A $70M agricultural data firm partnered with KUNGFU.AI to turn one of its most labor-intensive data challenges into a scalable, AI-powered pipeline.
AGDATA delivers data and insights to North America’s largest agricultural manufacturers — but its most critical data process couldn’t scale. Matching hundreds of thousands of incoming transactions to the right grower records was manual, error-prone, and dependent on inconsistent formats and missing fields. It was the single biggest constraint on the company’s growth. AGDATA needed a partner who could build AI that worked against messy, real-world data — not just in a lab.
KUNGFU.AI started with strategy — not code. The team assessed dozens of potential AI use cases across AGDATA’s operations and identified grower matching as the highest-impact starting point. From there, KUNGFU.AI built a custom matching system that learns to recognize when different transaction records belong to the same grower, even when the details vary significantly across records. High-confidence matches are fully automated. Uncertain cases route to human reviewers. The solution was delivered production-ready and integrated into AGDATA’s core pipeline, with retraining infrastructure so the system improves over time.
EXPANDABLE: UNDER THE HOOD
The matching system uses a transformer model trained with triplet loss to generate embeddings from combined fields — farm names, business owners, addresses, and contact information — clustering transactions belonging to the same grower even across significant data variation. Tenant-aware constraints enforce data permission rules at the matching layer. A confidence scoring system based on embedding distances drives the automation tiers. Retraining logic and deployment infrastructure were delivered for ongoing model improvement.
