Turbocharging the Internet Fulfillment Revolution
Sanchoy Das recalls his first jaw-dropping view of an Amazon fulfillment warehouse in 2013. The word “explosive” came to mind. Gone were the single-product sections where boxes of diapers and paper towels were stacked next to each other in neat rows. Instead, these products were dispersed throughout the building in hundreds of different locations, tucked into bins with unrelated items such as ketchup and motor oil.
But what looked like chaos, he learned, was in fact a sophisticated model optimized for speed. By offering free, expedited shipping, Amazon’s orders mushroomed. The e-tailer responded by building in efficiencies to compensate for the lost fees, as it shaves days, hours and minutes off fulfillment times, continually asking the question: “How fast can I get this order fulfilled?”
“This was an entirely new paradigm,” notes Das, a professor of mechanical and industrial engineering who devises optimization and analytical models for sectors ranging from retail supply chains to hospitals. Because Amazon’s models are proprietary, he set out to develop his own algorithms — for stocking, picking, packing, truck assignments and distribution, among other operations — to understand them and to create new ones other companies could use.
He and his research team visited Amazon fulfillment warehouses in Wilmington and Indianapolis to conduct an observational study and figure out the data, information, human and product flows throughout the system. He discovered that Amazon was capturing every detail about customer orders and then integrating this data into powerful decision models, which controlled every activity in the fulfillment supply chain. For example, they realized that item demand was not just an aggregate monthly number, but tracked by day of the week and time of the day, and then correlated with the demand of hundreds of other items.
Probability models are critical to increasing the efficiency of these warehouses.
“You want to accelerate the speed with which pickers fulfill orders, and that implies increasing the probability that ordered items are stocked close to each other. The company’s stocking algorithms are designed so that a picker can collect the 20 items she’s assembling within a few aisles. Her list pops up on a device that directs her to a succession of bins,” Das says, adding, “To increase delivery speed, however, orders might be split between picklists and warehouses in different locations. Consolidation algorithms put orders together, while minimizing the number of parcels shipped.
“I’ve seen a huge change in retail supply chains over the last 15 years, as customers increasingly buy products online, and expect free shipping and next-day fulfillment,” he notes. “This requires very smart data-driven supply chains. Amazon, for example, uses deep analytics to predict what orders will come in today.”
Competitors such as Target, Walmart and The Home Depot have spent several years trying to catch up. The capital requirements to build fulfillment warehouses with integrated robotics and automation systems, and the machine intelligence to run them, make it a challenge.
“Even today, most warehouses collect a few days of orders and then plan on how to ship them over several days, while Amazon warehouses operate in real time, making thousands of decisions every minute. It’s no longer a level playing field,” observes Das, adding, “Our research goal is to develop new control models and algorithms and make them available to a wider variety of companies. These include warehousing companies like Prologis, and small and big retailers like Walmart, Kohl’s and even the Dollar Store. Additionally, eCommerce software service companies, like Shopify, are expanding into the physical fulfillment side, and the models we are developing could be used in their new warehouses.”
Das says he assumed the gap between Amazon and its competitors would widen over the pandemic, but it did not. Many “shook up” their supply chains and used a lot more data and decision models to manage the shift from store to online buying. Indeed, over the first five months of the pandemic, The Home Depot reported a 100% increase in digital sales with 60% fulfillment from its stores, Walmart a 97% increase in digital sales and Target a 273% increase in same-day fulfillment digital sales, he says.
Disruptive new companies emerged. As the pandemic closed restaurants across the U.S., innovators at Brinker International, which owns the Chili’s chain, developed a new, virtual, pure-delivery restaurant called It’s Just Wings that leveraged the underutilized kitchen capacity of associated restaurants and the fulfillment network of DoorDash.
“Were these fulfillment systems not in place, the effect of the pandemic would have been a lot more severe for a lot of companies,” Das says, also noting that 52% of the products Amazon ships are from third parties. “Speed — the necessary condition — is what determined the victors.”
Going forward, he says, more retail will go online.
“While some shopping experiences can’t be replaced, I think we’ll see fewer commodities purchased in person, because these items have no in-store purchasing excitement. Think groceries.”
