Insurance Underwriting Intake Workflow: Submissions to Decision-Ready Files
Underwriters spend most of their day assembling files, not underwriting. The workflow takes broker submissions as they arrive — emails with ACORD forms, loss runs, schedules of values, spreadsheets — extracts the data, checks completeness, pulls third-party data, runs appetite and rating rules, and delivers a decision-ready file with the open questions listed. Clear-cut declines are answered the same day; underwriters spend their time on the risks that need judgement.
Written by Max Zeshut
Founder at Agentmelt · Last updated Sep 11, 2026
The problem
Submissions arrive as inconsistent attachments. Underwriting assistants re-key data into the policy system, chase brokers for missing information, and out-of-appetite risks sit in the queue for a week before being declined. Quote turnaround is the reason brokers send business elsewhere.
What changes when it runs
Every submission is triaged within an hour: out-of-appetite declined with a reason, incomplete ones bounced to the broker with a precise list, and the rest in the underwriter's queue with the data in the system, third-party checks done and the questions to ask. Quote turnaround and hit ratio both improve because the underwriter starts with a file, not a pile.
Trigger, then 8 steps
Trigger
Submissions inbox (Outlook / Gmail) + broker portal webhook
New emails to submissions@ with attachments; portal submissions arrive as webhooks.
Capture and classify attachments
Gmail TriggerEach attachment is classified — ACORD application, loss runs, schedule of values, financials, supplemental questionnaire, other — and stored against a new submission record.
Extract application data
Information ExtractorInsured details, operations description, revenue, employees, locations with construction/occupancy/protection data, requested limits and coverages, prior carrier and premium, loss history by year — with confidence per field.
Check completeness
CodeRequired fields per line of business; missing items generate a broker request drafted by the agent with the exact list, sent after a human click or automatically below a size threshold.
Pull third-party data
HTTP RequestProperty data (construction, roof age, flood zone), business verification, sanctions and OFAC checks, and loss-control reports where available; results attached to the file.
Apply appetite rules
IFClass of business, geography, revenue band, loss ratio and limits against the appetite guide. Clear declines are drafted with the reason; borderline cases are flagged for the underwriter with the rule that triggered.
Indicative rating
CodeFor in-appetite risks, an indicative premium from the rating algorithm or rater API, with the factors listed so the underwriter sees what drove it.
Assemble the file and questions
AI AgentThe agent writes the submission summary — the risk in one paragraph, key exposures, loss history narrative, third-party findings, indicative rating — and the specific questions an underwriter would ask this broker about this account.
Queue and track
HTTP RequestSubmission created in the policy admin or underwriting workbench with the data and documents; the underwriter is assigned by line and capacity; turnaround is tracked from receipt to quote.
Data it touches
- Submissions inbox and broker portal
- ACORD forms, loss runs, SOVs, financials (attachments)
- Third-party data (property, business verification, sanctions)
- Appetite guide and rating rules
- Policy admin / underwriting workbench
Guardrails
- No binding decisions are automated; the workflow triages and prepares. Declines outside appetite are drafted and sent after a human click unless the carrier chooses to automate below a threshold.
- Every extracted field carries confidence; low-confidence fields are highlighted for verification before rating.
- Appetite and rating rules are versioned and owned by underwriting management, not embedded in prompts.
- Sanctions screening is mandatory and blocks progress on a hit.
Automate intake and triage, not the underwriting decision
The underwriting judgement — is this risk priced right, what conditions, what capacity — stays with the underwriter. What the workflow removes is the two to four hours per submission spent turning attachments into a file: classifying documents, keying data, chasing missing pieces, running the same checks. That work is mechanical and it is exactly what delays quotes. Carriers that automate intake typically cut submission-to-quote time by days and free underwriters to look at more risks, which is where the hit-ratio improvement comes from.
Appetite rules as code, owned by underwriting
The appetite guide lives in a versioned rules table — class codes, states, revenue bands, loss-ratio thresholds, limit caps — that underwriting management edits. The workflow applies it and reports which rule fired. That is more transparent than the current state at most carriers, where appetite is partly in a PDF and partly in senior underwriters' heads, and it makes the fast decline honest: the broker gets the reason.
The summary and the questions
The agent's summary is the page an underwriter would write for themselves before starting: what the business does, the exposures that matter for this line, the loss history as a story rather than a table, what third-party data confirmed or contradicted, and the indicative premium with its drivers. The questions list — 'confirm sprinkler status at location 2', 'explain the 2024 liability claim' — goes to the broker in the same message as the acknowledgement, so answers arrive before the underwriter opens the file.
Tools in the stack
| Tool | Role in this workflow |
|---|---|
| n8n | Intake, extraction pipeline, rules, routing |
| Claude | Attachment classification, data extraction, summaries, broker requests |
| Policy admin / workbench (Duck Creek, Guidewire, Applied, or in-house) | System of record |
| Third-party data providers | Property, verification, sanctions |
| Outlook / Gmail | Submissions and broker correspondence |
Want this running without building it?
Automation workflow
$297/month
We set up, host and maintain this workflow on n8n and connect it to your tools. Setup included, cancel monthly, you keep the JSON.
Custom build
$8,000–15,000 one-time
Your systems, your rules, your edge cases. A one-off build on Claude and n8n, delivered with documentation and a walkthrough.
The subscription covers a single line of business with standard ACORD forms and up to 300 submissions a month. Multi-line appetite logic, rater integration and policy-admin integration are a custom build; carriers and MGAs usually need the custom path.
Frequently asked questions
Does it make underwriting decisions?
No. It triages against your appetite rules, prepares the file and drafts declines and requests. Binding, pricing and conditions remain with the underwriter. Automated declines below a threshold are optional and configured by you.
Which policy systems does it integrate with?
Any with an API — Duck Creek, Guidewire, Applied Epic, Majesco and in-house systems — as part of the custom build. The subscription version delivers the file to a queue (email, Slack or a sheet) without a system integration.
How does it handle loss runs from different carriers?
Loss-run extraction normalises different carrier formats into a standard claims table (date, cause, paid, reserved, status) with confidence scores; unusual formats are flagged for review.
Case study
AI Insurance Agent for a Regional Carrier: 45% Faster Claims Processing
How a regional P&C carrier used an AI insurance agent to automate FNOL intake and simple claims adjudication—cutting average claims cycle time by 45%.
The pillar
AI Insurance Agent
Automate claims processing, underwriting, policy management, and customer communication for insurance.