Trade Document Extractor

Extracts fields from invoices, bills of lading and LC documents.

ConfidentialCustomer PIIThird-partyLimited riskDocument AI · 2.2.0 · trade-document-extractor

Test in Sandbox

58 23K/30d 44

Field-F1 (invoices)

0.954

Field-F1 (bills of lading)

0.931

Discrepancy detection recall

0.88

Straight-through rate

62%

Median latency

1.9 s / page

About this model

Extracts fields from trade documents submitted for letters of credit and trade financing: commercial invoices, bills of lading, packing lists and LC applications. Returns applicant, beneficiary, goods description, HS code, quantities, amounts, ports and dates with per-field confidence, and flags discrepancies between documents in the same set.

Intended use

Straight-through data capture and first-pass document checking for Trade Operations. Checkers confirm every field below 0.90 confidence and every flagged discrepancy before the LC is processed.

Training data lineage

Lumen Risk Labs' document model, licensed and fine-tuned inside the Bank's Google Cloud project on 18,000 historical trade document sets linked to trade-finance-transactions in BigQuery. Documents never leave the Malaysia region. Artefacts registered in the Bank's Vertex AI model registry; vendor assessed under the Bank's third-party risk policy.

Limitations & bias notes

Handwritten annotations and stamped amendments are not read. Documents in languages other than English are routed to manual processing. Scanned faxes below 150 dpi lose about 8 points of field-F1.

#document-ai#trade-finance#letters-of-credit#invoices#bills-of-lading#vendor

Ownership and sensitivity

Owned by

Lumen Risk Labs Lumen Risk Labs delivery team

Accountable owner: Vendor account manager

Who approves access

Owner, Lumen Risk Labs delivery team, Vendor account manager
Operational & Third-party Risk, Aminah Yusof

Built on by SME Lending.

Confidential

Business-sensitive. Models and data products scoped to named business units.

Entitlement per business unit, approved by the owner; conditions attach.

Customer PII: Contains or processes personal data about customers (PDPA 2010).
Third-party: Supplied by an approved vendor; outsourcing and third-party risk controls apply.

Try it

Live sandbox

lc_mt700_specimen_00417.pdf

fixture

Evaluation only: restricted models run against synthetic or masked sample data until your business unit's access request is approved.

Feedback

Details

Updated: 2026-06-24
Latest version: 2.2.0
Licence: Vendor Licensed
Access: Restricted
Framework: PyTorch
Language: English

Trained on

trade-finance-transactions →