Amazon has launched Project PI (Private Investigator), an innovative artificial intelligence (AI)-based system designed to detect defective products before they leave logistics centers and reach customers. This initiative reflects the company's commitment to providing the best experience for its customers. Project PI is active in several distribution centers in North America and has shown promising results, so Amazon plans to integrate it into more centers later this year.
Project PI uses generative AI technology and computer vision to scan products and detect defects, from physical damage to labeling issues and misclassification by color or size. Products destined for customers pass through a tunnel where they are scanned and the vision program inspects them for damage. If a defect is detected, the package is isolated, the defect is assessed and a check is made to see if there is a problem with similar items or the batch to which it belongs to determine the cause of the problem.
Pingping Shan, Director of Perfect Order Experience at Amazon, explained: "We want to equip ourselves with the most powerful and scalable tools and levers to help us protect our customers' trust."
The system is designed to prevent the shipment of damaged or faulty products while taking preventative measures to avoid these issues in the future. Amazon is also contacting sellers to inform them of these setbacks and enable them to avoid similar problems in the future.
How does it work?
Project PI was derived from Amazon's product quality program and owes its success to a combination of advanced AI technologies and machine learning models. Packages are scanned with high-resolution cameras that capture detailed images of each product. The AI analyzes these images in real time to identify any visible defects. In addition, an OCR (Optical Character Recognition) model checks the inventory's labeling information and compares it with the Amazon database for discrepancies.
If a defect or problem is detected, such as an expired date, the product is removed from the shipping line. Depending on the severity of the problem, the product is sorted out, repaired, sold at a lower price or donated. In addition, Project PI analyzes the causes of the problems in order to make corrections and avoid similar problems in the future.
Amazon has used customer feedback to train the machine learning models and recognize the difference between normal and faulty items. It has also labeled errors and the types of defects that customers notice.
Amazon has implemented a multimodal large language model (MLLM) to investigate the cause of negative customer experiences from receiving defective products. The AI analyzes customer feedback and examines images taken by Project PI at the logistics center from which the item was shipped, as well as other data sources. In this way, the AI finds out where the error occurred and helps to make error data more accessible.
Using AI to avoid shipping faulty items that end up being returned benefits customers, sellers and the environment. Kara Hurst, Amazon's Vice President of Global Sustainability, explains: "Amazon is using AI to meet our sustainability commitments with the urgency demanded by climate change while improving the customer experience."
