AWS for Industries
WNS Malkom accelerates shipping operations with AWS Bedrock
In the fast-paced world of logistics, every minute counts. When trucks are idle waiting for paperwork processing, it costs companies dollars in wages and delays. Less-than-truckload (LTL) deliveries create a paperwork nightmare that typically causes next-day delays. Enter WNS Malkom’s groundbreaking solution: an AI-powered system that transforms hours of document processing into minutes, enabling same-day operations at unprecedented scale.
The challenge
Today’s shipping industry faces critical document processing challenges that impact operational efficiency.
Critical documentation hurdles:
- Document Variety: Shippers use countless custom bill of lading (BoL) formats, making template-based automation impossible
- Accuracy Demands: Even minor errors in digitization of key fields can trigger costly shipping mistakes and billing disputes
- Manual Elements: Handwritten notes and instructions require special handling and verification
- Complex Data Structures: Extracting data from non-standard tables and inline content requires sophisticated processing
Operational challenges:
- Peak Load Management: With 70-80 percent of daily documents arriving in just 3-5 hours, processing capacity must scale instantly
- Business Rule Complexity: Multiple cargo types, payment terms, and shipper requirements demand intelligent processing
- System Integration: Solutions must work seamlessly with both legacy systems and modern cloud platforms
The solution
WNS Malkom (WNS), a proprietary shipping automation platform, addresses these challenges by leveraging WNS’s proprietary systems and Amazon Web Services (AWS). It features a digitization desk that automates the processing of BoLs, commercial invoices (CIs) and other shipping documents.
The solution comprises three integrated parts that can be tailored for each customer to deliver an end-to-end digitization desk on AWS.
Part 1: Image/document ingestion and extraction
- Primary purpose: Ingest and process scanned documents upon arrival and extract text/data/information within the required turnaround time (TAT). According to WNS’ experience, large carriers may experience arrival volumes of up to 10,000 documents for each hour during peak times, with a typical TAT of 15–60 minutes.
- Key non-functional requirements include the ability to:
- Ingest various formats, types and resolutions of images
- Ingest images from multiple channels (for example, email attachments, file upload, API, and so on)
- Handle spikes in load to ensure processing within the specified TAT
- Auto classify incoming emails and documents
- Automatically extract information from the bills with high accuracy (over 90 percent) for shipper, consignee and bill-to details
- Extract information from handwritten documents
- Extract information from tables within the documents
- Continuously improve extraction accuracy through user feedback, aiming for touchless processing
- Apply business rules based on shipper/consignee and bill-to parties during the extraction process
Part 2: Document validation
- Primary purpose: Present digitized documents to an agent or auditor for validation or verification
- Key non-functional requirements include the ability to:
- Extract information from bills while in process by selecting a rectangular area within the image
- Visually present the agent the area of the image from which the information was extracted
- Configure business rules for special instructions and soft/hard validations based on shipper, consignee, bill-to and type of goods being transported
- Perform data validation against look-up tables and the master database
Part 3: Digitized data submission
- Primary purpose: Integrate with the system of record and submit validated data for downstream processing in the freight management system
- Key non-functional requirements include the ability to:
- Integrate with client enterprise resource planning/transportation planning and execution systems (ERP/TMS) leveraging multiple methods such as robotic process automation (RPA), file, API, and so on
- Handle exceptions
- Create an output in multiple formats such as XML, JSON and CSV
The following image is a reference architecture of the WNS Malkom solution.
To achieve these requirements, the solution leverages many AWS services. WNS’ solution is able to ingest various formats, across multiple channels, to receive scanned images through an API interface (using Amazon API Gateway), cloud storage (using Amazon Simple Storage Service (Amazon S3)) and email (using Amazon Simple Email Service (Amazon SES)).
These same services are also used for integration with client ERP/TMS systems through multiple methods (for example, RPA, as a file, API, and so on). A digitized BoL representation can be generated in JSON or XML format, which can be posted to the downstream system through a serverless event-based architecture.
Content is extracted automatically from documents of multiple types (for example, TIFF, PNG, PDF, and so on). A serverless extraction pipeline is triggered by an event originating either from an incoming email, a call to an API or a file being pushed into cloud storage. The AWS fully managed machine learning (ML) service Amazon Textract is leveraged to extract content from images and PDFs. A ML model is deployed to aid the extraction of specific data fields within the image.
We leverage Anthropic’s Claude 3.5 Haiku for image data extraction on Amazon Bedrock. Amazon Bedrock enables:
- High availability with cost-effective deployment
- Easy switching between AI models without code changes
- Consistent performance during peak loads through provisioned throughput
- Flexible choice between Anthropic’s Claude 3.5 Haiku, Mistral AI and other models
- Enhanced security through built-in protection using Amazon Bedrock Guardrails
WNS’ proprietary shipping contextualization engine is leveraged to extract information around shippers, consignees, bill-to parties, commodities, weights, cargo type, various types of shipping account reference numbers (ARN) and shipping instructions. The solution automatically detects shipping ARNs and populates fields for shippers, consignees and bill-to parties using information extracted from the image. This includes populating from short bill data and reference data, if available, and using machine prediction based on historical data in the system. AWS Lambda functions are leveraged to merge outputs of various extraction engines to improve both extraction coverage and extraction accuracy.
WNS’ solution follows AWS Well-Architected framework best practices for scalability, reliability and high availability (even during volume spikes and continuous operations). The solution has the ability to handle volume spikes with consistent performance throughput by leveraging the serverless capabilities of the AWS Cloud. Employing multiple AWS services (including Amazon Aurora Serverless, Application Load Balancer, AWS Auto Scaling, and Amazon CloudWatch) it can scale up to handle over 12,000 documents for each hour seamlessly.
With this architecture framework and combination of AWS services with proprietary systems, WNS is able to accurately digitize 70–90 percent of a BoL/CI without any human intervention or compromising accuracy.
Conclusion
The BoLs, CIs and other shipping documents accompanying all goods needs to be handled in an expedient and accurate manner. Doing so leads to saving costs for customers by eliminating delays, accurate billing and smoother operation for shipping and trucking company, which creates better, efficient supply chains.
The serverless architecture of WNS Malkom’s solution, combined with AWS services such as Amazon Bedrock, enable the processing capability to soar from a few documents each hour to over 12,000 documents for each hour, without any human intervention.
WNS Malkom’s accurate, automated solution is just what shipping and trucking companies need as clients continue to ask for their goods to be delivered faster and faster.
To learn more about transport and logistics, contact your AWS account team.
Further reading