About Carrier Selection and historical shipment data

Shipium's machine learning (ML) models are designed to help you achieve significant cost savings while maintaining reliable delivery promises to your customers. These models provide accurate estimated delivery dates (EDDs), enabling you to optimize Carrier Selection for speed, cost, or other priorities. One of these models, our machine learning TNT model, predicts time in transit by analyzing historical shipment data and considering various factors that can influence delivery times, such as weather patterns, seasonal trends, and route adjustments.

When you call our Carrier and Method Selection API service for carrier selection without generating labels, we rely on the delivery outcome data you provide to assess performance. Your shipment data provided for model training and evaluation plays a crucial role in ensuring the accuracy and ongoing evaluation of our ML models. In addition, providing comprehensive, accurate shipment information contributes to the TNT model's ability to generate reliable estimated delivery dates (EDDs), which in turn helps you make informed Carrier Selections based on your specific needs and priorities.

This guide outlines best practices for sharing your data to enable us to effectively monitor and evaluate delivery performance. You can find information on setting up your file transfers and the data needed in Shipment File Transfers.

Best practices for data contribution

To maximize model accuracy and the effectiveness of our performance monitoring and model evaluation, we recommend following these best practices when sharing your shipment data:

1. Reference identifier. On each Carrier Selection request, you can include a unique shipment identifier (e.g., your internal unique identifier for the shipment) via the partnerShipmentId shipment parameter. When providing Shipium with historical data, we ask that for each shipment observation you include that same partnerShipmentId as the partner_reference_identifier field so that we can link your outcome data to our internal Carrier Selection data. This enables us to evaluate your specific delivery performance, similar to how we do for Label Generation customers, and aids in customer support investigations. A complete list of field names and descriptions, including partner_reference_identifier, is in Shipment File Transfers.
2. Tracking number. Provide a carrier tracking number for each shipment to allow us to register your shipments and gather carrier-based ship date-time and delivery date-time information. This helps us understand potential discrepancies between the estimated ship date-time and the actual ship date-time, and allows for localized timezone evaluation.
3. Timely data updates. Regularly provide us with updated historical shipment data. Timely data ensures our models are trained on the latest information, enabling accurate predictions that reflect recent trends and changes in your shipping patterns. Shipium supports both SFTP and S3 direct file drops. If you would like to pursue setting up either a manual or automated file transfer, please reach out to your Shipium Implementation Team member or Customer Success Manager to start the process.
4. Consistent file format. Maintain a standardized and consistent file format for your data to minimize processing errors and ensure a smooth ingestion pipeline.
5. Complete shipment data. Include all required fields for each shipment, such as origin and destination information, carrier details, and package characteristics. This allows our models to account for all variables that may impact transit time estimates.

Resources

Your Shipium team member is available to help along the way. However, you might find these resources helpful:

• Terms and Definitions

• Troubleshooting Guide

• Contact Shipium Support