Cutting Contract Processing from Hours to Minutes

Descartes Systems Group is a logistics and supply chain technology leader headquartered in Waterloo, Canada, with more than 1,800 employees and a history dating to 1981. Its Global Price Management (GPM) system runs on rate data extracted from buy contracts issued by Vessel Operating Common Carriers. Every one of those contracts had to pass through human hands before the pricing engine could use it.

• Buy contracts arrived in highly varied formats: inconsistent tables, nested rows, scattered footnotes, and free-text sections, with no two carriers structuring them the same way

• Each contract was transcribed by hand into standardized templates for the GPM system, a task measured in hours per contract

• Manual transcription carried a high risk of misinterpreting the rate data that feeds pricing decisions

• Ingestion delays slowed the pricing operations that depend on current contract data

• Labour-intensive handling made contract volume a direct driver of cost, so growth meant more transcription, not more throughput

• Built a modular extraction and template generation pipeline on Azure Document Intelligence and Azure OpenAI, structured as five core modules

• Delivered a user interface for initiating contract uploads and tracking processing progress

• Engineered a data extraction engine to parse both structured and unstructured contract content, paired with a data organization module that merges and classifies complex, nested table data

• Added an interpretation layer for contextual analysis of paragraphs and footnotes, where rate exceptions and conditions typically hide

• Generated standardized Rate, Origin Arbitration, and Destination Arbitration templates as output, integrated with the GPM backend

• Designed the pipeline to absorb format variability across vendors and contract types, so new carriers do not require new manual handling

• Estimated 80% reduction in manual contract-processing effort demonstrated in the proof of value

• Contract processing that took hours per contract completed in minutes

• 5 core modules delivered: upload and tracking interface, extraction engine, data organization, interpretation layer, and template generation

• 3 standardized template types generated automatically: Rate, Origin Arbitration, and Destination Arbitration

• Structured tables, nested rows, footnotes, and free text handled across varied carrier formats in a single pipeline

• Output integrated directly with the GPM backend

• Pricing operations will no longer wait on transcription. The proof of value showed contract data will reach the GPM system in minutes instead of hours.

• Misinterpretation risk dropped at the source. Standardized, structured templates replaced hand-keyed data in the workflow the POC covered.

• Contract volume decoupled from headcount. The pipeline was designed to absorb growth across vendors and contract types without additional manual effort.

• Descartes now owns a validated pattern for AI-driven document ingestion, a starting point for broader automation across its pricing workflows.

This was a proof-of-value engagement. The outcomes here were demonstrated within the POC scope against real carrier contracts. Production savings will be measured once the pipeline runs against live contract volume.