Amazon’s PR offensive on worker training rolls on

Amazon got  a lot of positive buzz for its Amazon Technical Academy announcement. But before we all start leaping up and down, let’s unpack the announcement.

  • “Hundreds of Amazon employees have enrolled in Amazon Technical Academy since its launch in 2017.” Well, “hundreds” of enrollees over 3+ years is not exactly blowing the doors off. Amazon has 1.2 million employees.
  • How many complete the course? Oops, Amazon doesn’t tell us. Given that these kinds of programs typically have low graduation rates, I would be amazed if even 50% graduate.
  • Graduates to get software development jobs at Amazon, and their salaries do approximately double. Further, this is tuition-free, and Amazon offers a stipend to cover living expenses. All this is clearly admirable)
  • “$12 million invested in 2020” is not exactly busting Amazon’s staffing budget

If Amazon told us actual numbers of enrollees and graduates, I’d be happy to applaud. In the meantime, I’ll hold my fire, and suggest that this is a lot of PR over not much yet.

More on Bessemer

The recent unionization campaign at the Bessemer Alabama  warehouse has generated tremendous pushback from Amazon – tons of propaganda inside and outside the warehouse, efforts to eliminate voting by mail, even getting the Post office to add a postbox in the parking lot. And of course hiring well-known union busters to work on its behalf.

I’ve written elsewhere that even if the union wins in Bessemer, it’s a very long road, involving years of trench warfare, to unionize the rest of the warehouses. I hope the unions are successful in the long march toward a unionized Amazon, although for many reasons I don’t think they will be.

But there is one point I want to add. Amazon claims that “Our employees know the truth—starting wages of $15 or more, health care from day one, and a safe and inclusive workplace.” It does pay relatively well, and fulltime workers get health care benefits immediately (the much touted 401k and education benefits are more of a mixed bag). A spokesperson notes that it was the second best company in the world to work for, according to a Forbes (survey though that presumably mixes white- and blue-collar workers, and the methodology is a bit shaky anyway). It claims its warehouses are a great place to work.

But Amazon could easily prove that it;s a great employer for warehouse workers, simply by revealing the numbers that show labor turnover: how many workers last a year in the warehouse? What’s the average or median job stay? What’s the average turnover rate by warehouse? How well does it treat older workers, or disabled workers? it’s all in the numbers.

Amazon already knows all these numbers. If they were good, if they bolstered Amazon’s case, I have no doubt that Amazon would distribute them far and wide. But it doesn’t do that, so we can’t answer any of those questions. We do know that in 2017-2018 Amazon fired 684 out of about 2,000 workers at a Baltimore warehouse (revealed thanks to filing in a lawsuit). No doubt many more workers left voluntarily during the year, given the massive physical demands of the workplace.

So if Amazon really wants to convince us that its warehouses are a great place to work, it can easily do so: just provide the data. We can connect the dots. The fact that Amazon won’t do so is very strong evidence that its claims are more PR than physical reality.

Over to you, Amazon.

Into the Future: Amazon and the Coming Rise of Personal Healthcare

For 100 years, retailers have segmented the broad marketplace into groups – soccer moms, or Millennials, or New Yorkers, or readers…  pretty much every retailer uses target segments like those.  But Amazon has never been interested. Instead, it wants a segment of one: you. It gathers information about what you look at, what you buy, your browsing habits. To that it adds information from your purchases – address and credit cards on file – as well as your Amazon address book to find those close to you. It is also the second biggest tracker on the web, after Google, so it follows your activity far beyond the Amazon ecosystem. And it can access the standard sets of information that can easily be sucked in from outside: your credit score, your home ownership, criminal complaints and records…. Amazon  probably knows more about you than any other entity on the planet, including your mother and your spouse.

Put that aside for a moment.

Turn to healthcare. The pandemic has accelerated some trends here, notably telemedicine. That’s given a healthy push to an emerging field – remote diagnostics. Your phone is gradually – with some extensions – turning into a remote diagnostic tool to replace doctor visits and expensive tests. That’s fairly well known, even if it is extraordinary: blood oxygen levels are captured by a device costing less than $20, while an always-on cardiac monitor tracks heart activity, for example.

But that’s just the very tip of the iceberg. Remote diagnosis can transform the entire scientific basis of modern medicine. Currently, the gold standard for testing the safety and efficacy of treatments is the randomized control trial (RCT), in which some part of the trial group is treated while another part is not. Both are tracked to see whether the treatment worked, and to look for adverse events like additional illness or even deaths. Outcomes are assessed using standard statistical tools to compare the two groups.

This is the gold standard. But it is based on a single core assumption: that humans by and large react similarly to treatments, and hence that the best way to address disease is to identify treatments that work effectively for large numbers of patients. Ideally, treatments work for everyone, although sometimes RCTs and subsequent tracking find groups for which a treatment doesn’t work, or another for which it works especially well – maybe the old (or young), men (or women), or people with specific pre-existing conditions. Still, this is definitely mass-oriented  medicine: it’s based on the impact of treatments on what might be called the median patient.

Amazon is about to enter healthcare in a big way. It is already planning to offer Amazon Care (its primary care system organized around telehealth) not just to its 1.2 million employees, but to other employers as a service, much like it offered Amazon Web Services 15 years ago (and AWS is now the leading provider of internet infrastructure in the world).  It also purchased Pillpack and set up Amazon Pharmacy to deliver medications and other health products online for delivery. But the real revolution is coming inside the home. The Amazon Halo is a new health monitoring device (with some admittedly creepy privacy-related features). It is designed to apply the capabilities of AI to the needs of individual patients. And then there is Alexa, which is clearly going to be Amazon’s device of choice for wellness in the home. It’s already partnered with Sharecare to provide automated advice on 80,000 wellness and health questions, as a first line response to health concerns. Alexa will likely expand, gaining the capacity to integrate remote diagnostics and escalation into Amazon’s wider telehealth network.

Still, this mostly sounds like more of the same: using telehealth and the telemedicine capabilities, but mainly just extending what we do now, making it all a bit more convenient – possibly, a lot more convenient. But all these new tools, and especially diagnostics tools, make truly personal medicine possible – like Amazon, all this will be individual-specific: a market segment of you.

Let me tell you a personal story. I suffer from sleep apnea. I discovered this after going into a hospital for a sleep study. I then used a CPAP positive airflow machine while I slept, to push air into my lungs. A second sleep study confirmed this was working. A few years later, the in-hospital sleep study had been replaced by one I could do at home, using a machine the provider gave me for the night. It too confirmed that for that specific night the CPAP worked. But I hated it.

And then I found an app for my phone that tracked my sleep. More, it encouraged experimentation by tracking all manner of variables – whether I had had a drink before bed, the kind of pillow, room temperature – about 30 variables that might affect my sleep, along with a range of potential treatments. Every morning, I would check and the app would give me an apnea score along with other helpful sleep data.  So I experimented. After a while, I narrowed in on a small set of factors that made a difference for me. I used a dental device to hold my mouth open instead of CPAP; I tilted the mattress slightly; and I used a specific pillow. That did it! My apnea score dropped from an average of about 30 to 2.

Critically, I could now easily run endless experiments: I wasn’t limited by the need for expensive tech which had to be returned to the clinic, I didn’t have to depend on extremely expensive (and slow!) RCT’s that could only test one solution at a time. I could run new experiments every night, and I got new data every morning. And that’s transformative: I could move beyond large-scale RCTs, towards experiments that tracked treatment effects for me alone, assessing the impact of dozens of variables in the course of a few weeks. Al that was now possible because the data collection tools had advanced so dramatically.

The implications are enormous. This is not “personalized medicine.” It is personal medicine, designed by me for me alone. I have no idea if this works for others. But now imagine if Amazon was available to support this. Imagine if Amazon tracked the activities and outcomes for every experiment that every person ran using the app (or the Halo). Imagine if Alexa could say to you, “362 people found this to be useful, and another 201 found something different helps. Maybe you should consider using both these approaches, and let me know what happens.” Amazon can also crunch the numbers – I analyzed my own experiments, but Amazon could easily do that for you, becoming your own personal expert on your own specific individual health.

This approach is of course not limited to sleep apnea. It could be applied to many chronic diseases, like skin conditions, or joint pain, or muscular issues – or more serious problems like diabetes, where monitoring tools that work via phones are already available. In particular, it seems well-suited to managing digestion issues which plague millions of Americans. Currently, a lot of medical treatment is already hit-and-miss. Doctors give you something to try, then try something else when it doesn’t work. The difference is the granularity and the reporting. Further, the US spends billions on useless treatments, and this new approach is ideal for identifying expensive treatments that don’t work: for example, it could help people find specific supplements that work for them, eliminating many others.

In the aggregate, personal experiments could radically advance treatment. Imagine if Alexa was able to aggregate the data from thousands of individual sleep apnea experiments. It could quite quickly identify commonalities – groups for whom specific treatments worked best (or indeed worse). It could also rule out solutions that worked for no-one, and highlight those that might be prioritized. It would in effect be running an endless series of A/B tests (in the language of software development). They would not be RCT’s, although maybe subsequent RCT’s would be a useful way to confirm findings. But Alexa-collated experiments would be much quicker, much cheaper, and much more agile than any RCT.

Other companies for both the technology sector and healthcare have ambitions here too. But because Amazon has such an enormous customer base, very high levels of customer trust, and its existing mass of in-home health-related assets (in particular Alexa), Amazon is uniquely well-positioned to drive the next steps into the era of personal medicine.

Of course there need to be privacy controls. Amazon cannot be allowed to simply suck in all of my data and spit out ads, or worse still sell the data or use it when Amazon (inevitably) gets into the health and life insurance business. Personal information like this is very touchy, and HIPPA will probably need to be amended to protect us all.

But, on balance, I cannot wait for the day when my personal health care AI is waiting in my pocket to meet my specific needs.

Growing losses in Amazon Retail

Amazon’s most recent 10k filing with the SEC confirmed – accelerated – existing trends. While Amazon works hard to obfuscate results for its major segments, a careful analysis shows that Amazon Retail’s losses have accelerated every year since 2014, and reached the staggering total of $43 billion in 2020: that’s a negative operating margin of 21.8 percent, up from 20.7 percent in 2019, while retail revenues grew by $56 billion.

Source: Amazon 10k reports; author calculation. A detailed discussion of the methodology used is in chapter 9 of Behemoth, Amazon Rising: Power and Seduction in the Age of Amazon

Why is the ocean of red ink growing? There are a few obvious reasons:

  1. Too many supply chains to manage. By 2016, Amazon has 13 millions for sale through its retail division. That number has undoubtedly grown since then. Managing supply chains for many millions of items is beyond challenging.
  2. Extension into the wrong retail segments. Challenging IKEA on price is a fool’s errand. In too many segments, Amazon faces well established competitors with a better understanding of their business.
  3. The Marketplace challenge. Amazon’s Marketplace has attracted 2 million sellers. Many specialize in a handful of SKU’s, which they understand completely, and where they can compete effectively against the great white whale that is Amazon Retail.
  4. Chinese manufacturers. By opening the direct pathway between Chinese manufacturers and American consumers, Amazon has in effect disintermediated itself.

In my next post, I’ll explain how Amazon can shrink the red ink quickly and efficiently.

  1. Failed automation? Amazon has over the past few years tried to automate the product buyer function, replacing industry buyers through the “Hands off the wheel” initiative. There is no sign that this is being successful. In the contrary.

My new book about Amazon released!

Just a note to say that my book Behemoth, Amazon Rising: Power and Seduction in the Age of Amazon has been released in paperback and as an ebook. I hope you’ll give it a look, and if you buy it, do make sure to leave reviews at Amazon and/or Goodreads (the modern currency of the book world). It’s available at most bookstores and of course Amazon. You can find more links on my book page.

I’m doing a launch event as a roundtable at GWU with three other experts on Amazon at 12pm on Wednesday March 3rd. I’ll be speaking around 12,30 if you have limited time. Details here. To register, just send an email to They will send you the zoom link. The other speakers at least are very knowledgeable and entertaining.

Behemoth has had some initially favorable press, and I hope you can help spread the word, as I’m looking for more opportunities to reach out. Anyway, happy to talk more about it and to connect more generally – I’m now emerging from the long dark tunnel of authorship…

Amazon and Advertising: Watch out Facebook (and Amazon’s sellers)

Over the past four years, Amazon has uncorked a gusher of cash by driving up advertising on its Marketplace. Ad revenues are up from $1.7 billion in 2015 to $14 billion in 2019, and to a staggering $8 billion in the fourth quarter of 2020 alone – up 66% in a single year. This is oil boom territory, and Amazon advertising revenues have grown fast enough to make a noticeable dent in the Google-Facebook duopoly, accounting for more than 10% of digital ad sales in the US for the first time. Digital ads are the oil of the digital economy, and Amazon is clearly positioned right over a gusher.  But that’s not entirely beneficial – to sellers, to customers, or even to Amazon itself. Continue reading “Amazon and Advertising: Watch out Facebook (and Amazon’s sellers)”

Bezos goes, Amazon booms, but the real story is the Marketplace

Amazon has been all over the news the last 24 hours, with Bezos stepping down and Amazon roaring past $100 billion in revenue in Q4 2020. Those are both important stories – but under the hood, it’s the accelerating shift away from Amazon retail and toward the Amazon Marketplace that really matters.

The chart below from Marketplace Pulse tells the story. Total GMV (sales) across the Amazon retail platform reached almost $500 billion. Marketplace sales are up 50% – even though non-essentials sellers were deprioritized in favor of COVID necessities. So Amazon’s total sales have almost surpassed Walmart, and will do almost certainly in 2021.  Another year like 2020 and Walmart will be receding fast in the rear-view mirror.

But there’s an even more important important story here. Amazon is continuing it’s shift away from its own retail operations (Amazon retail), towards its enormous third-party Amazon Marketplace. Over the past 15 years or so, that shift has averaged around 2-2.5% annually. In 2020, the Marketplace share reached more than 60% for the first time, at 61.2%.

This ongoing shift is not surprising. Amazon retail loses money – a lot of money – as I show in Behemoth, Amazon Rising. Conversely, Marketplace made about $17.8 billion in operating revenue in 2019, a margin of 33.1%. It makes perfect sense for Amazon to transition away from its own retail business and into it’s increasingly dominant role as the platform manager.



Groceries: Amazon’s Afghanistan?

In 2017, Amazon bought Whole Foods. Since then, it’s opened a handful of new Amazon Fresh grocery stores and plans many more, along with Amazon 4-star stores, small cashier-less Amazon Go convenience stores, small Amazon Go grocery stores, and a few physical bookstores. It’s diving deep into physical retail, focused primarily on groceries.

Plenty of excitement, then. But sadly misplaced.

Amazon has no business being in groceries at all. Take it from one who knows. In his 2005 letter to shareholders, Jeff Bezos said this: “I often get asked, “When are you going to open physical stores?” That’s an expansion opportunity we’ve resisted. It fails all but one of the tests outlined above. The potential size of a network of physical stores is exciting. However: we don’t know how to do it with low capital and high returns; physical-world retailing is a cagey and ancient business that’s already well served; and we don’t have any ideas for how to build a physical world store experience that’s meaningfully differentiated for customers.”

So have things changed since 2006? Not really. Grocery stores are high capital/low margin projects. Unlike books and electronics and even apparel, there are no fat and happy bricks-and-mortar incumbents with tempting margins. On the contrary; the groceries pool is filled with great white sharks. So where are Amazon’s competitive advantages? There aren’t any:

  • Amazon’s huge competitive advantages online mean nothing in groceries. That enormous catalog and the great search and recommendation system doesn’t matter much… milk and eggs is milk and eggs.
  • Amazon’s extraordinary logistics network is tuned to goods, not groceries. Doorstep delivery costs make free delivery unsustainable, Amazon has no competitive advantages here anyway, and Whole Foods stores are not well placed to act as delivery hubs.
  • Amazon’s past record in groceries is poor: Amazon Fresh – its long-running grocery delivery service – has gained no traction. And Whole foods at best breaks even.

Amazon has no expertise in groceries, no track record of success, and will be running uphill against bigger and more experienced competitors with much better supply chains, and with deep established relationships with both suppliers and customers.

But what if Amazon builds the grocery stores of tomorrow, not today? Maybe Amazon is betting on the grocery business of 2031, not 2021. That means automation behind the store, through automated stocking systems like Ocado’s; automation within the store, through cashier-less tech like Amazon Go; and delivery to the home, likely through more AVs such as the Scout that Amazon is testing now in Tennessee. Automation will cut costs, perhaps significantly. However, it’s being heavily pursued by other grocery chains as well, so won’t provide Amazon with a significant or durable advantage.

Still, the real train-wreck comes from huge organizational and financial challenges. A medium-sized grocery does about $554,000 in weekly sales ($28 million annually).To match the scale of Aldi’s – the 4th largest US groceries chain with about $31 billion in revenues, Amazon would need about 4,500 average-sized supermarkets. But this not digital, with effortless scaling. It’s a physical business where every store needs a location, staff, local buy-in, and regulatory approvals. Every individual store is a challenge.

Assume that somehow Amazon sprinkles magic dust and builds a network of 4,700 supermarkets in the next few years. Running them is an enormous undertaking. Each store comes with dust and rats and physical problems to be solved every day. And there’s no evidence that Amazon has a secret formula: Whole Foods is break even at best.

Leaving the standup and operational problems aside, if somehow Amazon became as successful as Aldi, that still wouldn’t move the needle. Aldi’s 2019 profits were about 1.4% of revenues, or about $430 million. That’s barely 3% of Amazon’s operating income.

Amazon may have other reasons for getting into groceries. Maybe it just wants to prove and then license technologies. But Bezos is still right. Groceries is a high cost/low margin business with well-established competitors, where Amazon has no sustainable competitive advantages. It also requires huge startup costs and offers low returns. That’s why groceries could become a quagmire for Amazon: a huge long-term financial and energy drain, with low and uncertain returns. Perhaps Bezos should re-read that letter.

To receive an expanded version of this post plus other news, signup up for the weekly Amazon Rising newsletter at substack.