Amazon ships 20 million packages a day in 19 countries, requiring an astronomical amount of paper, cardboard and plastic packaging. Figuring out how to pack all those items efficiently is good for the environment and for its own profits.

So in 2019, Amazon launched its own proprietary AI model to reduce packaging waste. Five years later, it’s helping to save at least 500,000 tons of packaging a year, according to company data. That’s roughly equivalent to the weight of 7,750 Boeing 737 airplanes.

To handle the volume, Amazon researchers built the AI model, which it calls the Package Decision Engine, to predict the most efficient packaging choice – making sure that a set of dinner plates gets a sturdy box, say, while a blanket does without.

The model uses natural language processing and text-based data about each item from its online store, including basics like the product’s name and description. It also collects feedback from returns and product reviews, incorporating information about products that arrived damaged. It combines all that data with photos taken when items arrive at an Amazon warehouse using a special computer vision tunnel. Those photos give the company details on each object’s exact dimensions and capture images from multiple angles, helping it determine the best way to pack.

Over time, the model has added more nuance in identifying specific items so that, for example, personal items like adult diapers won’t be shipped without packaging in order to save on materials, and products that have strong magnets get enough protection so that they don’t get stuck on conveyors.

In a 2021 paper, Amazon researchers Prasanth Meiyappan, an applied scientist with a PhD in atmospheric sciences from the University of Illinois Urbana-Champaign, and Matthew Bales, a physicist in charge of machine learning in Amazon’s customer experience team, wrote that combining data from both visuals and text improved the model’s performance by as much as 30%.

Using AI to make packaging decisions is unusual, said Euihark Lee, an assistant professor at Michigan State University’s School of Packaging. “I think Amazon is in the forefront of this AI implementation on the packaging side because they have a lot of data,” he said. “Other companies don’t have that amount of data.”

Amazon said that its packaging model helped it to cut more than 2 million tons of packaging between 2015 and 2022. That’s an increase of 500,000 tons since the last time the retail giant announced its packaging reduction in 2021, when it reported saving more than 1.5 million tons since 2015. Amazon does not disclose how many tons of packaging material it uses each year, making it tough to know how significantly the AI model has reduced its total packaging use.

Experts told Forbes that any reductions are positive. “I think it is substantial,” said David Feber, a McKinsey senior partner focused on packaging. Rafael Auras, a professor of packaging sustainability at Michigan State University's School of Packaging, said that it was an impressive amount for one company to reduce, but the problem is so widespread that this hardly makes a dent. “There is a large opportunity not only for Amazon but for the whole industry to use AI to reduce packaging waste,” he said.

Source:

Forbes (2024.4.16) How AI Is Helping Amazon Save Half A Million Tons Of Packaging Per Year