Inside an Amazon warehouse, when an order gets ready to ship, a worker doesn’t just make a decision about what package seems best. Instead, a suite of machine-learning tools analyze the combination of items that are shipping and choose the size and type of packaging—or whether a box is needed at all—as part of the company’s ongoing quest to shrink its giant packaging footprint.

The company won’t share how much cardboard it uses in a year, though as the world’s largest online retailer, the number is huge. One recent analyst estimate suggests that Amazon shipped 415 million packages in the month of July, for example. But the company says that over the last five years, it has managed to reduce the amount of packaging used per shipment by roughly a third, eliminating 915,000 tons of packaging material, or the equivalent of 1.6 billion boxes that it would have otherwise used. In the last two years, its software has allowed the company to avoid shipping more than 500,000 tons of waste.

In many cases, orders now come in padded mailers rather than cardboard boxes. This helps lower the carbon footprint of the delivery, which drops along with the size and weight of the package. “We can reduce the wastefulness of the packaging, and the space it takes up on trucks, which reduces our overall fuel to transport to the customer,” says Kim Houchens, Amazon’s director of customer packaging experience. “And then, of course, when it arrives on the doorstep, we have less for the customer to recycle or to waste.”

Bubble envelopes aren’t perfect: Amazon typically uses plastic mailers that can’t be recycled in curbside bins, and which can even cause problems in recycling centers if they’re put in the system and jam up equipment. “It’s actually a huge burden on [the recycling centers],” says David Pinsky, plastics campaigner for Greenpeace.
Amazon notes that it’s possible to drop off the mailers at some stores for recycling, and that it’s donating money to help improve recycling infrastructure. But it’s also working on alternatives, including a new all-paper mailer now in use for some shipments. Launched in 2019, the new mailer uses four layers of paper and a unique treatment with a water-soluble glue. “When you heat it up, it actually puffs up and makes what looks to be like Styrofoam,” Houchens says. (The appearance is a challenge, and the company is also working with recycling centers to explain that it is, in fact, recyclable.)

AI tells warehouse workers—or, in some cases, machines that are packing orders automatically—when a padded mailer can be used instead of a box, pulling in data from customer complaints and returns to understand when a particular product should have had more packaging, or when it was overpackaged, as in a shipment where a single small item is inexplicably delivered in a giant box.
The sheer scale of the range of products that the retailer offers means that it isn’t feasible for someone working in a warehouse to quickly choose a package for any particular order, or even to have a database listing which package to use for any given product.

“It appears perhaps simple at first glance—that you could just let the packer make the right decision, or that you could properly measure and keep a catalog of everything perfectly,” Houchens says. “But Amazon’s ever-changing and vast catalog has hundreds of millions of different products. If you then try to combine them in any combination of orders that a customer can place, we’ve got near-infinite permutations of customer shipment options. This is unlike most everyone else in the retail business. Even the largest e-commerce companies don’t have the catalog that we have. And not only is it hundreds of millions of products today, but it’s changing every day. We might have 35% new products between now and Christmas to learn about.”

Algorithms can also judge how products can be packaged together—which items are flexible, for example, and can be packed around items that are rigid. (This involves building in shortcuts, because fully solving a problem like how to fit a five-item order together “would take greater than the time we have here on Earth in order to calculate all those possibilities,” says Matthew Bales, research science manager for Amazon’s customer packaging experience team.) In another bid to reduce packaging, the company offers Prime customers the choice of “Amazon Day,” meaning multiple orders will wait to ship and be delivered on a particular day of the week, which lets them be combined into fewer packages.

The algorithms also tell packers when a product can ship without an outer box, through the company’s “frustration-free” packaging program, where manufacturers are prodded by Amazon to develop packages that can ship alone. Tide laundry detergent, for example, now ships in its own packaging, a wine-box-like container instead of a plastic jug. In Amazon’s Packaging Lab, engineers work on problems like how to safely ship liquids with as little packaging as possible.

The company expects that the amount of packaging it uses per order will continue to shrink. Still, the challenge is immense. Data analytics company Fastmarkets says that shopping online uses seven times more cardboard than shopping in a store. And as online shopping continues to grow, Greenpeace’s Pinsky argues that retailers like Amazon will need to move to new reusable packaging models like those pioneered by companies like Loop, both for product packaging and the packaging used in shipping. “We need to eliminate this question of single-use packaging,” he says, “whether plastic, cardboard, paper, whatever it is—and embrace reuse.”