Amazon Prime: How AI And Forecasting Power One-Hour Deliveries Amazon executives often evoke magic when talking about fast shipping. Now in a race for one-hour deliveries, few retailers can afford to keep up. And few rely quite so much on artificial intelligence.

Optimized Prime: How AI And Anticipation Power Amazon's 1-Hour Deliveries

  • Download
  • <iframe src="" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript


Americans take for granted that most things they buy online will arrive on their doorstep two days after they clicked the order button. When Amazon introduced two-day shipping, it was a huge shift in retail thinking. To stay competitive, most major stores now offer similar speedy delivery. But getting you that package in just two days - or in some cases in just one hour - is extremely expensive. NPR's Alina Selyukh got a rare look at how Amazon makes it work.

ALINA SELYUKH, BYLINE: By the time someone clicks buy on Amazon, usually Jenny Freshwater had long predicted it.

JENNY FRESHWATER: In advance of them even knowing they want the product, in many cases.

SELYUKH: Her team's job is to anticipate demand for every single thing sold by Amazon in the world - not just how many blouses should Amazon sell but what color, size, sleeve length and most importantly...

FRESHWATER: Where do we actually put the product so that our customers can get it when they click buy?

SELYUKH: This is key to how Amazon cuts down delivery time, by stocking items as close as possible to the people who will buy them. That's why Amazon has been building smaller warehouses closer to city centers, where delivery might be within hours. And we should note, Amazon is one of NPR's sponsors.

FRESHWATER: We've started to build these AI algorithms, and the more we build, the better they get.

SELYUKH: AI algorithms, as in artificial intelligence, as in computers analyzing massive amounts of data. For example, AI knows that new doesn't always mean more sales. Like, with tax software, sure it does. Everyone wants the latest version. But a new DSLR camera actually creates more demand for the older version, which is cheaper. AI also knows that people often abandon their online grocery cart if bananas are sold out. And think about sunscreen. AI found many surges in the winter and spring.

FRESHWATER: Right around the holidays and then again when schools typically have spring breaks.

SELYUKH: Both AI and forecasting are not unique to Amazon. All retail stores work hard to plan for the future, and all major ones have their own algorithms and automated warehouses and delivery tricks. But it was Amazon Prime that got Americans hooked on two-day shipping. Now it's a race for a one-hour delivery with Amazon Prime Now. Few companies can afford that, and few rely quite so much on AI to control costs while growing.

CEM SIBAY: Really, AI is an underpinner technology for almost the entire Prime experience.

SELYUKH: Cem Sibay is an executive at Prime whose promise of free two-day shipping is the main reason millions of Americans shell out $119 a year for membership. Sibay and other Amazon executives push this illusion that fast delivery is magic. Like, the code name for Prime Now was Houdini. But the reality is forecasting on steroids and a meticulous shaving off of minutes and seconds on the journey of the package.

SIBAY: A lot of it is sort of end-to-end control of the experience as well.

SELYUKH: As in, Amazon has control of the entire process from the website to the warehouses to the actual delivery to your doorstep. In corporate lingo, that's first mile, middle mile and last mile. AI is woven through it all. In the first mile, when you order, AI analyzes your search and tells you upfront how fast your item could ship. In the warehouse - that's the middle mile - AI powers the Kiva robots. They look like large Roombas carrying yellow shelving units. In traditional warehouses, it's the people who walk to the shelves. Here, it's the robots that bring the shelves to people. The machines know what to bring and when to get each order packed in time for delivery. It's AI keeping track of all items in almost a million square feet of this warehouse. AI constantly arranges those shelves so that things you're about to buy are ready to go.

BRAD PORTER: There's a mix of industrial automation, manual processes and more sophisticated robotics.

SELYUKH: Brad Porter is the head of robotics for Amazon operations. This is controversial work in retail, where layoffs are rampant just as automation reshapes the workforce. Economists are divided on how much exactly AI will eliminate or create jobs, especially for lower-income Americans. In its defense, Amazon often points to how much it's actually been hiring. To Porter, we are in the latest chapter of industrialization.

PORTER: Industrial automation and robotics are here. They've been here for a long time.

SELYUKH: One area where AI has created a new type of job is in deliveries in the last mile. In busy cities, Amazon has to pull out all the stops. The company took a page from Uber and now hires drivers for side gigs, making superfast deliveries that pay as much as $25 an hour. AI fuels this. It matches package size to car size and even recommends what package to put in last. Sibay says when you're in a one-hour race, every minute counts. So AI's timing estimates consider traffic and building entry codes, and it learns from its mistakes.

SIBAY: The driver forgets his key at reception and has to walk a little bit longer. The driver is delivering a package, and it's an elderly lady. And they, you know, talk a little bit.

SELYUKH: It's hard for AI to predict all these scenarios, he says. But next time, maybe the address with a chatty Cathy will get a few more minutes baked into the algorithm.

Alina Selyukh, NPR News, Seattle.


Copyright © 2018 NPR. All rights reserved. Visit our website terms of use and permissions pages at for further information.

NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.