SMBSC modernizes beet sugar intake quality screening with AI on AWS

During peak harvest, SMBSC manages tens of thousands of truckloads of beets per season, each requiring inspection for impurities such as excess dirt and leafy greens before entering production. The cooperative’s previous process relied on manual visual grading and a binary greens present/not present assessment, which introduced inconsistent grading, labor-intensive review processes, and throughput bottlenecks during peak intake windows. These challenges compounded across the season, creating delayed visibility into grower quality trends and meaningful financial risk from off-specification load payments.

Greens alone could represent multimillion-dollar annual inefficiencies because of downstream sugar quality impact and grower payment adjustments. “During peak harvest, it just wasn’t realistic to manually inspect every load with consistency,” says an SMBSC operations leader. “We needed a way to standardize grading, reduce false negatives, and scale quality screening without disrupting throughput.” SMBSC needed a system that could perform reliably across variable lighting and weather conditions throughout the most critical production window of the year.

Combining technical capabilities with operational experience, Tactical Edge AI was a natural choice for SMBSC. The partner designs models that perform reliably outside controlled conditions, accounting for variability in lighting, weather, and input quality—common challenges in agricultural settings. Early in the engagement, Tactical Edge AI conducted onsite assessments to understand the intake process, environmental conditions, and peak-season constraints. The team took ownership across the full build—from camera setup and data pipeline design through model deployment and NRT alerting—to fully incorporate the solution into SMBSC’s existing workflows.

Using Amazon SageMaker AI, a fully managed service that brings together a comprehensive set of AI tools and capabilities, Tactical Edge AI built, trained, and deployed a computer vision model. The model can detect impurity levels in intake images, replacing SMBSC’s binary inspection with continuous 0–100 quality scoring. This supports more precise, consistent grading and threshold-based exception handling across every load. To support rapid development while minimizing upfront investment, the team architected the solution by using AWS serverless services. The object storage service Amazon Simple Storage Service (Amazon S3) saves intake images and routes them to processing workflows through event-driven notifications. AWS Lambda handles all infrastructure management, orchestrating backend processing and dashboard pipeline updates in NRT. By performing NRT impurity detection and scoring across the high-volume seasonal intake, Amazon SageMaker AI endpoints maintain consistent performance across lighting and weather variations throughout the harvest window.

When a load exceeds established impurity thresholds, the system automatically generates alerts to operations leaders, which reduces time to action from days to minutes. “The AWS based cloud architecture that Tactical Edge AI delivered created a scalable, sustainable foundation for automating one of our most critical workflows,” says Mike Skucius, executive director of transformation and information services at SMBSC. “The transparent support and integrated ticketing system keep our teams connected and responsive.”

Delivering more than 90 percent accuracy in impurity detection, the AI-powered intake-screening system has transformed SMBSC’s operations. Grading time dropped to under 3 seconds per load, significantly improving throughput at the receiving stations while reducing the manual reinspection burden. The automated reporting capabilities removed time-consuming weekly compilation work, and the system’s consistent performance across environmental conditions gives SMBSC a reliable foundation for ongoing quality management. From a financial standpoint, SMBSC projects a potential USD $5 million annual savings by reducing payments for off-specification loads and improving intake decisions, with a rapid return on investment.

Beyond operational gains, the continuous-scoring model creates a faster and more transparent feedback loop with growers. Near-immediate identification of below-threshold loads improves accountability and supports long-term inbound quality improvements across the cooperative’s grower network. “It’s not only more efficient—it gives us confidence that we’re applying quality standards consistently across every load,” says an SMBSC operations leader. SMBSC is now building on this data foundation as it continues its journey toward adopting agentic AI by using Amazon Bedrock, a service for building generative AI applications and agents at production scale. The cooperative sees this work as the first step in a broader data-driven transformation: one that connects intake quality to smarter decisions across the entire production chain.