Submission Ingestion & Clearance
Read a multi-line commercial submission and hand the underwriter a prepared package overview — clearance, appetite, and completeness already assessed.
The business problem
A commercial submission arrives as a stack of separately-filed documents — an application, coverage-line supplements, schedules, loss runs, an SOV. Before pricing can start, someone has to read all of it, confirm what's being requested, check clearance and appetite, and flag what's missing. Across fifteen-plus document types, that first read is real time spent before the underwriting work begins.
The Submission Ingestion & Clearance template automates it: classifies and extracts every document, synthesizes package-level clearance, appetite, completeness, risk, and exposure, and routes every submission to an underwriter with the analysis prepared.
The analysis is decision support, not a decision. The agent lays out findings and evidence; it doesn't clear, decline, or bind. An underwriter reviews, can override anything, and approves before the item is finalized.
Trigger and source data
The template ships with Manual File Upload. Other triggers can be configured; inbound webhooks, scheduled/timer triggers, and SFTP are not currently available. To start runs from another system, use the Bevaya API. See Configure triggers.
Upload a submission package — typically pre-split, one form or schedule per file — containing one or more of:
- A main application
- One or more ACORD-style coverage-line supplements (General Liability, Auto, Workers Compensation, Umbrella, Property, Inland Marine)
- A vehicle or driver schedule
- Loss runs
- A statement of values (SOV)
- A supplemental application or Emod worksheet
- Correspondence, if any
The flow works entirely from the submitted documents — it doesn't look up policy or claim data from a system of record. A cover sheet or title page whose header merely names a line of business, with no actual coverage data, classifies as "Other" rather than that line's supplement type.
Template starting point
Create an AI agent from the Submission Ingestion & Clearance template for a working flow out of the box. Open it in the canvas, select an environment, click Start Editing, and tune it to your lines of business and standards — field prompts, the submission-insights prompt, validation thresholds. See Create an AI Agent.
What the template preconfigures
| Step in the flow | Node it uses | Category |
|---|---|---|
| Prepare the uploaded files | Read Files | Utility |
| Process each document in the submission | For Loop | Control |
| Classify each document and extract per-type fields | InsurGPT: Custom | InsurGPT |
| Route each document to the right extractor | Switch | Control |
| Generate package-level submission insights (clearance, appetite, overview) | InsurGPT: Custom | InsurGPT |
| Summarize coverage lines, risk profile, and exposure | InsurGPT: Custom | InsurGPT |
| Check extracted and analyzed values against rules | Field Validation | Utility |
| Route the prepared analysis to an underwriter | Review | Utility |
| Close out the item | Complete | Action |
The summary step produces coverage-line detail, a completeness check with broker follow-ups, risk and submission flags, an exposure summary (payroll, revenue, loss history), a clearance check, and an appetite triage.
What builders must configure before deployment
- Extraction, analysis, and validation. Tune classification and extraction fields, the submission-insights prompts, and validation rules to your lines of business and appetite.
- Validation thresholds and review routing. Set the confidence threshold and reviewer assignment. See Utility nodes.
How the AI agent behaves
The agent classifies the submission and each document, extracts fields across all recognized document types, and synthesizes package-level insights: lines requested, clearance, appetite, and what's missing. Each finding — clearance, appetite, completeness — carries a plain status token and supporting evidence rather than a bare verdict.
The agent works only from the supplied documents and doesn't invent values. A document that self-discloses its own incompleteness (a schedule stating units were omitted) is treated as a stated fact, not an inference. Low-confidence or missing fields are flagged by validation, never fabricated. The agent assembles and explains; the underwriter clears, prices, and binds.
Human review model
The underwriter reviews the item in the Bevaya Platform, across two tabs:
- Insights — the decision-ready briefing: an exception summary for hard stops, an AI-recommended next step, key-insights cards (overview, clearance, appetite), a lines-of-business breakdown, a completeness breakdown with broker follow-ups, a risk profile, an exposure summary, source documents, and an audit trail.
- Review — field-by-field verification (insured identity, lines requested, effective date, every extracted fact) with source document and page, where the underwriter confirms, edits, and approves.
Every submission pauses at Review until the underwriter submits. Reviewers can work items and reviews but can't build or run flows. See Human review.
Run status and reporting
Each submission is an item moving through In Progress, Review, Complete, Failed, Canceled. Watch runs in Run history; see live counts by status in Item status reporting.
Example run
A trucking and cross-dock logistics operator submits a main application, an Auto Supplement with a vehicle schedule, a WC Supplement, correspondence, and loss runs.
- The flow ingests the package and prepares the files.
- It classifies each document and extracts the application, vehicle schedule, WC supplement, and loss run fields.
- Submission insights confirm the lines requested and run the clearance check and appetite triage.
- The summary step produces risk and submission flags, an exposure summary, and a completeness check against what the broker has and hasn't supplied.
- Validation checks fields against the confidence threshold.
- The underwriter reads the Insights briefing, confirms the detail on Review, and approves.
Common failure modes
- A cover sheet misclassified as a coverage-line supplement. A title page naming a line of business but carrying no coverage data should classify as "Other" — confirm your classification prompt guards against this.
- Low-confidence or missing fields. Poor scans or incomplete packages produce unconfirmed values; validation flags them for review rather than fabricating.
- A genuinely incomplete submission read as merely low-confidence. A schedule that states its own excerpting ("29 units omitted") is a fact the agent should check for, distinct from softer confidence flags.
- A misclassified document. The underwriter corrects it during review and resubmits so extraction re-runs for the right type; fix persistent patterns in the classification prompt.
Recommended rollout path
- Build and test on sample submissions spanning your lines of business with Run Draft, inspecting each step. See Test and debug a draft.
- Validate the clearance and appetite triage against known submissions.
- Pilot with every submission routing to review while you build trust in the findings.
- Promote to production once accuracy is proven out. See Drafts and publishing.
- Keep the human in the loop — the underwriter review gate is the safety control.
Where to go next
- Utility nodes — Field Validation and Review.
- InsurGPT nodes — classification, extraction, and submission insights.
- Human review — the Insights and Review tabs.
- Item status reporting — track submissions by status.
- Use cases overview — the full catalog of supported patterns.