Frustrated customers often receive incorrect products, damaged items, or packages in poor condition. To tackle these issues, Amazon is deploying computer vision and AI to ensure products arrive in pristine condition and to bolster its sustainability efforts. Dubbed “Project P.I.” (short for “private investigator”), this initiative operates within Amazon’s North American fulfilment centres, scanning millions of products daily for defects.
Project P.I. utilizes generative AI and computer vision technologies to detect issues like damaged products or incorrect colors and sizes before they reach customers. The AI not only identifies defects but also uncovers their root causes, allowing Amazon to implement preventative measures upstream. This system has proven highly effective, accurately identifying product issues among the vast number of items processed each month.
Before dispatch, items pass through an imaging tunnel where Project P.I. evaluates their condition. If a defect is detected, the item is isolated for further investigation to determine if similar products are affected. Amazon associates review these flagged items and decide whether to resell them at a discount via Amazon’s Second Chance site, donate them, or find alternative uses. This technology acts as an extra pair of eyes, enhancing manual inspections across North American fulfilment centres, with plans for expansion throughout 2024.
Project P.I. also supports Amazon’s sustainability initiatives. By preventing damaged or defective items from reaching customers, the system helps reduce unwanted returns, wasted packaging, and unnecessary carbon emissions from additional transportation. This AI-driven approach ensures high-quality items reach customers while contributing to Amazon’s environmental goals.
In parallel, Amazon is using a generative AI system equipped with a Multi-Modal LLM (MLLM) to investigate the root causes of negative customer experiences. When defects reported by customers slip through initial checks, this system reviews customer feedback and analyses images from fulfilment centres to understand what went wrong, such as incorrect product sizes.
This technology benefits Amazon’s selling partners, especially small and medium-sized businesses that make up over 60% of Amazon’s sales. By making defect data more accessible, Amazon helps these sellers quickly rectify issues and reduce future errors, enhancing overall customer satisfaction and operational efficiency.