

That is the shape of the problem. The capacity already exists inside the desk you have, but it is being spent on document handling.

The capacity model: five inputs

There is no published benchmark to measure yourself against. We looked through Capgemini's World Property and Casualty Insurance Report for 2024, 2025 and 2026, the Accenture and Institutes P&C underwriting survey, Accenture's Underwriting Rewritten, AM Best's MGA and DUAE market segment reporting, Conning's MGA study and Deloitte's 2026 outlook. None of them publishes a submissions-per-underwriter figure. Aon's Ward benchmarking practice runs workload and productivity benchmarking for participating insurers, but those figures go to participants rather than being published, and they cover carriers rather than MGAs.

The closest public reference point belongs to a carrier. On Kinsale Capital Group's fourth-quarter 2025 earnings call, CEO Michael Kehoe put the company's volume at "give or take, 1 million submissions last year," and Chief Underwriting Officer Stuart Winston described "the ability to quote more than 70% of all new business submissions" as a competitive advantage. That is a high-volume, small-account E&S book, and a different operating model from most specialty MGAs.

So the number has to be derived. Everything that governs how much your desk can carry reduces to five values. Pull them for one line of business over one recent quarter, not for the whole book, because mixing a binding-authority program with a large-account property book produces an average that describes neither.

1. Arrival rate (S). Submissions received per underwriter per week. Count everything that enters the queue, including duplicates and out-of-appetite risks, because those consume handling time before they are dismissed.

2. Clearance ratio (c). The share of arrivals that survive triage and get a genuine underwriting look. Everything else is dismissed on appetite, duplication or incompleteness.

3. Touch time (t). Average hours of actual underwriter attention on a submission that gets a full look, from opening it to issuing a quote or declining with a reason.

4. Rework multiplier (1+w). The proportion of files reopened after the first pass because data was missing, wrong, or had to be chased from the broker. A file touched twice costs roughly twice as much.

5. Referral load (ref × r). The share of files that go to a senior underwriter, multiplied by the handling time each referral adds on the originating desk. This excludes the referee's own time, which is a separate constraint.

Hours your desk needs each week come out as:

(S × c × t × (1+w)) + (S × c × ref × r)

Divide that by the hours each underwriter genuinely has available for new-business submission work, and you have utilization. Not a productivity score. A ratio that predicts your turnaround, and that can exceed 1, at which point the queue is formally unstable and the backlog grows until something is turned away.

A worked example

The inputs below are illustrative, chosen to be plausible for an upper-mid-size commercial desk. Substitute your own; the structure is the point, not the figures.

An MGA takes roughly 470 new-business submissions a month across six underwriters, so about 18 a week each. Around 45% clear triage into a full look, so eight files. Each takes about 1.2 hours of real attention. A quarter get reopened for missing information. Roughly 15% are referred, adding half an hour each on the originating desk.

The available-hours figure is the one people get wrong. It is not a 40-hour week. New business competes with renewals, endorsements, broker meetings and internal work, and Capgemini's time-allocation research puts core underwriting at roughly a third of an underwriter's time. Fourteen hours a week for new-business submission handling is a reasonable working figure.

‍

‍

That desk is not underperforming. It clears everything that arrives, with over an hour to spare, and on any dashboard measuring output it looks fine. It is also sitting five points past the threshold where the queue stops behaving, which is why brokers notice before anyone internally does.

The 85% rule, and why growth breaks suddenly

Waiting time in a queue with variable arrivals does not rise in proportion to how busy the server is. It rises with the ratio ρ/(1−ρ), where ρ is utilization. This is the utilization term in Kingman's approximation for queue waiting time, published in 1961 and used across operations management ever since. Two caveats worth carrying: it is a single-server approximation, most accurate as utilization approaches 1, and a pooled desk where any underwriter can pick up any file sits on a flatter curve than a set of individually-owned queues. Both are arguments for measuring utilization per desk rather than across the book. Little's Law, from the same year, gives the companion result: the amount of work in progress equals the arrival rate multiplied by the average time each item spends in the system.

Turnaround always degrades faster than volume grows; what changes above roughly 85% is that it stops being survivable. Eighty-five percent is a working threshold rather than a constant derived from the formula, but it is where the curve turns steeply enough that a normal quarter of growth becomes a service failure. The practical consequence is a table every operations leader should be able to draw from memory.

‍

‍

Two desks at 85% and 95% both look busy. The second makes brokers wait more than three times as long.

Now apply growth. A 10% increase in submissions does very different things depending on where you start:

‍

‍

This is why the pattern is so consistent and so hard to anticipate. An MGA absorbs three consecutive years of growth with no visible strain, then takes on a modest new program and turnaround collapses within a quarter. Nothing changed except position on a curve that is nearly flat on the left and vertical on the right.

It also explains the shape of the fix. Return to the worked desk at 90% utilization and suppose touch time falls from 1.2 hours to 0.7 through automated intake, extraction and pre-fill, while the rework rate drops from 25% to 8% because data arrives structured and validated rather than re-keyed. Required hours fall from 12.6 to 6.6. Utilization lands at 47.5%, and the queue multiplier goes from 9.0 to 0.9.

A 42% reduction in handling time produced a tenfold improvement in how long brokers wait. That is not a rounding artifact. It is what happens when you move a system off the steep part of the curve, and it is the single most important reason capacity work outperforms hiring at the margin: an extra underwriter adds capacity linearly, while removing handling time buys back queue position non-linearly.

The same arithmetic sets your growth headroom. A desk at 50% utilization can absorb roughly 70% more volume before it reaches 85%. A desk already at 85% can absorb none.

What breaks at 50, 200 and 500 submissions a month

Volume alone does not determine the bottleneck. What changes with scale is which constraint binds first, and each band calls for a different first move. The boundaries below are our own framework drawn from client work rather than published research, so treat them as soft; what travels is the order in which the constraints arrive, not the exact thresholds.

Under ~50 submissions a month

One or two underwriters, often the founders, and enough slack that queueing never becomes the issue. The constraint is straightforward manual handling: someone retyping ACORD forms, building the schedule of values by hand, chasing the broker for the loss runs. Utilization is low enough that a bad week is absorbed rather than compounded.

The right first move here is intake capture, and it usually does not require a platform commitment. Configurable automation gets an MGA of this size a long way, and we have covered the options for that segment in a separate guide to no-code intake platforms for small MGAs. That is the better starting point if you are in this band. The rest of this article addresses constraints you do not have yet.

Roughly 50 to 200 submissions a month

Three to six underwriters, and the first band where queueing behavior appears. The binding constraint moves from data entry to triage. Enough submissions now arrive that deciding what deserves attention consumes real capacity, and out-of-appetite risks are absorbing time before anyone establishes they were never writable.

Clearance is the function to automate first. Ivans found in its 2026 connectivity survey of 702 independent agents that commercial-lines-heavy agencies re-key data at a rate of 70%, against 62% for personal-lines-focused agencies, and that those same commercial-lines-heavy agencies rely on email 49% of the time versus dedicated submission capture software at 7%. Most of what arrives at this band arrives as email attachments, and until something reads and classifies them automatically, an underwriter is doing it.

The measurable target is the clearance ratio itself. If you cannot state what share of arrivals get a full look, that is the number to instrument before buying anything.

Roughly 200 to 500 submissions a month

Six to fifteen underwriters, usually across several programs, and the band where most mid-size MGAs discover the 85% problem. Triage is handled. The constraint has moved to extraction and referral dwell.

Two things bind here at once. First, the work of turning submission documents into a structured risk view is now the largest single consumer of touch time, particularly on property schedules and loss runs. Second, referrals begin sitting. A 15% referral rate against eight full looks is barely more than one file a week per underwriter, which sounds trivial until you multiply it across a six-person desk and seven or eight files a week are landing on senior underwriters who carry their own books as well. The delay lands on the accounts most worth winning, because referrals skew toward larger and more complex risks.

Automate extraction first, then referral routing. The reconciliation problem underneath this, where the same risk is described differently by an SOV, an inspection report and a broker email, is worked through in detail in our guide to aggregating multi-source risk data.

Capgemini's 2026 report documents what this is worth when it works. AG Insurance built an assistant that synthesizes information from internal systems, inspection reports, public databases, media sources and company disclosures into a unified view. Capgemini records that "on average, the tool delivers time savings of 40-60% per file, freeing underwriters to respond more quickly to submissions, generate more quotes, and support portfolio growth." A 40-60% cut in per-file time is the same order of magnitude as the worked example above, and produces the same non-linear effect on queue position.

Above ~500 submissions a month

Fifteen or more underwriters across multiple programs and carrier relationships. Throughput is no longer the primary risk. Divergence is.

At this scale no single person reads a representative sample of the files any more, and programs start documenting differently. The failure surfaces at audit rather than at bind, and it lands on whichever team documented least. The fix is a governed standard for what a finished file contains, which is a different discipline from capacity and is covered in our guide to automating underwriting summaries with audit capabilities. If you are a large MGA whose main question is standardizing summary creation across teams, start there rather than here.

Deloitte's 2026 global insurance outlook records what large-scale submission triage now looks like: "AIG launched a gen AI-powered underwriting assistant with Anthropic and Palantir, which ingests and prioritizes every new excess and surplus submission, allowing review of additional policies without adding new staff." Every new submission, prioritized, with the headcount unchanged.

What to automate first, by band

‍

‍

The sequence matters more than the destination. Automating extraction before clearance means paying to structure documents that should never have reached an underwriter. Automating referral routing before extraction routes files that still need an hour of manual assembly once they arrive. Each band earns the next.

For a stage-by-stage view of which capabilities cover which part of the process, our 2026 buyer's guide to underwriting workflow software maps the market by workflow stage rather than by vendor.

What this looked like at Lynx Specialty

Lynx Specialty is growing about 35% this year, and the mechanism is exactly the one the model predicts. In a shifting specialty market, the firm got there not by adding brokers but by responding faster to the ones it already worked with.

Paul Ritter, Senior Vice President of Lynx Specialty, described the loop directly: "More brokers within our existing relationships are sending more submissions in, because we're responding so quickly. That means more quotes out the door, more bind orders, and in a changing market, that's been crucial for us to continue to grow at about a 35% cliff this year so far."

Read that against the headroom arithmetic. A desk at 50% utilization can absorb roughly 70% more volume before it reaches 85%. Lynx's growth sits inside that envelope, which is one explanation for why the additional submissions produced additional quotes rather than a lengthening queue. We do not have Lynx's utilization figures, so treat that as a consistent reading rather than a demonstrated cause.

What the arithmetic does show is the counterfactual. Had the same 35% landed on a desk already running at 85%, required hours would have exceeded available hours outright. At that point the queue does not lengthen, it grows without bound, and business gets turned away by the calendar rather than by a decision.

The pattern generalizes because broker behavior is responsive rather than fixed. Ivans found that 90% of agents have reduced business with a carrier due to friction around submissions. Turnaround is not a service metric that sits downstream of growth. It is an input to growth, and the same capacity work that protects it also produces it.

Scale shows up elsewhere in the same shape. One of the largest MGAs in the US, writing over $1.5 billion in premium across more than 20 programs, cut average time-to-clear a submission from roughly 32 minutes to about one minute, which the case study records as a 30x improvement, with a 200%+ gain in underwriting efficiency in the first three months. In that window the workflow handled over $20 billion in total insured value and saved more than 2,000 hours of manual effort at close to 100% accuracy. Thirty-one minutes recovered per submission, at that volume, is a capacity release no hiring plan would have matched.

Instrument these four numbers first

Most mid-size MGAs cannot currently calculate their own utilization, which is the actual obstacle. Four measurements make the model runnable, and none requires new software to start.

Clearance ratio. Files that receive a full underwriting look, divided by arrivals, per program. Tells you how much capacity triage is consuming before any underwriting happens.

Touch time. Median underwriter hours per full look. The number automation moves most directly, and the one worth sampling by hand for a fortnight if no system captures it.

Rework rate. Share of files reopened after first pass. A leading indicator of intake quality, and usually the most under-measured of the four.

Referral dwell. Calendar hours between referral and decision. Distinct from referral rate; a 20% referral rate with same-day decisions is healthy, and a 10% rate with three-day dwell is not.

Track them monthly by program rather than across the book. Utilization is a per-desk property, and a blended figure will hide the one program that is quietly at 95%.

Few systems of any kind capture these four measures as standard, and underwriting tooling generally remains thin on the ground. In Capgemini's 2026 survey, only 31% of underwriters said they use a workbench with AI recommendations, and those who do were approximately 1.4 times more likely to increase book growth. Until the tooling catches up, a two-week hand count beats waiting for a system to produce the number.

How FurtherAI fits

FurtherAI is built for the middle two bands, where the constraint is handling time rather than tooling gaps or governance. Submissions arrive as email attachments, ACORD forms, schedules of values and loss runs; the platform reads them, structures them, checks them against appetite and carrier guidelines, and presents a decision-ready file with its sources traceable. The effect on the model is direct: touch time and rework both fall, which moves utilization down the curve where small improvements produce large gains in turnaround.

The results above come from documented customer work rather than projections. If you want to see the arithmetic run against your own numbers, a demo is the fastest way to get there.

Frequently asked questions

‍

REFERENCES

^{Accenture. "Underwriting Rewritten." Accenture, August 2025. accenture.com}

^{AM Best. "Best's Market Segment Report: MGA Premiums Showed Double-Digit Growth for Fourth Straight Year in 2024." AM Best, June 4, 2025. businesswire.com}

^{Aon. "Benchmarking Solutions for Insurers." Ward Benchmarking, Aon. aon.com}

^{Applied Systems. "2026 Insurance Agency-Carrier Connectivity Trends Survey Report." Ivans, August 26, 2026. ivans.com}

^{Capgemini. "Unleashing Growth: The Evolving Role of Underwriters." Capgemini Research Institute, 2024. capgemini.com} 

^{Capgemini. "World Property and Casualty Insurance Report 2026." Capgemini Research Institute, May 2026. capgemini.com}

^{Conning. "U.S. MGA Premiums Reach $128 Billion as Market Evolution Continues." Conning, July 28, 2026. conning.com}

^{Deloitte. "2026 Global Insurance Outlook." Deloitte Insights, 2026. deloitte.com}

^{FurtherAI. "How Lynx Specialty Grew 35% by Responding Faster, Not Adding Brokers." FurtherAI, June 1, 2026. furtherai.com}

^{FurtherAI. "Submissions Processing AI." FurtherAI Customer Stories. furtherai.com}

^{Kingman, J. F. C. "The Single Server Queue in Heavy Traffic." Mathematical Proceedings of the Cambridge Philosophical Society 57, no. 4 (1961): 902–904. doi.org/10.1017/S0305004100036094}

^{Kinsale Capital Group. "Q4 2025 Earnings Call Transcript." February 13, 2026. fool.com}

^{Little, John D. C. "A Proof for the Queuing Formula: L = λW." Operations Research 9, no. 3 (1961): 383–387. doi.org/10.1287/opre.9.3.383}

^{The Jacobson Group and Aon. "Q1 2026 Insurance Labor Market Study." The Jacobson Group, March 3, 2026. jacobsononline.com}

‍

DISCLAIMER 

This article is for general informational purposes only and does not constitute legal, regulatory, compliance, underwriting, or other professional advice. The content reflects information available as of the date of publication, and FurtherAI undertakes no obligation to update it as laws, regulations, or AI technologies evolve.