When I was working the E&S desk at a Hartford carrier, we had a running joke: every underwriter needed two sets of skills — one for actually underwriting, and one for locating the relevant data inside a 60-page broker packet. It was not funny. It was accurate.
That experience is what eventually became Undwrlyft. But before getting into what we built and why, it's worth being precise about the nature of the problem — because "the submission bottleneck" gets thrown around a lot in insurtech circles without the operational specificity that makes it actionable.
What the bottleneck actually is
Specialty lines submissions arrive at carriers in a format that was designed for human comprehension, not data processing. A typical commercial property E&S submission might include a broker cover letter, an ACORD 140 or 146 form, supplemental applications specific to the risk class, five to ten years of loss run PDFs from prior carriers, a statement of values or schedule of locations, and various supporting documents — environmental reports, engineering surveys, financial statements depending on the risk type.
The underwriter's job is to assess whether this risk fits appetite, at what pricing, under what conditions. That is the judgment work. The intake work — finding the right figures across those documents, validating them for consistency, and entering them into a pricing model — is a prerequisite, not the job itself.
The ratio of intake work to judgment work is where the bottleneck lives. For a complex E&S submission, intake alone can consume three to five hours of an underwriter's day. If you have a 12-person underwriting team handling 700 submissions per month, that's a material share of total available underwriting capacity being spent on document triage rather than risk assessment.
Why automation has lagged in specialty lines
Standard commercial lines — commercial auto, BOP, workers' comp — automated earlier and faster. The reason is structural. Standard lines submissions follow predictable formats, relatively stable NAIC class codes, and well-defined ISO rating structures. ACORD forms for standard lines carry consistent field populations because the underlying risk classes are well-defined and carriers have aligned on what information they need.
Specialty and E&S risks don't have that luxury. An environmental contractor submission is structurally different from a technology E&O submission, which is different from a surplus lines property schedule. The supplemental applications are carrier-specific and often class-specific. Loss run formats vary by prior carrier — a loss run from a regional specialty writer looks nothing like one from a London market syndicate. Prior liability limits, attachment points, and retention structures appear in different sections of different documents.
This variability killed the early rule-based automation attempts. You cannot write a template-based parser that reliably handles the format diversity of E&S submissions, because E&S is, by definition, the lines that exist because standard forms don't fit.
What changed — and what the hype misses
The capability shift that makes submission automation viable now is the combination of large language model document understanding with structured extraction pipelines trained on insurance-specific document corpora. This is not magic — it's a meaningful technical advance in the ability to find semantically equivalent fields across structurally different documents.
We should be clear about what this is not, however: it is not a general AI that "understands" insurance. We're not saying that document extraction replaces underwriting judgment. An automated system can reliably extract five years of incurred losses, identify the prior carrier, flag the attachment point structure, and populate the standard pricing model fields. It cannot determine whether the insured's claims management practices suggest adverse development risk, or whether the description of operations matches the NAICS code the broker assigned. Those remain human decisions.
The insurtech hype machine tends to oversell the judgment layer because judgment sounds more impressive than data entry reduction. The operational reality is that eliminating the data entry layer is where the capacity gain actually lives.
The volume problem is not going away
E&S premium volume has grown materially over the past several years as admitted market capacity has tightened in coastal property, habitational risks, and certain casualty classes. NAIC data on E&S market trends shows consistent written premium growth as risks migrate from admitted to non-admitted markets. That growth compounds the submission volume problem: more submissions are arriving at E&S carriers, each requiring the same manual intake process, with underwriting teams that have not grown proportionally.
Consider a mid-size E&S carrier processing 800 submissions per month across a 10-person underwriting team. Even modest automation of the intake function — say, reducing per-submission intake time from 90 minutes to 20 minutes — translates to several hundred hours per month of recaptured capacity. That's underwriting capacity redirected from data entry to actual risk assessment, which improves quote turnaround, reduces declination rates for submissions that might otherwise have timed out, and lets underwriters take on more complex accounts that genuinely require their expertise.
Broker relationships and the hidden cost
There's a less-measured aspect of the submission bottleneck that deserves attention: what happens to broker relationships when turnaround time lags. Specialty brokers place business with wholesale brokers and carriers who give them fast, reliable responses. When a submission sits in a queue for five or six business days before an underwriter even confirms receipt, that broker starts allocating their best submissions elsewhere. The carrier doesn't see the business it didn't get.
This dynamic is particularly sharp in hard market conditions for specific risk classes. When admitted market capacity exits a class abruptly — as happened in parts of the coastal property market after successive hurricane seasons — E&S capacity becomes the only option for brokers placing those risks. The carriers that can quote quickly capture the business. The ones with submission clearance backlogs do not.
The realistic state of the art in 2025
Building in Hartford means I spend a lot of time in conversations with CUOs and VP Underwriting roles at specialty carriers. The market is past the "should we care about submission automation" question. The question now is about accuracy, integration, and data governance.
What extraction accuracy do you achieve on our specific submission types? How does your system handle the carrier-specific loss run formats we receive from our largest markets? Does the extracted data go directly into our existing pricing model, or do we need to rebuild our workflow? Does your system train on our loss data in a way that could leak to competitors?
These are the right questions. They reflect an operational sophistication about what automation actually involves — not the 2019-era "will AI replace underwriters" framing that was never the actual value proposition.
The submission bottleneck is structural. The volume growth in E&S is structural. The technical capability to address it with meaningful accuracy is now real. That combination is what makes this a different moment than the false starts of the previous decade.
