All case studies
Retail · Computer Vision · Production engagement
Automating retail shelf compliance with computer vision
Replaced a fully manual shelf-audit process with a production CV pipeline that turns in-store images into structured data on card placement, SKU identity, and planogram compliance.
Hundreds saved / mo
Manual inspection hours
Exception-based review
Workflow shift
Scales across stores
Throughput
Near real time
Feedback loop
Challenge
Avanti Press manages thousands of greeting-card displays across retail stores. Their team was spending hundreds of hours each month reviewing shelf images to verify cards were placed correctly. The process was manual, hard to scale across locations, prone to inconsistency, and delayed in surfacing merchandising issues. They needed to automatically identify every card on a shelf, determine its exact position, match it to the correct SKU, and flag missing or misplaced cards.
Client
Avanti Press
Our approach
- Built a multi-stage pipeline: shelf detection → card detection → cropping → classification → compliance logic → JSON output
- Isolated valid shelf regions from full in-store images, filtering out floors, signage, and adjacent aisles
- Trained a detection model to identify individual cards, empty slots, and skewed or misaligned placements
- Built a classification model on Avanti's catalog that returns predicted SKU, confidence, and alternative matches
- Combined card location, SKU, confidence, and the expected planogram to flag correct, misplaced, missing, and low-confidence cases
- Deployed the full pipeline as an API endpoint that powers an internal tool with visual card positions and real-time compliance flags