Amazon is most certainly automating away jobs

Plus some more important stuff that didn't make it into the Guardian article

I’m in The Guardian this morning on what Amazon’s robotics deployments mean for its human workforce. First a quick summary, and then some important stuff that hit the cutting room floor.

Amazon claims that it’s a “myth” that robots are taking jobs, and their total workforce size hasn’t dipped too much after the pandemic high, so it’s difficult at the macro level to understand the precise consequences of its automation efforts. But if we focus on those facilities where Amazon is primarily deploying its robots - Fulfillment Centers - it’s pretty clear that their workforce is shrinking dramatically where it’s automating most. Using Occupational Safety and Health Administration data, I’ve determined that employment levels have dropped by 25% over two years at Amazon Robotics Sortable (ARS) Fulfillment Centers, accounting for growth in productivity.

That’s the main takeaway, but there’s some other important stuff that didn’t make it into the final version of the article that I’m including here.

First, in the article I note that Delivery Station average employment has jumped from 208 to 250 employees in the past two years, a 20% growth that mirrors the 20% growth in overall package volume. Attentive readers will note that that number only includes Delivery Station workers directly employed by Amazon and not its vast, subcontracted Delivery Service Provider driver workforce - the only part of its logistics chain that it does not keep in-house. But looking at OSHA data on Amazon subcontractors, the average number of employees at Amazon DSPs also increased by 24%, from 71 in 2022 to 88 in 2024. These are not the most telling numbers, as DSPs come and go, and many DSPs will operate out of a single Delivery Station, but it’s still a helpful reference point.

So for last-mile delivery - a notoriously costly part of any package’s journey, and one that is pretty difficult to automate - employment levels at Amazon and its DSPs have risen by more or less the same amount as package volume over the past two years.

Second, if this were any other company, there would be good reason to be skeptical that the investments made in SHV1 in Shreveport, LA - the most automated of its Fulfillment Centers - will be rolled out system-wide, leading to an even more dramatic displacement of Fulfillment Center jobs. Put simply, it costs a lot of money and takes a lot of time to build and deploy these robots. The United Parcel Service has made a great show of its automation experiments recently, but a peek inside one of its recently opened “automated” distribution hubs indicates a noticeable absence of such machines. Robots are a great show for investors; it’s much more difficult to actually put them to use.

For comparable retailers, there’s traditionally a great reluctance to devote too much to warehouse automation. Warehousing costs are generally included under “Selling, General, and Administrative Expenses” (SGA) on Security and Exchange Commission filings, and this is in turn typically presented as a percentage of net sales. In 2024, Walmart’s was 20.4%; Target’s 20.6%; Home Depot’s 18%. They present it in this way because investors want to see that percentage go down, not up. With its warehouse-as-store model, Costco is perennially a winner here, and in 2024 their SGA percentage was 9.14%.

Amazon is the anti-Costco. It reports its expenses quite differently from other retailers, but in 2024, its Fulfillment expenses alone were 15.4% of net sales. That doesn’t include its General and administrative expenses (1.8%), its shipping expenses (15%), or the portion of its Technology and infrastructure expenses (13.9%) that contribute to the retail side of its operation, all of which would arguably be included in SGA. As an e-commerce company, Amazon is saved the expense of stores and retail labor, but it spends a fortune maintaining and improving its dynamic distribution network.

In short, Amazon’s built to change and “innovate.” Much was made of its recent $100 billion capex announcement (most of which will go to building out its AI infrastructure, but a quarter of which will go to improving its retail network), but that’s a one-time announcement. Technological investments are built into Amazon’s regular operation.

In one sense, this makes Amazon an even more daunting organizing target than Walmart. Not only is it willing to deploy all the same vicious anti-union stuff as Walmart, but it’s also a moving target, constantly opening and closing facilities, moving workers around, and now aggressively automating away jobs. But given the speed at which it has packages flying around the country, labor has structural power in being capable of disrupting the flow of goods through Amazon’s network. As I argue in a recent article in New Labor Forum, this is where Amazon workers’ power lies.