Amazon is the co-founder of the Climate Pledge.

The signatories of this text undertake to achieve the objectives of the Paris Agreement ten years in advance.

Their ultimate challenge is to achieve net zero carbon by 2040.

To achieve this, Amazon has developed a machine learning (ML) solution powered by the Amazon SageMaker machine learning platform.

Its main goal is to create a more sustainable packaging process, while keeping the customer experience bar high.

For example, in a blog post, the company describes how artificial intelligence helps it determine how to use the right amount of packaging for the hundreds of millions of products it ships.

A 36% reduction in packaging weight

To limit its waste production, the e-commerce giant uses its machine-learning as well as a combination of natural language processing and computer vision.

According to Amazon, these tools have allowed it to reduce the weight of packaging per shipment by 36%.

That's more than one million tons of packages, the equivalent of more than 2 billion shipping boxes, over the past six years.

Specifically, these technologies allow the company to refine the choice of packaging for each product.

Indeed, given the size of Amazon's catalog, finding the right amount of packaging to ship an item is a tall order.

A machine learning model

Machine learning helps predict whether Amazon can safely ship a product in a particular type of packaging. This technology relies on textual data from online product listings, such as item name, description, price, and package dimensions.

For its proper functioning, customer feedback is paramount.

Indeed, they are the ones that feed the statistical tests necessary for machine learning.

Clearly, when a packaging does not protect a product enough, Amazon has access to almost direct feedback.

Customers report poor condition through several online feedback mechanisms.

The artificial intelligence is based either on the data of the products delivered successfully, or on those of the products that arrived damaged because they were poorly protected.

The importance of keywords

When choosing the package, some keywords are particularly important. A particularity that the ML system developed by Amazon has integrated well. For example, the keywords “multipack” and “bag” suggest that the product might already have initial protective packaging. For the latter, envelopes seem to be the most relevant solution. Conversely, the keywords "ceramic", "cup" and "glass" indicate that a shipment in a padded envelope is not suitable. In order not to damage them, Amazon will therefore have to send these products in a box.

If considered alone, keywords and customer reviews are not accurate enough data.

Indeed, the type of packaging that the seller chooses to package his product before sending it to a fulfillment center is just as important information.

Identifying these primary packaging choices helps define the best final shipping packaging.

To collect this type of information, product images posted by sellers are not enough.

Amazon takes the example of LED bulbs.

Let's say a seller presents a pack of these bulbs with a photo of a single unpackaged bulb.

The artificial intelligence may see the product as fragile, when in reality, the seller will pack the pack securely and in a way that no additional protection is needed.

Similarly, a shipper may choose to pack a teapot in clear bubble wrap or in a sturdy box.

To overcome this risk, Amazon has developed artificial vision technology.

Artificial vision as a solution

Once delivered to the execution centers, the products continue their journey on conveyor belts placed in very special tunnels. In fact, in its tunnels, Amazon uses cameras with software-driven sensors. The latter take images of the products from several angles and provide additional data for the choice of the final package. This AI also makes it possible to plan grouped shipments if they share the same recipient.

"When the model is certain of the best type of packaging for a given product, we allow it to self-certify for that type of packaging," says Matthew Bales, research science manager at Amazon.

He adds, "when the model is less certain, it flags a product and its packaging for human testing."

According to Amazon, integrating text and visual data improved ML model performance by 30% compared to using text data alone.

A greener production line

In addition to reducing the number of boxes needed for shipments, the grouping of deliveries reduces the traffic of delivery trucks.

In other words, to reduce the carbon footprint of deliveries.

All of these goals align with those of Amazon's "Shipment Zero" initiative.

This program hopes to see 50% of Amazon's shipments become carbon free by 2030. The end goal is to achieve zero carbon emissions for all shipments.

Currently, Amazon is only using AI for US and European customers.

To hope to achieve the carbon footprint reduction rate it has set itself, the company should extend its model to more countries in the world.


Delivered catering urged to reduce the environmental impact of its packaging


Amazon will open its first high-tech clothing store in 2022

  • ecology

  • Delivery

  • Economy

  • Amazon

  • Artificial intelligence

  • Waste

  • high tech

  • 0 comment

  • 0 share

    • Share on Messenger

    • Share on Facebook

    • Share on Twitter

    • Share on Flipboard

    • Share on Pinterest

    • Share on Linkedin

    • Send by Mail

  • To safeguard

  • A fault ?

  • To print