LMC Software: How Algorithms run Amazon's Warehouses

Amazon's vast, 40,000 sq m, fulfilment centre in Hemel Hempstead (Credit: Amazon)

By Chris Baraniuk, 18 August 2015

When you click “buy’ on Amazon, a flurry of activity begins inside a nearby warehouse – all managed by smart computer code. What’s it like to work there?

“Work Hard. Have Fun. Make History.” So reads a sign above the entrance to Amazon’s newest UK “fulfilment centre” (or warehouse) in Hemel Hempstead.

Inside lies more than 40,000 sq m of shelving, packing lines and millions of products. One area where larger products are stored is known as “pallet land” and adjacent, at another end of the space sits “the tower”, where several floors of shelving are stacked on top of each other. Pickers constantly, and more or less silently, walk up and down the tower’s lengthy aisles, pushing carts into which they deposit items purchased by someone, somewhere online.

Founded at the dawn of the web in 1994, Amazon is now reportedly worth $247bn (£157bn). But the company is not, in fact, hugely profitable. What keeps Amazon afloat? As for any business with tight margins, efficiency is key.

A recent New York Times investigation revealed that the corporate culture inside Amazon is highly driven by data: personal performance, for example, is continually checked with a software system called the Anytime Feedback Tool that allows employees to share praise or criticism about their colleagues.

At the company’s warehouses, the workers are also guided and monitored by software, but in a much more direct way. When you order an item online, the Amazon system quickly works out where the item sits in its inventory, and dispatches a human picker to go fetch it. “It’s not about learning where things are, in your head, or having to memorise,” explains general manager Henry Low. “We make the task as simple as possible.”

One of the first things that strikes you about the Amazon fulfilment centres is that the products aren’t organised logically – or at least in the kind of fashion that a human would use. For instance, products on shelves are not organised by category. Instead, they are placed on shelves as if by random. An HDMI cable lies near to five copies of some Harry Potter sheet music. A brand of baby’s bottle is across the aisle from a drain water diverter. But there is method to this apparent madness. “Imagine picking one model of HDMI cable from a shelf of hundreds of them,” says Low. The pickers are not meant to have to think too long about what they’re retrieving – the whole process is designed to be as streamlined as humanly possible.

Preparing for Christmas in Peterborough, Cambridgeshire (Credit: Rex)

When an item is collected by hand, the picker scans it with a handheld device to ensure that the correct object has been taken. And every item’s progress throughout the warehouse is constantly monitored thanks to a series of points at which it is scanned again – for example at the moment of being labelled with the customer’s name and address. “We are able to track where the item is at any one time at the fulfillment center,” says Low, who is both confident and clearly proud of the attention to detail.

The BBC’s Panorama programme has reported in the past on the high levels of accuracy and productivity expected of employees in Amazon warehouses. Reportedly, the pickers’ handheld devices in some warehouses count down the seconds that they have to retrieve the next product, in order to meet their performance targets.

Scanners instruct the pickers where to go, and reportedly count down in seconds how long it should take (Credit: Rex)

Initially, Low denies this is the case at Hemel Hampstead. However, when asked to see one of the scanners it becomes immediately apparent that such a countdown does indeed exist.

Thomas Owens, a picker at the centre estimates that he picks, on average, 1,000 items a day. “It depends on what process path you’re doing. But between break-times it’s normally between 200 and 300, so I’d say it’s about 1,000 or 1,200 a day,” he explains. In a 10-hour shift, that equates to almost two items every minute. Still, Owens adds that missing a handful of these countdowns is perfectly acceptable, and does not prevent pickers from meeting their overall productivity targets.

Workers in Peterborough prepare for 'Cyber Monday', the busiest online shopping day of the year (Credit: Rex)

Even so, the human element is arguably the weak link in the efficiency chain, and new computer-based systems that promise to automate Amazon’s operations even further might one day take over the ferrying of products themselves. Kiva Systems is a little-known Amazon subsidiary which develops hi-tech warehouse robotics. Instead of human pickers going to and from shelves in a large space, the shelves themselves are mobile and travel on wheels to stationary pickers who simply lift off the required item. Dutifully, the robot shelves then return to their place.

Another company, called Sick, has invented sophisticated sensing equipment that can be used in highly automated warehouses. Human eyes may be less and less useful in the future since, as a company press release about a visual sensor boasts, “the Sick Inspector P30 […] enables the automated crane system to pick and place more quickly, saving typically 10 seconds per pick (potentially 15% extra picking cycles per hour).”

Pickers, here at Peterborough, wander the shelves almost silently (Credit: Rex)

And yet, human involvement in the whole business of getting Amazon products from these massive fulfilment centres to your doormat is, for the time being, still crucial and evident. Human workers don’t just lift items off the shelves at the centre, they’re also responsible for packing products in boxes (the size of which is predetermined by an algorithm) and stuffing packing paper and vouchers in along with the purchased item.

Several rows of workers tasked with this part of the process are busily filling boxes, keeping packages moving onto conveyor belts at a steady pace. Later, of course, these packages will be driven to homes and businesses up and down the country by human drivers for delivery. Still, with Amazon now experimenting with the idea of delivery-by-drone, it’s not clear for how long exactly that part of the chain will go unmodified.

Items on shelves are not organised logically, by category, but seemingly at random (Credit: Rex)

When it’s time for a break, a notification sounds and the workers flood dutifully to the canteen, which seats at least 100 of them easily. There they enjoy the mid-shift break, eating, reading, chatting quietly.

“Work Hard,” said the sign near the entrance. On the warehouse floor, it’s apparent that the largely anonymous people who handle your online purchases take the first part of this mantra seriously.