Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Amazon Built Its One-Day Delivery Supply Chain
Transcript
- Lucas: In April 2019, Amazon announced it was moving Prime shipping from two days to one day. Wall Street yawned. Logistics engineers, I think, felt their jaws hit the floor. Luna: Because two-day was already the gold standard. One-day meant cutting the delivery window by fifty percent across millions of SKUs. Lucas: Right. And the thing is, two-day delivery wasn't actually two days for a lot of items — Amazon had built an incredibly complex network where some stuff shipped from a warehouse an hour away, other stuff from across the country. The two-day promise worked because they had time to route things. One-day compressed that entire system. Luna: Before we go deep — I want to mention something. These episodes take a lot of research and time to put together, and we keep them ad-free because a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo. It's not a big ask, just something that keeps this going. Lucas: Yeah, genuinely. If you get value out of these deep dives, that's the place. And if not, no worries — let's keep going with how Amazon actually pulled this off. Lucas: So the core problem was inventory placement. Under two-day, Amazon could store items in a few regional fulfillment centers and still meet the promise. For one-day, the inventory had to be much closer to customers — ideally within a hundred miles. Luna: Which means you're essentially splitting your inventory into smaller batches and distributing it across many more locations. Lucas: Exactly. Amazon had around a hundred fulfillment centers in the US in 2018. For one-day, they needed to roughly double that — plus add a whole new layer of sortation centers and delivery stations. They broke it down into what they called 'inventory placement groups' — clusters of products that tend to be ordered together, stored together, near the customers who order them most. Luna: And that's where machine learning came in. They had to predict, for each SKU, where demand would come from at the zip-code level. Lucas: Right. They built a system that would look at historical order data, seasonality, even local events — and then assign each item a primary location. The algorithm would minimize the number of zones an item had to ship across. If you live in Chicago and order a book, the system tries to have that book in a Chicago-area fulfillment center before you click buy. Luna: But that's risky. If you over-predict demand for a particular item in a particular region, you end up with dead inventory. Lucas: And Amazon did accept higher inventory carrying costs. In their 2019 annual report, inventory increased by about 25 percent year-over-year. That's billions of dollars sitting in more warehouses. But they calculated that the revenue lift from faster delivery — plus the competitive moat — was worth it. Luna: And they also had to rethink the transportation network. Two-day could rely on UPS and FedEx. One-day pushed Amazon to build its own logistics arm. Lucas: Yeah, the timing was strategic. In 2018, Amazon launched its own delivery service — Amazon Logistics — and started leasing cargo planes. By 2020, they had over seventy planes. The centerpiece was the Cincinnati Northern Kentucky Airport hub — a 1.5 billion dollar investment, designed to handle a hundred planes on the tarmac at once. Luna: So you've got inventory placement, you've got your own air network, but the last mile is still the hardest part. How did they solve that? Lucas: They built delivery stations — small warehouses in urban areas, often just a few blocks from residential zones. These act as the final transfer point. A package goes from a fulfillment center to a sortation center, then to a delivery station, then to a driver. That driver might be a Flex driver — essentially a gig worker using their own car. Luna: And that's the part that competitors can't replicate easily. The density of delivery stations and the driver network. Lucas: Right. Walmart has stores that serve as pickup points, but they don't have the same last-mile coverage. Target acquired Shipt, but it's a different model. Amazon essentially built a logistics network that rivals UPS and FedEx in terms of reach, but optimized for one thing: getting a box to your door in under 24 hours. Luna: There's a trade-off though. This system increases carbon emissions significantly. Amazon has committed to net zero by 2040, but one-day delivery means more vehicles on the road, more air freight. Lucas: They've tried to offset with electric delivery vans — they ordered a hundred thousand Rivian vans. But the reality is, speed and sustainability are in tension. Amazon's own sustainability report in 2025 showed that transportation emissions were still rising, even as they improved efficiency per package. Luna: Let's talk about the economics. Analysts estimated that one-day delivery added about three to five billion dollars annually to Amazon's shipping costs. Lucas: And they didn't raise Prime fees — not immediately. They absorbed it, betting that increased order frequency would make up the difference. And it worked: Prime members started ordering more often, average order value held steady, and Amazon actually grew its market share in e-commerce from about 38 percent in 2019 to over 50 percent in 2025. Luna: So the architecture wasn't just about software — it was a physical network redesign. And the software had to control that physical network in real time. Lucas: Exactly. They built what they call the 'supply chain optimization engine' — a set of algorithms that decide, for every order, which fulfillment center should ship it, which carrier should handle it, and which route it should take. All in milliseconds. Luna: And it has to be resilient. If a fulfillment center goes down or a snowstorm hits a region, the system has to reroute inventory and orders dynamically. Lucas: Right. They use a technique called 'zone skipping' — where inventory is pre-positioned in zones based on probabilistic demand, and then orders are fulfilled from the closest zone. If that zone is disrupted, the system falls back to the next closest. It's like a distributed database with geographic sharding, but for physical boxes. Luna: One last thing — the impact on warehouse workers. One-day delivery puts enormous pressure on fulfillment center employees. The pace is faster, the targets are tighter. Lucas: There's no question about that. Amazon has faced criticism and unionization efforts, partly because of the speed demands. The engineering challenge of one-day delivery is impressive, but it's built on a human cost that's worth acknowledging. Amazon says it's invested in robotics to reduce physical strain, but the pace remains high. Luna: So when we look back at the 2019 announcement, it was a bet that reshaped not just Amazon, but the entire retail logistics industry. Lucas: And it forced competitors to respond. Walmart accelerated its own delivery efforts, Target expanded same-day, and even FedEx and UPS had to rethink their networks. One engineering decision cascaded through the whole economy. Luna: It's a case study in how a company can use software and operations research to turn a seemingly impossible promise into a competitive advantage. Lucas: And it's still evolving. Amazon is now testing drone delivery for some markets, and the next frontier might be hour delivery for select items. The architecture keeps getting more granular. Luna: I wonder if there's a point where speed stops mattering to customers. Or if convenience always wins. Lucas: That's the billion-dollar question. For now, the network keeps getting faster.