About the company

The Client is headquartered in Kuala Lumpur, Malaysia, is a multinational Global Business Services and Solutions provider and one of the earliest movers in Malaysia’s shared-services sector. The company specializes in Finance & Accounting, Human Resources, IT Project & Operations, Shipping Documentation Services, Change Management, Recruitment Process Outsourcing, Corporate Function Services, and Digital Transformation, delivering these services to more than 300 client organizations across industries ranging from manufacturing to oil & gas and renewable energy.

Challenges

As the Global Business Services arm of a multinational, resource-based industrial group, the client’s shipping-document processing was entirely manual — time-consuming, labor-intensive, and prone to error when interpreting low-quality scanned trade forms. Every invoice, Letter of Credit, and bank form required full manual re-keying, with no systematic way to distinguish straightforward documents from genuinely ambiguous ones.

Two structural issues compounded the problem:

• Document heterogeneity — incoming paperwork spanned high-fidelity structured formats (e.g., ERP-generated invoices) and complex, often poor-quality unstructured scans, each requiring different handling.
• No confidence-based triage — every document received the same full manual review regardless of how much of it was actually unambiguous, meaning review effort scaled directly with volume rather than with genuine complexity.

This dependency on full manual review constrained processing speed and consistency, while introducing meaningful error risk — particularly for low-quality scans, where misreads could trigger downstream correction cycles and processing delays.

Searce Solution

Our team of solvers developed a solution on AWS, deploying a dual-layer AI parsing pipeline combined with an intelligent Human-in-the-Loop verification interface:

• Dual-layer extraction — Amazon Textract performs the initial structured extraction pass across invoices, Letters of Credit, and bank forms, with Amazon Bedrock’s Vision-enabled LLM applied as a secondary pass for fields Textract can’t confidently resolve.
• Confidence-based routing — extracted data routes automatically into high- and low-confidence paths, orchestrated via AWS Lambda, narrowing human involvement to genuinely ambiguous fields rather than full-document review.
• Human-in-the-Loop verification interface — automated highlighting and overlay flags exactly which fields fall below the confidence threshold, letting reviewers verify in seconds rather than re-keying entire documents.
• Precision field mapping and bank-form population — an LLM-driven metadata-matching engine maps extracted values directly into target bank-form fields, automating what was previously a fully manual re-keying step.
• Enterprise-grade, decoupled storage on Amazon S3 — maintains strict data isolation while handling large file volumes, with a managed layer tracking each document through its full lifecycle.

The solution was validated across multiple markets, and Searce has since generalized its methodology — the dual-layer AI parsing pipeline and Human-in-the-Loop verification approach — into a repeatable framework, published under AWS Marketplace’s Professional Services category as “LLM Powered Intelligent Document Processing (IDP) by Searce,” enabling the same approach to be applied across future engagements.

Business Impact

By pairing automated extraction with intelligent confidence routing, the platform dramatically reduced processing overhead while raising extraction accuracy across international markets.

• 94.10% global extraction accuracy achieved across all document types and markets in the pilot, directly validating the dual-layer AI parsing approach.
• 90%+ estimated reduction in manual review effort per document batch, by confining human attention to flagged low-confidence fields rather than full documents.
• Only 4–10% of extracted fields required human correction — 95.98% field-match accuracy in Malaysia and 89.53% in Singapore — versus 100% manual re-keying in the prior process.
• Directly reduced labor hours per shipping document processed, freeing review staff from routine re-keying to focus on genuinely ambiguous cases.
• Reduced downstream error risk from manual interpretation of low-quality scans, avoiding the processing delays and correction cycles that error-prone manual review previously introduced.

By transforming a rigid data-entry workflow into an intelligent, exception-based routing pipeline, Searce eliminated major friction points in the client’s global trade operations. This successful transformation was later generalized into Searce’s repeatable AWS Marketplace framework, giving the client a scalable, future-ready foundation for document processing worldwide.