Amazon pioneers artificial intelligence machine learning for packaging waste reductions

06 Jan 2022 --- Amazon is reducing packaging waste at scale with several artificial intelligence (AI) technologies. A combination of deep learning, natural language processing, and computer vision is helping the e-commerce giant determine the right amount of packaging for each product.

Over the past six years, these tools have reportedly helped Amazon reduce per-shipment packaging weight by 36% and eliminate more than a million tons of packaging, equivalent to more than 2 billion shipping boxes. The technologies are currently being applied to North American and European product lines.

As research science manager Matthew Bales, who heads up machine learning within Amazon’s Customer Packaging Experience team, explains in a corporate blog post: “Customer feedback is paramount, it powers all of our [Amazon’s] statistical testing.”

Bales and his colleagues built a machine learning model mainly based on the text-based data customers find on the Amazon Store – such as item name, description, price and package dimensions – to predict whether a particular product could be safely shipped in a specific package type.

The company’s Shipment Zero goal is to deliver 50% of shipments with net-zero carbon by 2030. From a packaging perspective, this target requires products to be shipped without added Amazon packaging or in carbon-neutral packaging.

Mastering machine learning
Amazon’s machine learning model has been trained on millions of examples of products successfully delivered in various packaging types but also examples of products that arrived damaged.

Meanwhile, real-time feedback on when a product is not sufficiently protected by its packaging is revealed when customers report it via the Online Returns Center and product reviews.

Amazon’s AI tools have enabled it to reduce per-shipment packaging weight by 36% in six years.The model learned that specific keywords are crucial when making packaging decisions. For example, keywords that suggested mailers were the right choice included “multipack” and “bag,” as the product might already have some form of protective packaging.

“The portion of the model that's learning from the Amazon Store has learned really well what the product is, and about its dimensions,” notes Bales.

Tunnel vision optimization
Equally important is understanding how the vendor packaged the product before sending it to a fulfillment center. Here, computer vision must be deployed to identify product packaging at scale, as specific products can be delivered in different pack types.

Bales’ team repurposed Amazon’s image data, using conveyor belts with special computer-vision tunnels equipped with cameras that capture images of the products from multiple angles. These tunnels can identify product dimensions and detect defects.

“Our model detects the packaging edges to determine shape, identifies a perforation, a bag around the product, or light shining through a glass bottle,” describes Prasanth Meiyappan, an Amazon applied scientist, who expanded the model’s training.

“But to some extent, how the model makes its judgement about what it detects in images is hard for a human to discern because the product features identified and weighted by the model tend to be complex.”

“The important thing,” Bales adds, “is the packaging decisions generated by the model are empirically accurate.”

“When the model is certain of the best package type for a given product, we allow it to auto-certify it for that pack type. When the model is less certain, it flags a product and its packaging for testing by a human.”The e-commerce giant is incentivizing vendors to create optimized e-commerce packaging through its Frustration Free Packaging program.

“It’s a triple win,” he says. “Reduced waste, increased customer satisfaction, and lower costs.”

The model’s performance improved by 30% after incorporating both text-based and visual data compared with using text-based data alone.

Room for improvement
The machine learning team is keen to expand the technology by training the model to understand all customers’ languages while incorporating the unique fulfillment demands of each country.

While Amazon scientists continue investigating other ways to utilize machine learning to eliminate waste, the company is also working to reduce packaging waste throughout the e-commerce supply chain.

Notably, the e-commerce giant is increasingly incentivizing its vendors to create optimized e-commerce packaging for themselves that saves space and materials without compromising product protection.

Meanwhile, Amazon US recently replaced plastic bubble bags with recycled paper for frozen groceries, while Amazon France announced it would stop packaging items in single-use plastic sleeves.

However, not all the company’s environmental sustainability moves are met with universal praise. Recently, Greenpeace accused Amazon Fresh in Singapore of greenwashing after it replaced unrecyclable gel packs with recyclable frozen water bottles to maintain the freshness of frozen and chilled products during deliveries.

Amazon’s wider Climate Pledge targets net-zero carbon by 2040, a decade earlier than the 2050 emissions target of the Paris Agreement.

By Joshua Poole