How Can Insurance Brokers Use AI for Underwriting Assist on Mid-Market Risks?

Most mid-market insurance brokerages know the same hard truth. Submission package quality is the single biggest predictor of placement success. The best producers write submissions that underwriters open first, take to authority faster, and quote at preferred terms. Newer producers write submissions that get queued, get partial quotes, and come back with a list of follow-up questions that take a week to answer. Same risk. Same market. Different outcomes, often by 15 to 25 percent in premium and 30 to 40 percent in placement rate.

The gap is not underwriting knowledge. It is the time the producer has to assemble a submission that tells the risk's story well. A senior producer can write a strong narrative in 90 minutes. A three-year producer needs four to six hours and the result is still 70 percent of what the senior producer would write. The brokerage cannot scale by hiring more senior producers because they do not exist in sufficient supply.

AI solves this by handling the assembly and drafting work between the data in the AMS and the polished narrative the underwriter sees. The producer still owns the underwriter relationship and decides which markets to approach. The AI just means a three-year producer can produce an eighteen-year producer's package in 90 minutes instead of six hours.

This guide walks through the underwriting assist workflow that holds up under state insurance regulations, E&O exposure, and modern carrier expectations. It covers the four sections AI handles well, the one it does not, and the workflow that protects the producer's license while compressing assembly time.

Why this matters for mid-market insurance brokerages specifically

Mid-market brokerages (10 to 75 producers, $5M to $50M in revenue, accounts in the $10K to $500K premium range) are uniquely under-served by current AI tooling. The mega-brokerages built internal tools they do not share. The InsurTech vendors (Indio, Insurmi, Federato) are real but expensive and built for retail or carrier-side use. AMS-bundled AI features are mostly limited to Acord auto-fill and email summarization. The actual submission narrative work is still manual.

The brokerages that figure out the workflow get hours back per producer per submission, place a higher percentage of their submissions, and produce a more consistent client experience across producers of different tenure. They also build an audit trail that holds up under E&O claim and regulatory inquiry. Brokerages that wait usually end up dealing with the producer development problem and the placement rate problem in the same year.

What AI for underwriting assist actually does

The useful tools are general-purpose large language models (Claude, ChatGPT, Microsoft Copilot) running on Team or Enterprise tier with a DPA and training on inputs disabled. They take inputs (Acord application data, loss runs, prior policies, operations descriptions) and produce structured submission packages: narrative cover letters, risk improvement summaries, market lists with rationale, and underwriter follow-up question responses.

Three things make these tools different from the AI features bundled into your AMS:

• They handle long-form drafting consistently. AMS-bundled features summarize Acord fields. The general-purpose tools take a 30-page application, five years of loss runs, and a declarations page and produce a 1,500-word narrative that reads like a senior producer wrote it.
• They produce output in the brokerage's voice and the underwriter's preferred format. Each major underwriter has preferences. The AI matches the format to the market.
• They iterate in plain English. The producer can say "add a paragraph on the new safety program, lead with the loss-frequency improvement, drop the section on building age" and the AI updates.

Think of it as a senior associate underwriter at the brokerage's desk who has read every account file, knows the markets cold, and produces narratives in the format each underwriter prefers. The producer still owns every market decision and coverage recommendation.

Before you start

You need:

• A Claude Team, ChatGPT Team or Enterprise, or Microsoft Copilot for Microsoft 365 account with training-on-inputs disabled and a signed DPA. $25 to $60 per seat per month.
• About 60 minutes for your first session, mostly to build the brokerage's prompt template library.
• One real submission you are working on or one renewal you are preparing in the next two weeks. The workflow is not abstract. You will be doing real submission work.
• Read access to the brokerage's AMS (Applied Epic, Vertafore AMS360, Sagitta, EZLynx, or whatever you run) and the markets and contacts list.
• Brokerage management and the brokerage's E&O contact on a 30-minute call to confirm the workflow is covered before any client account uses it.

One thing to settle before any insured data goes into an AI tool: state insurance regulations, the NAIC Insurance Data Security Model Law as adopted in your state, the brokerage's E&O exposure, and the brokerage's information security program all shape what AI can and cannot do here. We have a dedicated section on this below. It is non-negotiable.

Material 1: Submission narrative cover letters

The failure pattern: a producer drafts a submission cover letter that lists the operations, the years in business, the locations, and the limit needed. It is factual and complete and does not give the underwriter a reason to write the account ahead of the other 30 submissions in the queue. Underwriters quote it last, at standard terms, because there is no story.

What to ask the AI for instead:

Below are five inputs: (1) the Acord 125 commercial general application for [Insured Name], (2) the operations description from the insured, (3) the five-year loss runs from the prior carriers, (4) a list of risk improvements the insured has made in the last 24 months, and (5) a one-paragraph note from the producer on what makes this risk better than the average risk in its class. Draft a 1,200 to 1,500 word submission narrative cover letter for an underwriter at [Carrier Name], whose preferred format I will describe below. Structure: (1) Two-paragraph operations and ownership summary that gives the underwriter a quick picture, (2) Risk strengths section with three to five specific items the underwriter should know, (3) Loss history section that addresses the loss runs honestly without minimizing, including the cause of any large losses and the corrective action taken, (4) Risk improvements section with the changes made in the last 24 months and the operational impact, (5) Coverage and limit request with the producer's commentary. Do not characterize the risk with adjectives like premier, leading, or best-in-class. Lead with what is true. Address what is unfavorable honestly. Use the producer's voice (sample below).

The constraint that matters: "address what is unfavorable honestly." AI tools default to minimizing weaknesses. Underwriters notice. A submission that minimizes a known weakness loses credibility before the rest of the package is read. The honest treatment plus the corrective action is the move that earns underwriter trust.

For different lines (workers' compensation, commercial auto, property, professional liability), the structure stays the same and the section emphasis shifts. WC narratives lead with the experience modification and the safety program. Commercial auto narratives lead with the driver list, the MVRs, and the fleet management practices. Property narratives lead with the construction, occupancy, protection, and exposure. Professional liability narratives lead with the operations, the procedures, and the prior claims context.

Material 2: Market selection and rationale

The failure pattern: the producer marks 10 to 15 markets on every submission because that is the brokerage's standard practice. Half of those markets are wrong for the risk and the producer knows it but does not have time to think through which five would be right. The submission goes wide, gets partial responses, and the producer chases follow-up questions on accounts that were never going to write.

The AI version produces a narrowed market list with rationale, which the producer reviews and approves before any submission goes out.

Below are three inputs: (1) the operations and risk profile for [Insured Name], (2) the brokerage's standard market list for this class of business, and (3) the brokerage's recent placement history with each market on that list (which markets wrote, declined, or quoted last 12 months for similar accounts). Produce a narrowed market list of 5 to 8 markets the brokerage should approach for this account. For each market, write two sentences: (1) Why this market is a likely fit based on the risk profile and the brokerage's recent history, (2) What the producer should emphasize in the submission to that specific market. Mark any market that has declined a similar account in the last 12 months and explain why this account is different (or recommend not approaching them). Do not include markets the brokerage does not have an active appointment with.

The AI is not deciding the markets. The producer is. The AI is doing the assembly work that helps the producer think through the list. The producer reviews, edits, and approves. The submission goes out to a tighter, better-fit market list, which improves placement rate and reduces the time spent on follow-up questions for markets that were never going to write.

For specialty lines (cyber, EPL, D&O, environmental), the same pattern works with one addition: the AI flags any market that has tightened appetite or pulled out of the line in recent industry news, so the producer does not waste a market slot on a carrier that will not write the class.

Material 3: Underwriter follow-up question responses

The failure pattern: the underwriter sends a list of seven follow-up questions. The producer pastes them into an email to the insured, gets answers a week later, and tries to remember what the underwriter actually wanted with each question. The response email goes back to the underwriter with answers that technically address the questions but do not anticipate the underwriter's next question, so a second round of follow-up adds another week.

The AI version reads the underwriter's questions and the insured's answers and produces a response email that addresses the questions plus the underlying concern.

Below are three inputs: (1) the underwriter's seven follow-up questions on [Insured Name], (2) the insured's responses, (3) the original submission narrative I sent. Draft a response email to the underwriter that addresses each of the seven questions clearly and proactively addresses the likely follow-up concern behind each question. Format: brief intro, then numbered responses 1 through 7 with the underwriter's question restated and the answer. For each answer, lead with the factual response and follow with one sentence addressing the underwriting concern (loss control, exposure, frequency, severity, or whatever the question is really getting at). Use the producer's voice (sample below). Do not characterize the risk with comparative adjectives. Where the insured's answer does not fully resolve the question, mark that for the producer to follow up with the insured before sending the response.

The move that compresses cycle time: anticipating the underwriter's next question. A response that addresses the underlying concern often closes the loop in one round. A response that just answers the literal question opens a second round. The difference is one to two weeks on the placement timeline.

For renewal underwriting where the underwriter is asking about the prior year's claims activity, the AI version pulls the loss history into the response and frames the corrective actions. The underwriter gets the full picture in one response instead of three.

Material 4: Renewal review summaries and stewardship reports

Renewal review summaries and annual stewardship reports are the highest-volume client deliverable a brokerage produces. They go to the insured before each renewal to summarize the year, markets approached, placement decisions, and recommended changes. Most brokerages produce them in templates that get manually filled out in 90 to 180 minutes per account.

Below are four inputs: (1) prior policy information for [Insured Name] as of last renewal, (2) loss runs from the prior term, (3) markets approached at last renewal and the outcomes, (4) producer's notes from the year on coverage changes, exposure changes, or claims activity. Produce a 1,200 word renewal review and stewardship report: (1) Year in review, summarizing operations, exposure, and claims changes, (2) Markets approached and outcomes, factual summary, (3) Recommended renewal strategy with two to three specific recommendations and reasoning, (4) Coverage changes the producer suggests considering, with rationale, (5) Open items for the renewal meeting. Use the brokerage's voice (sample below). Do not project specific premium outcomes. Where the loss runs show meaningful activity, address it factually with cause and response.

The stewardship report becomes the basis for the renewal meeting. The producer walks the insured through it and makes adjustments based on the conversation. The AMS captures the document. The brokerage has a clean record of the renewal discussion, which matters for E&O defense if a coverage gap shows up later.

Material 5: Coverage gap analysis and exposure summaries

Coverage gap analysis is the highest-value advisory work a producer does and the most under-systematized. Senior producers do it well. Newer producers do not, because the analysis requires knowing which exposures to look for, which coverages to check, and which industry-specific risks to flag. That knowledge takes years.

Below are three inputs: (1) the operations description for [Insured Name], including industry, size, locations, and any unusual exposures, (2) the current insurance program with limits and forms, (3) any concerns the insured has raised in the last 12 months. Produce a coverage gap analysis with: (1) Standard exposures for this industry and size, with typical coverage approach, (2) Current program assessment with each line and any apparent gap or under-limit, (3) Industry-specific risks not addressed in the current program, (4) Items for the producer to discuss with the insured. Frame as items to discuss, not as definitive coverage advice. Do not promise specific premium outcomes.

The constraint that protects the producer: "items to discuss, not definitive coverage advice." The coverage decision is the producer's. That separation matters for E&O exposure and for state insurance regulation around coverage advice. For specialty exposures (cyber, EPL, D&O, environmental, professional liability), the AI version works particularly well because the senior producer's expertise is hard to codify and the AI captures it once in the prompt template.

Material 6: Producer development and submission training

The producer development program is the single highest-return investment a brokerage makes and the most under-systematized. Most brokerages run informal mentorship where a senior producer teaches a newer producer through observation. The pace is whatever the senior producer's calendar allows.

Below are five strong submission narratives from the brokerage's senior producers across different lines. Below them is a draft narrative I just wrote for [Insured Name]. Compare my draft to the senior producer patterns and produce a coaching summary: (1) What my draft does well, (2) What is missing or weak compared to the senior producer patterns, (3) Specific edits the senior producer would make and why, (4) The development pattern this illustrates. Be direct. Do not soften the feedback. The goal is to help me improve faster, not to make me feel good about a weak draft.

The coaching summary becomes a structured way for newer producers to develop submission writing without waiting for senior producer time. Senior producers spend time on cases that need their judgment, not on red-pencil edits. For onboarding, new producers write practice submissions on de-identified examples and develop patterns faster than they would through observation alone.

The broker-specific prompts that actually work

After watching mid-market brokerages run AI through a couple of renewal cycles, the difference between useful output and generic output comes down to four prompt moves.

Specify the line of business and the carrier. "Workers comp submission for a 75-employee manufacturer in Ohio with a 0.92 X-mod and a clean three-year loss history" lands very differently than "a workers comp submission." The AI calibrates the narrative emphasis to the actual risk and the carrier's preferences.

Specify the constraint that actually matters. For brokerages the constraints are: do not characterize the risk with comparative adjectives, do not minimize unfavorable items, do not project specific premium outcomes, frame coverage recommendations as items to discuss not as definitive advice. State them explicitly. The AI honors explicit constraints. It minimizes badly when the constraints are implicit.

Specify the producer voice and the underwriter format. Paste a sample of how the producer or the brokerage actually writes, and one of how the target underwriter has historically responded best. The AI matches both.

Specify what stays static and what changes. For recurring documents (renewal stewardship reports, market lists, follow-up question responses), tell the AI what is fixed (brokerage format, standard sections, producer sign-off) and what is variable (this account's situation, this market's appetite, this loss history). The structure stays the same. The content scales across the book.

The state insurance regulation and E&O non-negotiables

This section is short because the rules are simple, but it is the most important section in this guide.

Do not put any of the following into the consumer tier of any AI tool:

• Insured names paired with specific coverage details
• SSNs, EINs, or driver's license numbers
• Specific loss details that identify the insured (claimant names, settlement amounts paired with the insured's name)
• Health information from any application or claim file
• Financials, financial statements, or any document that could identify the insured by name and financial detail
• Anything that could identify an insured to a third party

Use AI for templates, prompt patterns, market list research on de-identified accounts, and submission narrative drafting on the Team or Enterprise tier with a Data Processing Addendum signed and training on inputs disabled. For client-identifiable data, the brokerage's information security program needs to address the AI workflow explicitly. Most state insurance regulators have adopted some version of the NAIC Insurance Data Security Model Law, which requires a written information security program, regular risk assessments, and notification of cybersecurity events.

State insurance departments are issuing AI guidance rapidly. Check the NAIC AI Working Group output and your domiciliary state's insurance department guidance. Some states (Colorado, New York, California) have specific rules around AI use in underwriting and claims that affect what brokers can and cannot do. The rules are evolving. Document the workflow and update the documentation as the rules update.

The brokerage's E&O carrier should be informed of the AI workflow. Most major E&O carriers have published guidance and most are comfortable with documented internal-use workflows. The carrier's confirmation in writing is the document you want in the file before any claim ever happens.

If your brokerage has signed an Anthropic Business agreement, an OpenAI Enterprise agreement, or a Microsoft 365 Copilot agreement with a DPA, the data-handling rules can be different. Ask your management and your IT team what is covered. Ask the E&O carrier separately about coverage. Document everything. Do not assume.

When NOT to use AI for underwriting assist

AI is a generalist tool. It will not be the right answer for every part of a brokerage's submission and renewal workflow.

Skip it for:

• Coverage recommendations to insureds. The producer makes the call. Always. The AI surfaces the question. The producer decides. The brokerage's recommendation goes to the insured in writing under the producer's signature.
• Direct insured-facing AI interactions about specific policies. Letting the AI draft an email is fine. Letting the AI answer specific coverage questions about a specific policy crosses state insurance regulation, E&O exposure, and duty-of-care lines.
• Claims advocacy and coverage litigation strategy. The claims work involves legal judgment and carrier negotiation. The AI can summarize a claims file. It cannot substitute for the claims advocate or coverage counsel.
• Anything that requires a producer's licensed judgment. Suitability, coverage adequacy, market access decisions. The producer decides.

A simple rule: AI is an unfair advantage on the 80% of submission and renewal work where the work is assembly and drafting. Trust the licensed producer's judgment for the 20% where the document or the recommendation has E&O or regulatory weight.

The quick-start template

Here is the prompt scaffold that works across most brokerage underwriting use cases. Copy it, fill in the brackets, paste into Claude Team or ChatGPT Team after the workflow addendum is in place.

Build me a [submission narrative / market list / follow-up question response / renewal stewardship report / coverage gap analysis / producer coaching summary] for [Insured Name] for the [renewal date / submission deadline].

Inputs: [paste Acord application data, loss runs, prior policies, operations description, producer notes, target carrier preferences].

Output structure: [list 3 to 6 named sections with one sentence on each].

Voice and format: [paste 2-3 paragraphs of how the producer or brokerage actually writes].

Constraints: Do not characterize the risk with comparative adjectives. Do not minimize unfavorable items. Do not project specific premium outcomes. Frame any coverage recommendation as an item to discuss with the producer's licensed judgment. Where the inputs are silent, mark blank rather than infer.

Output format: [draft email to underwriter, formal narrative for AMS attachment, internal coaching memo].

For recurring documents, save the populated template once and reuse the structure. The account-specific inputs change. The structure does not.

Bigger wins beyond submission drafting

Once the underwriting assist workflow is running, the next layer of value shows up in places that affect the brokerage's economics, not just per-submission throughput.

Market intelligence library. Capture the brokerage's recent placement history, each market's appetite signals, and the patterns of which markets quote which classes well. That document becomes the basis for every market list. New producers ramp faster.

Cross-sell opportunity surfacing. AI can read the existing book and surface accounts that are candidates for additional lines (GL-only insureds that should have cyber, property insureds that should have business income). The producer makes the call. The AI surfaces the candidates across thousands of accounts.

Producer onboarding acceleration. New producer ramp at a brokerage is typically 18 to 36 months. AI compresses the ramp by giving new producers the brokerage's submission patterns, market patterns, and coaching feedback from day one.

Renewal calendar load balancing. AI can pre-draft renewal stewardship reports for the next 90 days of renewals, so the producer walks into each renewal review meeting with the document already 70 percent done. Across a book of 200 renewals a year, that is hundreds of hours recovered.

The financial services AI consulting connection

This is one tool in one category of a brokerage's workflow. The bigger question for the financial services niche is structural: insurance buyer expectations are shifting toward faster placement and more transparent stewardship, while state insurance regulation around AI use is tightening. Brokerages that figure out the workflow end up with producers who get hours back per submission, placement rates that improve, and a documented compliance posture that holds up under E&O claim and regulatory inquiry.

If your brokerage is wrestling with the bigger AI question, the AI Consulting in Financial Services page covers the full scope: where AI actually fits in insurance brokerages, RIAs, wealth management firms, and mid-size CPA firms, what the common failure modes look like, and what an engagement looks like when it works.

For individual brokerage owners, start with this guide. Pick one of the six workflows and build the prompt template this week. Run it on three real submissions next week. The case for the rest follows from the throughput change.

Closing

The goal is not for producers to become prompt engineers. It is for producers to spend their judgment hours on relationships, market access, and coverage decisions instead of on document assembly. AI is the closest tool I have seen to that goal for mid-market brokerages specifically. It rewards specificity and respects the regulatory frame.

Pick one workflow. Build the template. Run it on three submissions this month. See the throughput change.

If you want to talk about how AI fits into your brokerage at the practice level, the AI Consulting in Financial Services page lays out the full picture and how an engagement works.