It’s fair for Thomas Hoe to describe his career path as “super nonlinear”. After all, his career trajectory has veered from prize-winning video gamer (before that was really a thing), to computer science student, to chef in Michelin-starred restaurants, to economics PhD, to leader of UK-government economics teams, to healthcare researcher at Cornell University before, finally, he launched Amazon Stores’ first European economic-science team.
But there is a theme to this two-decade rollercoaster: obsession. When something catches Hoe’s attention, it catches all of his attention. It was in 2019, as the UK’s looming departure from the European Union was catching Amazon’s attention, that a company recruiter collared the then assistant professor Hoe after his presentation at the Econometrics Society’s annual conference in Seattle. That meeting ultimately led him to join Amazon as the first dedicated economist in its European Stores retail arm.
“I was intrigued, because Amazon ticked a lot of boxes for me: big data, economics, optimization, problems that require deep science work, and the chance to work closely with executives,” says Hoe.
Brexit represented a challenge for Amazon’s European retail business, so it was a baptism of fire for Hoe. “I modeled all the increased investments we would need to make to reduce Brexit’s impact.” His work played a key role in enabling Amazon to continue meeting the wide variety of its customers’ needs while seamlessly shipping many millions of orders across the new Brexit border. Not a bad start.
One size doesn’t fit all
Brexit helped to reinforce the need for local economics expertise in Europe, says Hoe. The retailer’s economic-science teams are largely US based, so its processes and decision-making systems are typically built and optimized for the US initially, then adapted and rolled out across the rest of the world.
Europe is very fragmented and there’s different — and more varied — competition.
“They work very well, for sure, but one size simply cannot fit all,” says Hoe. “Europe is very fragmented, and there’s different — and more varied — competition.” His now well-established team, Economic Decision Science (EDS), addresses uniquely Eurocentric challenges using the latest techniques at the intersection of microeconomics, statistics, and machine learning.
Back in 2021, though, with several quick wins — and a few misses — under his belt and his first few economists hired, Hoe was looking for ways to have an impact. “We felt like an economic-science startup within Amazon. We were the weirdos in the corner,” says Hoe. The European business teams weren’t always sure what to make of Hoe’s EDS team, nor how and when to integrate economists into their problem solving.
In order to demonstrate where its value lay, the team started solving problems that no one was asking about. In 2020, for example, their analysis found that customer demand was much higher among third-party sellers that provided free-shipping offers. Based on this research, Amazon shared the insights from this work directly with sellers to help them optimize their Amazon offers and be more successful on the store.
Economic Decision Science in action
Now, four years later, the EDS team is in demand. “What I like best about my job is the huge breadth of opportunities, and the challenge of trying to identify the projects that will have the largest long-term positive impact on the business and on our customers,” says Hoe.
For example, at the start of 2023, Amazon’s U.S. fulfillment network successfully restructured into eight largely self-sufficient regional networks. In Europe, the network had historically transformed in the other direction, originally being individually built across the five biggest European Stores — UK, Germany, France, Italy, and Spain — and later being sewn together into a single European Fulfillment Network, enabling goods to flow in a frictionless way across countries. Hoe’s team has been working on ways to optimize this unique European network.
A recent focus has been trying to cut down on long-distance fulfillment when equally suitable versions of products are available much closer to the customer. Think of a USB stick that might be shipped from France to fulfill a UK order, when a different but similar option is sitting in Amazon’s UK warehouses.
With European countries more densely populated than the US, customers often live closer to physical competitors, so our value proposition can differ. We need economic science to help us understand where and when customers value faster deliveries most.
The EDS team developed a model, incorporating data from Amazon’s US-based Supply Chain Optimization Technologies team, to explore what would happen if Amazon re-optimized its European ordering systems so items are placed closer to customers in the network, avoiding cross-country shipments. Their finding? Customer choice can be maintained but costs and shipping times slashed. That is good news for both Amazon’s carbon footprint and its customers, as the savings allow the company to invest in making even more products locally available. Hoe’s team is continuing to test its model and hopes to roll it out this year.
Other differences between Europe and the US are the way in which customers engage with online purchases and the alternatives available from physical stores. This makes understanding local customer preferences critical.
“Part of Amazon’s value proposition is the huge convenience we offer to customers by saving them a trip to the physical store and offering a range of fast delivery options. With European countries more densely populated than the US, customers often live closer to physical competitors, so our value proposition can differ,” says Hoe. “We need economic science to help us understand where and when customers value faster deliveries most, to make sure we provide that.”
Hoe and his team are currently building a model to help them understand customer expectations for online-delivery speeds and the level of convenience Amazon must offer to customers in Europe. “Amazon is continually investing in faster deliveries, and we want to make sure those investments are delighting as many customers as possible,” says Hoe.
Europe’s economic insights ripple out
While the EDS team was created for European impact, Hoe is keen that the fruits of its projects start to flow back to the US and beyond. Consider its creation of a machine learning algorithm trained to highlight the best deals for customers across Amazon’s vast inventory.
First, Hoe and his colleagues needed data on what customers thought of various combinations of products and prices. So they surveyed a large swath of Amazon customers, asking them a total of six million hypothetical pricing questions. The team fed this big chunk of customer feedback into a machine learning model, which taught it how people perceive the value of a range of Amazon products and prices. But here’s the special sauce: the model trained on customer preferences could then be applied to millions of live products across Amazon’s European inventory, looking for more instances of particularly attractive pricing. It’s akin to having the voice of a customer telling you where all the best deals are.
“By directly incorporating customer perceptions into our algorithms, we’ve consistently found that we can display a compelling selection of products that increase customer engagement,” says Hoe.
Some of the products identified by the model could then be highlighted in customer searches, on the Amazon home page, or in marketing campaigns to help even more customers find the best deals on Amazon. After several iterations of successful prototyping in Europe, the technology has recently been trialed in the US. “I love that we’ve got innovation going the other way now,” says Hoe.
Now that Hoe’s EDS team is established and its capabilities in demand, Hoe considers it a success that Amazon’s European teams have a clearer understanding of how economic science can help them tackle the unique business challenges that they face in Europe.
“Even at Amazon, where we have some of the most advanced systems in the world, economic science is still in its infancy when applied at this scale,” says Hoe. “We’re excited about the path ahead.”
