Deterministic submission compiler for E&S MGAs that turns broker packets into appetite checks and referral memos.

1. Deterministic workflow architectures have now shown they can outperform autonomous agents in high-stakes enterprise procedures.
2. Named financial and insurance customers have already bought when process adherence became measurable, reducing category risk for adjacent insurer workflows.
3. Underwriting is explicitly identified as a rule-dense procedure full of undocumented logic, making it a natural early wedge for execution-first AI.
4. Eight-figure ARR, profitability, and perfect pilot-to-production conversion imply reliable workflow software is crossing from experiment to budgeted rollout.

Catalyst. Poetic's proof that deterministic workflow software can reach 99% accuracy and 100% process adherence in financial and insurance operations makes insurer back offices newly ready to buy execution-first automation.

Build an underwriting workbench that sits on top of submission inboxes, broker portals, and core policy systems. The product ingests broker packets, loss runs, prior referrals, and carrier guidelines, then compiles each program's rules into a deterministic decision graph that classifies submissions, requests missing evidence, and drafts referral memos with cited facts. Human underwriters approve or override the recommendation, while every exception becomes structured rule feedback instead of another hidden tribal-memory note. Teams start with one program where quote backlog is painful, prove faster turnaround and cleaner appetite enforcement, then roll the compiler across additional carrier programs and renewal workflows.

What's different. This is not another underwriting copilot, BPO, or rules-only workbench. The product's advantage is a compiler layer that turns messy carrier guidelines and referral practices into executable, auditable workflows with human approvals and exception capture built in. That creates a growing moat in program-specific decision graphs, override history, and integrations across the systems MGAs already use.

Jobs to be done

flowchart LR
  Buyer[Chief Underwriting Officer] --> Pain[Submission backlog and inconsistent appetite decisions]
  Pain --> Product[Deterministic underwriting playbook compiler]
  Product --> Outcome[Faster quote turnaround and cleaner referral audit trails]

Executive takeaways

• The best initial market is not generic “AI for underwriting” but high-volume small-commercial E&S MGAs: MGA premium estimates moved from $80B+ in 2023 to $114B+ in 2024, while delegated-underwriting and E&S channels kept growing faster than the broader P&C market. [3][4][5]
• The operational pain is acute and measurable: multiple sources place 30%-70% of underwriter time outside core risk work, and buyers increasingly lose business when first-touch and quote turnaround lag. [17][18][19][20]
• Competitive intensity is high around intake, triage, and underwriting workbenches, but the reviewed players mostly optimize ingestion, prioritization, or platform breadth rather than deterministic translation of carrier playbooks into executable referral logic. [21][23][25][27][30]
• A credible year-3 foothold exists if the startup lands program by program, proves faster turnaround plus cleaner appetite enforcement quickly, and only then expands into renewals, delegated-authority QA, and adjacent workflows. [21][25][28][30]

Market definition

An auditable underwriting-compiler layer for U.S. small-commercial E&S MGAs: software that ingests broker submissions, maps them to carrier and program rules, and produces deterministic triage, missing-information requests, and cited referral memos before final human approval. [8][9][22][26][31]

Customer and buyer

Primary users are line underwriters and underwriting-operations managers at E&S MGAs handling dense submission inboxes; the economic buyer is usually the CUO or COO when growth, hiring pressure, or a new program makes turnaround and consistency board-level issues. [3][17][18][20][25]

Buying triggers

• Submission surges during hard-market, catastrophe, or renewal spikes expose shared-inbox triage delays and off-appetite waste. [5][18][19]
• Launching or expanding a delegated-authority program increases pressure to encode carrier rules consistently and prove oversight to partners. [7][9][26]
• Leadership wants more quote capacity without adding underwriter or admin headcount. [17][20][21]

Willingness to pay

Budget is likely justified from recovered capacity and faster win conversion rather than AI novelty: buyers already fund platforms that cut document-handling time, slash cycle times, or materially improve underwriting productivity, which supports low- to mid-five-figure annual spend per active program after ROI proof. [18][21][23][28][30]

Category dynamics

Growth signal 16.0% YoY (U.S. MGA direct premiums, 2024 estimate)

Validation signals

• Poetic and its customer references show deterministic workflow automation can cross 99% accuracy in high-stakes insurance-adjacent work.
• Markel reported a 113% productivity uplift after deploying Cytora.
• Velocity Risk used Federato to reduce time to quote by 89%.
• Allstate cut underwriting and renewal document review from seven days to under 15 minutes with Indico.
• Paragon doubled quote-to-bind and achieved 99% submission accuracy with Kalepa.

Regulatory & technical constraints

• AI-supported underwriting decisions must comply with applicable insurance law and keep governance, documentation, and unfair-discrimination controls examiner-ready.
• Insurance data-security obligations require an information security program, third-party oversight, and cyber incident reporting discipline.
• Surplus-lines and delegated-authority workflows require clear reason codes, exception handling, and carrier oversight rather than opaque model outputs.
• Submission ingestion must reconcile ACORD forms with messy email attachments, SOVs, and loss runs while preserving human exception handling.

Competition is dense around intake, triage, workbenches, and delegated-underwriting platforms. Cytora, Federato, Send, Convr, Kalepa, and adjacent intake vendors like Indico each solve meaningful parts of the workflow, but the evidence base still clusters around better ingestion, prioritization, or platform breadth—not deterministic translation of carrier playbooks into executable referral logic with audit-ready memo output. [21][23][25][27][28][30]

Why incumbents do not win by default

• Underwriting workbenches. Cytora and Kalepa improve intake and prioritization, but they do not clearly own program-by-program referral memo logic across carrier appetite changes.
• Delegated-underwriting platforms. Send addresses broad delegated-underwriting operations and compliance, yet its value proposition is platform breadth rather than compiler-like capture of tacit underwriter logic.
• AI intake and orchestration layers. Convr and Indico help structure submissions and context, but underwriters still need a deterministic layer that turns program rules into executable decisions and referral packets.
• Portfolio steering and prioritization tools. Federato is strong at winnability and appetite prioritization, but that is adjacent to codifying every referral threshold, missing-information rule, and memo template.

This company should start with a deterministic submission compiler for U.S. small-commercial E&S MGAs, where 10-50 underwriters still triage broker emails, PDFs, and carrier guides by hand. The urgent pain is not summarization; it is deciding faster without violating carrier appetite, creating off-appetite touches, or asking senior underwriters to write the same referral memo repeatedly. The first product should ingest email, PDFs, ACORD forms, and loss runs for one carrier program, classify each submission against explicit rules, request missing evidence, and draft cited referral memos for human approval. The first sale should be a paid historical replay plus a 60-day live pilot at an MGA with 1,000-plus monthly submissions, triggered by backlog spikes, a new delegated-authority program, or a mandate to improve turnaround without hiring. Research supports a roughly $75.0M beachhead SAM and a $2.5M year-3 SOM, but the venture case depends on proving one-program deployments can expand into additional programs, renewals, delegated-authority QA, and selected carrier workflows. The company can win because adjacent vendors mostly improve ingestion, prioritization, or workflow breadth, while this wedge centers on deterministic rule compilation, cited outputs, and auditable override capture. The biggest disconfirming risk is that each carrier program requires too much bespoke rule mapping or deep PAS integration, which would collapse software margins into services work. The first 6 months therefore need to prove replay accuracy, export-only deployment viability, and pricing acceptance before the company adds a broad sales team or adjacent workflow modules.

Problem

• High-volume E&S MGAs still work broker submissions from shared inboxes, PDFs, spreadsheets, and prior examples, so quote turnaround slows and off-appetite risks consume scarce underwriter time.
• Generic AI copilots can draft text, but they do not reliably encode carrier appetite, referral thresholds, change control, and audit-ready reasoning in a way CUOs will trust.

Solution

• Compile carrier guidelines, referral rules, and missing-document logic into a deterministic decision graph that turns each submission into accept, decline, request-more-info, or refer recommendations with cited facts.
• Give underwriters a workbench that surfaces missing evidence, drafts referral memos, logs every override, and converts exceptions into reusable rule updates instead of hidden tribal knowledge.

Why we win

• The product sells into a narrow, budget-worthy workflow where backlog, quote speed, and carrier oversight are already measured program by program.
• Deterministic rule compilation plus cited memo output is a clearer differentiation than generic intake or prioritization features bundled into broader underwriting platforms.
• Program-specific decision graphs, override history, and rule-change logs can compound into an operational moat that is hard for services firms or incumbent point features to recreate.

Milestones

flowchart LR
  Wedge[One carrier-program submission triage wedge] --> MVP[Deterministic compiler MVP]
  MVP --> Proof[Faster quote turnaround plus cited referral memos]
  Proof --> Expansion[More programs then renewals and carrier oversight]

Founding team

Experiment roadmap

Risk assessment

What must be true

• At least 25 beachhead MGAs must have one program with enough submission density and backlog pain to justify roughly $50k annual spend.
• Historical replays must show at least 90% agreement on scoped triage logic and at least 30% manual triage time reduction before live deployment.
• Email, PDF, and ACORD ingestion plus memo export must be enough to start at least half of early pilots without PAS write-back.
• At least half of production customers must expand from one live program to a second within 9-12 months.
• Buyers must prefer a deterministic compiler overlay to bundled workbench or intake features in multiple competitive evaluations.

Open diligence questions

• Which classes and carrier programs create the highest recurring referral and missing-information volume?
• Which budget line funds the purchase in practice: underwriting technology, operations productivity, or carrier-program support?
• How frequently do carrier guidelines change, and who owns rule updates and certification inside the MGA?
• Can design partners start with export-only deployment, or does PAS or workbench write-back block adoption?
• Why will MGAs buy this overlay instead of extending Cytora, Federato, Send, Convr, Kalepa, or internal ops tooling?

Model sanity

• Revenue engine. Base-case revenue comes from growing active live programs from 3 at Y1 exit to 45 at Y3 exit at a $55K blended program-year value, with second-program expansion and partners doing most of the work after Y2.
• Must go right. The model needs export-first deployments and reusable rule templates to keep first-program launches under 45 days and hold gross margin at 70% while one solutions engineer supports early scale.
• Model breaks if. If ARPU falls toward $50K or pilot-to-production stretches toward 6-7 months, cash compresses toward the downside case low point of roughly $157K and the seed round likely needs extension capital.
• Next-round proof. A credible Series A setup is 8-10 production MGAs, 15-20 live programs, one reusable PAS/workbench connector, and two active distribution partners before the cash buffer drops below about six months.

Scenarios

Sensitivity

flowchart LR
  ReplayDemand --> PaidPilots
  PaidPilots --> LivePrograms
  LivePrograms --> SecondPrograms
  SecondPrograms --> Revenue
  Revenue --> GrossProfit
  GrossProfit --> Cash

Flags: The base case still requires the company to scale from 16 to 45 live programs in Y3, so second-program expansion and partner channels must replace pure founder-led selling. · The $55K blended program-year assumes some setup and volume-band revenue, so recurring subscription ARR is modestly lower than total reported revenue. · Cash bottoms near $0.6M in the base case, leaving limited room if compliance review or connector work pushes sales cycles toward the downside case. · Revenue per FTE reaches only about $233K by Y3, which is acceptable for a services-assisted wedge but below best-in-class SaaS efficiency.

• Implementation sprawl. Every carrier program has its own appetite quirks, which can turn the business into custom services. Mitigation: Start with one high-volume class and ship reusable authoring tools plus templates for the first 10-20 common rule patterns.
• Incumbent workflow capture. Underwriting-workbench or policy-admin vendors could add basic AI triage features and bundle them. Mitigation: Stay system-neutral and go deeper on rule compilation, exception feedback, and cross-program audit trails than incumbent point features can.
• Rule drift blind spots. If carrier guidelines change faster than the workflow graph is updated, the product could create false confidence. Mitigation: Require human approval on referral decisions, add change-management alerts, and treat every override as a monitored rule-update event.