Shipping, customs, and insurance claims share one document problem: high volumes of mixed, variable operational paperwork arriving from countless carriers, brokers, and claimants in inconsistent layouts.
Bills of lading, commercial invoices, customs declarations, loss notices, damage estimates, and claims invoices all turn up in the same queue, and template-based tools break on that variance. LandingAI Agentic Document Extraction (ADE) separates mixed bundles, reads high-variance layouts without templates, and returns structured data grounded to the source for verification.
The Shared Problem
Every shipment or claim arrives as a bundle of different document types, in different formats, from different senders. A template tuned to one carrier or claimant breaks when the next uses a different layout, and volume compounds the maintenance cost rather than amortising it.
The requirement is the same in both domains: separate the bundle, read each document regardless of layout, and return verifiable structured data at operational volume.
Separating Mixed Document Bundles
A single shipping packet or claim file often holds several documents inside one PDF, which a field-by-field extractor cannot handle on its own.
- Split runs after parsing and separates the file into classified sub-documents, so a bill of lading, commercial invoice, and packing list in one PDF each route to the right extraction schema instead of being flattened together.
- You define the document types to look for, and pages Split cannot place come back as Uncategorized rather than being forced into a category.
- Where a bundle holds several instances of the same type, an identifier such as invoice number, order ID, or date creates a separate split per unique value, which is the common shape of a freight bundle or a claims file.
- Classify labels each page by type to route pages before parsing, when you need triage earlier in the pipeline.
Both run on the ADE v1 API and are in Preview. The Split documentation states directly that the feature is still in development, may not return accurate results, and should not be used in production environments. Treat bundle separation as a pilot capability and keep a manual or rules-based fallback on the critical path until it reaches general availability.
Reading High-Variance Layouts Without Templates
- Schema-driven extraction defines what to extract rather than where to find it, so one schema handles commercial invoices from any exporter or loss notices from any claimant.
- Dense multi-page tables such as line items, packing lists, and damage estimates keep their structure through layout-aware parsing, with every table cell grounded individually.
- Multi-language content, scans, photos, and the stamps and signatures common in customs and claims paperwork run through one pipeline, with signature and stamp regions classified as their own grounded elements.
Verifiable Data for Customs and Claims
Every value carries a trace back to where it came from, which is what lets a customs broker or claims adjuster check a figure without re-reading the file.
- Parsing grounds every element with its page and a bounding box on that page, and grounds each table cell independently.
- Extraction ties every value to the span of source text it was read from, and marks values the model inferred rather than copied.
- Fields route to review on that grounding and on your own business validation rules, so the routing logic stays explicit and auditable.
- A field ADE cannot find comes back empty rather than guessed, giving the pipeline a second signal distinct from a located value.
That chain supports customs and trade compliance and defensible claims decisions backed by a complete audit trail.
Shipping, Customs, and Logistics
- Bills of lading and shipping documents: shipment details, carrier information, and cargo data.
- Customs and trade compliance: required fields from customs declarations and trade documents to speed clearance, as part of LandingAI's logistics workflows.
- Freight invoices and proof of delivery: billing data and delivery confirmation passed into downstream systems.
Insurance Claims
- First Notice of Loss and investigative reports: claimant, policy, and incident details.
- Repair estimates, receipts, and claims invoices: line-item costs and totals for adjudication.
- Subrogation and recovery: assembling title, lien, and settlement evidence into structured case files, as part of LandingAI's insurance claims workflows.
Scale and Integration
- Asynchronous processing handles high-volume operational pipelines, with rate limits that scale by plan tier and are customizable on Enterprise plans.
- A REST API with Python and TypeScript libraries, plus a Snowflake Native App, delivers structured output into transportation management, claims, and case systems.
- SOC 2 Type II, GDPR, and a Zero Data Retention option are documented on the Security and Compliance page. Confirm current status for your requirements through the Trust Center.
- Test against your own bundles in the Playground before any integration work.
FAQ
How does ADE handle a bundle with several document types in one file?
Split runs after parsing and separates the file into classified sub-documents, and Classify labels pages by type before parsing, so a bill of lading, invoice, or repair estimate each route to the right schema. Both run on the v1 API in Preview, and the Split documentation advises against production use while the feature is still in development, so keep a fallback in place for now.
How does ADE cope with so many carrier and claimant formats?
Extraction is schema-driven and template-free. The schema names what to extract rather than where it sits on the page, so one schema handles a document type across many issuers without a rebuild when a format shifts.
How are extracted values verified for customs or claims decisions?
Each value traces back through the span of source text it was read from to the page and region that text occupies, so a broker or adjuster can follow any figure to its origin. That provenance is what makes customs compliance and claims decisions defensible in an audit.
Can ADE handle high operational volume?
Yes. Asynchronous processing supports high-volume pipelines, and rate limits scale by plan tier and are customizable on Enterprise plans, so throughput does not depend on manual batching.
When is ADE not the right fit?
For a single fixed-template form processed at high volume with no audit requirement, a cheaper template-based tool may be more cost-efficient. ADE is built for mixed, variable operational documents where layouts change constantly and every value has to be verifiable.