How Amazon Marketplace evaluates content quality with Amazon SageMaker AI

Amazon's Marketplace team describes a multi-model fusion system on Amazon SageMaker AI that evaluates A+ Content quality holistically rather than asset by asset.

• A+ Content is the rich media widget on the product detail page, and Amazon says selling partners using high-quality A+ Content boost their sales by 8–20 percent compared to those without it.
• Because brands range from large in-house design teams to bootstrapped sellers, content quality varies widely, and checking each image or bullet in isolation misses problems that only appear when the page is viewed as a whole.
• The approach renders the full page as a single snapshot and has a vision-language model score it against a rubric, then combines those semantic scores with OCR readability signals in a CatBoost classifier trained on expert annotations.
• Keeping the VLM and prompt fixed and retraining only the classifier avoids GPU training infrastructure and the overfitting risk of fine-tuning on a few thousand labeled examples.
• SHAP explains the classifier's quality decisions and produces improvement recommendations for selling partners, and the pipeline runs end to end on SageMaker AI, from data ingestion and annotation through model hosting, explainability and production monitoring.