Amazon: Deconstructing the Retail P&L

Disclaimer: This interview is for informational purposes only and should not be relied upon as a basis for investment decisions. In Practise is an independent publisher and all opinions expressed by guests are solely their own opinions and do not reflect the opinion of In Practise.

Can we just start with a short introduction to your background, specifically, at Amazon?

I joined Amazon, leading their central pricing team, basically, covering the product pricing, across the consumer business, globally. That included subsidiaries, but with a principal focus on consumer. I then moved into global financial consolidations, for Amazon; how they put together operating plans, for the board and for the senior executives. This included proforma P&Ls, financial bridges, actual versus planned and planned versus planned analysis, including guidance analyses.

Following that work, I moved into global third party growth strategy and helped lead a 500 basis point acceleration in top line growth, relative to our goals, for the time that I was in that role. It was a plan to accelerate growth from 50% to 55%, year over year, in the third party business, and we accelerated the growth up to 60%, year over year.

How does the S-team analyze Amazon?

They divide the business into two high-level buckets; AWS and then the consumer business. When they are looking at the AWS business, it’s broken up into the product groups – as you would see on the AWS site – which are seen as the revenue centers for that business. On the consumer side, internally, they actually break the business out between North America Established Consumer and International Established Consumer. This is, principally, how they are looking at the business and, typically, doing so ex ads on the consumer side.

As investment areas, they separately break out North America Emerging and International Emerging business units, since they are, typically, not profitable on their recognized investment areas.

Why do they exclude ads from retail?

If you include ads in the core consumer business, you are unable to assess the health of the business, at its fundamentals. When you think about the core ecommerce, because the fulfilment network and those investments are so massive, there is a relentless focus on optimizing the opex/capex outlays associated with those operations. That includes those fulfilment centers, the staffing of those centers and the ultimate operation of the site, as a core business.

Since operating income, from the consumer business, is largely attributed to advertising – it drives an outsized proportion of operating income – and the S-team is largely looking at that business, in terms of its operating income, if you include the advertising numbers in the core consumer business, it entirely obfuscates how that business is operating independently of the advertising. You look at third-party sales and fulfilment to consumers; how profitable is that? Look at third party and fulfilment to consumers; how profitable is that?

So they aim to be profitable at a business unit level, such as FBA, third party, first party?

Yes. When you look at the break out, for example, of North America Established, firstly, you aggregate all the product categories and you break down that operating unit per unit, by separate channels; first party versus third party. Then further down, separate from that, you have your operation costs, per unit. You have the centers of the business that are driving the revenue and then, below that, you have the operations that underpin both the first and third party fulfilled parts of the business.

Taking North America, for the US, they report physical stores, third party and then first party. Would you include all of those in one revenue bucket?

No; when looking at North America, your revenue buckets would show as US, as a column – they would actually break US out of Canada – and a separate column for physical stores, as well. You are looking just at an established geography and the ecommerce business. Then you are able to break out, in rows, the different channels of the business and the operations that underlie those assets.

First party and third party?

Yes. Then you also have third-party non-Amazon fulfilled, also known as MFN.

Within third party, would you have FBA and then the marketplace business?

Yes.

How has the split evolved, roughly, in FBA versus MFN, over the last decade?

During Covid, there was a surge in MFN and they are very happy to see growth in MFN, in order to meet customer demand. As you know, there have been some significant supply chain and inbound constraints, during and since Covid, so the ability to provide customers with access to third-party fulfilment networks, for products that are sold on Amazon, was definitely a boon and much welcomed. In total, we’ve seen the third-party business continue to grow, between 200 and 300 basis points per year, in the proportion of revenue attributable to third party versus first party.

I think, within that third party, globally, it’s roughly 63% in 2021. How much of that do you think is FBA?

70% to 80%.

The rest is customers listing on Amazon, Amazon taking a take rate for, basically, lead gen?

Yes, exactly. And it’s fulfilled by those third parties, as opposed to being fulfilled through Amazon’s network.

Now they’ve got this supply chain business, where they are doing 3PL stuff – Buy with Prime – are they lumping that into FBA or is that a separate line?

That would be included in FBA, I believe.

Back to this point about excluding advertising, do you think it makes sense?

Absolutely. If you want to look at the core fundamentals of an ecommerce and you include advertising, you don’t know if you are actually a profitable ecommerce business or if you are a profitable advertising business. You need to look at both views.

Is there a target or goal for a certain margin, excluding ads, for those established ecomm businesses?

Yes, absolutely. The reason you want that ex ads view, largely, is to maintain strong control over your operations costs, per unit, which excludes returns, importantly; it’s just all the units shipped.

You don’t include returns?

Precisely. When you include returns, you net out the returns against the units shipped. At a high level, often, when they look at, how many units did we sell, that is net of returns. But if you are assessing your operations cost, you are actually just looking at the number of units shipped, versus the cost of shipping those units, regardless of how many returns there were.

Don’t returns distort the economics, if it’s high?

It does, yes. If you say, what is my operating profit, per unit, you can’t include all the things that were returned in that number.

Are the S-team looking at it on a per unit basis? Take North America Established, do they look at gross margin or contribution margin, per unit, excluding ads and excluding returns?

Yes. They look at the gross margin and they look at the operating margin, principally.

On this ad revenue point, can we look at it from a supplier’s perspective? Let’s say I’m a big clothing supplier, such as Levi’s, and I’m advertising on the platform and also selling wholesale, to Amazon and putting it in FBA, doesn’t my advertising spend come into that margin question, on the wholesale side, therefore, you have to treat them as one bucket, if they are going to flex them up and down, either way; lower margin, higher ad spend and vice versa?

When Amazon assess the business, they don’t look at it on a by-account basis; they look at it by geography and channel and then product category, at a high level. To answer your question, if they were looking at the business and they wanted to understand the soft lines category, which includes apparel, they would look at it based on that geography and channel, excluding physical stores and excluding advertising.

Typically, for example, when the ads team actually looks at the numbers, they are looking at revenue, instead of gross merchandise sales, and they are looking at total operating income dollars.

Going back to the top of the Amazon retail P&L, you’ve got geography, then down the rows you list whether it’s first party, third party, with third party split again by FBA and non-FBA. Then cost of sales which, I guess, is pretty straightforward and should be mainly the COGS on the consolidated income statement, minus some video and music spend which, I believe, is in that line too.

When they look at the consumer P&L, they entirely exclude video and music.

Why?

Again, in order to look at a pure e-commerce business. If you include video related advertising, royalties and so on, it is awfully hard to say, am I shipping it, effectively, as a business?

The cost of sales line, that has music and video expense in it, isn’t that just the amortization of the licensing and the video assets they have, for Prime?

Largely; there are different P&L views. You have your AWS P&L view, Prime view, with video, and music. You have a core consumer P&L view, which includes the physical stores in separate columns or completely separate, and the ecommerce business, both with and without ads. Those are the three high-level views that you have of the business. Then you can have consolidated financial statements.

Why would you exclude video and music, which clearly is an expense, to drive Prime subscriptions, if you're including Prime in the revenue for consumers as well?

You do break out Prime subscription revenue in the consumer P&L, but it is broken out as a separate line item.

You include it in the revenue as income, but you want to break it out so you can see what the pure ecommerce profit per unit is?

Exactly; it's all about getting a clean view of the business, in order to drive better decision-making. They're all about details, and they have the data.

If I'm going to look at the profitability, I should probably include the video cost spend, and if I'm including the Prime revenue, I should probably include an expense line with that?

Yes.

That is the cost of goods; you can break that out as a line just to see it clearly, then you get a gross margin. If I do that just for the last 10 years, to 2021, if I include the COGS and the video expenses, I get a 20.5% gross profit margin for retail. How would you split out physical, or have you already split out physical from that?

Yes, we already split out physical and you have to remember that physical didn't exist.

If I just assume that a 30% gross profit margin for Whole Foods and net that out, I get to 20% online retail profit margin, which has declined 200 basis points. It’s 22% to 23% over the last five years. Does that seem roughly in the right ballpark to you?

Yes. When you overlay those material fixed costs, your operating margins come down significantly.

Then we've got those two revenue streams; there's the size of ad revenue and subscription revenue, which is obviously a huge chunk. Taking a step back. Do you look at the ad revenue and sub revenue on top of that as well, or do you add that at the end of the P&L, to get some kind of gross profit margin, including ads and subs, which could be 37% to 38% last year?

That would typically go to the bottom of the P&L. They make a little mini P&L that says, where did we start, where did we end, what was the ecommerce line in there, without ads, and then, what was the ads?

What's your view on the average fulfillment expense as a percentage of sales, and the outlook for that?

It has materially increased over the past five years, with the speed of investments that were made, at the pace of something like $3 billion a quarter, just to get toward same-day delivery – especially in the US geography – and with the next day delivery and in other geographies as a desire to become the ‘kings of speed’, to attempt to maintain that as a differentiated value proposition. As a result of that investment, significant capex outlays, that didn't take into account slowing the ecommerce top line revenue from 17% year over year, to around 75 to 10% year over year.

Then the actual transport costs that have increased. Lastly, the $4 an hour increases in the first-line worker compensation, from an average of about $12 to an average of about $16 per person. That 33% increase that I cited a moment ago, across two million workers is a significant headwind to profitability.

That said, these decisions will enable Amazon to retain qualified talent, to uplevel that talent, in order to build veterans within the company. It will enable them to coast on that preceding investment during a recessionary period, now that that point of differentiation is established. They're very well positioned to outcompete during a recessionary environment relative to their competitors. But yes, it's trending up and should stabilize.

Back to this margin question, do they mainly look at contribution margin or gross profit margin per unit?

The individual businesses will look at contribution margin and contribution profit. However, at the higher-level view, they look at gross margin, gross profit per unit, gross profit, operating margin, operating profit per unit, and operating income.

How do you look at the difference between the gross profit of first party versus third party?

The first party includes the product cost of goods sold, whereas the third party includes revenue tailwind from the actual FBA fulfillment fees that are charged to merchants. That's principally how you look at the gross margin side of things. Below the gross margin threshold line, moving toward operating margin, you obviously have the same fixed outlay for your fulfillment. The same processes are applied so you can look at them in proportion to each other.

How does Amazon set the gross margin target per unit for first party?

That’s typically managed at the business level. Generally, when there's a separate operations team and operations finance team, they look at the overall Amazon fulfillment network, meaning first-party and third-party fulfilled by Amazon, also called AFN. They look at total shift units, as opposed to placing a special emphasis on what our shipping efficiency is for first party versus third party, since those shift units are going through the same business processes and they are warehoused in the same fulfillment centers and so on.

Let's just go down the P&L again. Revenue minus cost of sales, minus fulfillment expenses, gets you a contribution margin, per unit, for the retail business that's been minus 1.7% in 2021. If you look at the gross profit at 20%, which seems quite solid, the fulfillment expenses increased massively, which pressures the contribution margin. How do you look at the operating expense lines? For example, in 2021, if we look at the consolidated accounts, there was 32 billion in marketing expenses. Of that, 16.9 billion was advertising cost which is roughly half of the total marketing line item in the consolidated P&L. Is all of that advertising cost typically for consumer? Or is the pure ‘advertising spend’ attributable to AWS?

For the view that you are looking at, assuming that both AWS and consumer are reflected in that P&L – it’s an external facing P&L – you would have both consumer and AWS advertising expenses from the consumer side, and then you also have the Media Group advertising. The Thursday Night Football advertisements that are content in the States, for example, the Alexa business, the Super Bowl commercials, or the consumer business Super Bowl commercials.

It’s mainly consumer business advertising spend, that sub line item in marketing. The rest of the marketing expense seems to be salespeople, both selling advertising and selling AWS?

AWS sales marketing expenses are significant. You're renting out very large prestige venues. I think that Amazon carries a mega yacht on its P&L that's used for customers who are going to purchase AWS services. A great deal of money is spent on wining and dining.

Is that sales team for AWS in the marketing expense, or in the retail tech and contents expense?

That would be in the marketing expense.

The labor?

Yes. Your sales team would have a budget for client entertainment, in order to close those sometimes multi-billion-dollar contracts.

What do you estimate the split in marketing expense to be between retail and AWS?

You're talking about marketing ex advertising?

Yes.

Ex mass media advertising?

Yes.

I'd say probably 80% AWS.

It’s a big portion.

It is a B2B motion. You have the go-to-market expenses that are very common in any B2B sales motion, where you're trying to close very large government contracts or five to 10 year $10 billion contracts with the world's largest retailers and so on.

The total marketing expense in 2015 was $5.2 billion, and of that, $3.8 billion was advertising costs, which is consumer advertising, first Night Football stuff like that. As AWS has grown, you see the advertising mix of marketing has come down to only 50%.

In the past, Amazon was very strongly of the opinion that companies who advertise heavily have a lack of product value, and so they're trying to sell it through advertising. That was kind of the anecdotal perception of advertising history.

We can assume the rule of thumb is, if we attribute all of the advertising costs in the marketing line to consumers, and 80% of that non-advertising spend on marketing is AWS, that gives you some kind of mixed to retail. The tech and content line is 56 billion, nearly 12% of total sales. What would you estimate as a split between AWS and retail in that business?

{audio:32:49} Typically, that Alexa group and the other media groups are held separately, but I would expect that largely hit the consumer business, in the view that you're looking at, if the Alexa teams aren't split out separately in that P&L or the Alexa products and product development fit within the Media Group.

What else is included in the Media Group?

Amazon Books. It was originally a book company, and they have a pretty significant presence there Whether it's a physical, digital or audio book, it's usually bought through Amazon. There's Amazon Photos, which carries pretty heavy costs in terms of servers and supported unlimited photo and video storage, up to a degree. You have Prime Video, which is your main cost driver, and you have Amazon Music.

What else is in that retail tech and content spend? Would there be satellites in there, or the other stuff they're doing in crazy spend?

I believe Project Kuiper might be in there, but I can't guarantee that. I am just speculating. That's their satellite constellation for near earth internet.

How many people are working on that, roughly?

I can't quote the headcount estimates for those teams. Very few people internally get any visibility into those brand challenge equivalent projects. They're extremely confidential and usually have code names and you need special clearance to get anywhere near them, even in the same building.

Are they that secretive?

Extremely secretive. There are few businesses that are as secretive, I think. Everything is classified, data access extremely limited. I once was serving a request from one of our CFOs to get an understanding of EPS for a business unit and I pulled it from our system. I was one of the only people in the company to have access to that kind of information. Maybe there were six people who could actually access and pull that information. For other people, it would simply be reported to them, but I pulled it, and a week later my VP comes talk to me and says, do you really need that access and I said, I could give it up.

They are very secretive, and they have a very close sense of the pulse of which data is being accessed and how easy it is to decrypt, so to speak. Be it code names, be it privileges to certain folders. Any folder that you want on the company's internal server, you must request access for. Typically, when you are on board, you look at your peers and you request access to all the folders that they have access to, and you get denied a bunch of them. You get the ones that you need; it's highly partitioned.

How many people do you think are working in the Alexa group then?

10,000.

When I look at that retail tech and content line, it's a huge amount of money. How much do you think is in that line which you would call science projects, something like the satellite expendtiures, which is really pure venture spend? How much do you think in that bucket of tech and consent is really this VC kind of spend?

15% to 20%; maybe more. Let me help you view the business in the way that Brian Olsavsky views the business; he’s the CFO. All of the CFO, not just one of the many CFOs of this country and that country. The way Brian is looking at it is called the contributions and investments page. The first half of this page is all the contributions, in terms of operating income, and it's broken out one column North America, one column international, one column stores, one column total.

The rows are the business units, the actual businesses. You have AWS, consumer business and so on; all the businesses that are contributing money. Everything that is below the line is a science project, and you see that P&L view, that contributions investment page, net out pretty quickly.

The way you think about the business is, we're building some profitable businesses and we're going to make a bunch of investments. We're structured as the world's greatest entrepreneurial organization. Every team is no bigger than two pizzas can serve and they're all inventing all the time.

Do they purposely almost net these out? We make this much, let's spend it and they don’t want to show any bottom line?

Sometimes you make an investment and then you have the Alexa Group. They also invested the same $200 million initially, in the Amazon Fire phone, and that was a total failure. They also put 200 million in Alexa as the, ‘who knows if this thing is going to work’. It seems like it's more likely to fail.

How big is Alexa?

Many households have an Alexa device in the United States and the Alexa service, meaning the equivalent of the platform as a service. It's like a SaaS application, really. It is embedded in many consumer products.

You’ve got the marketing expense; we went through that. Retail tech and content, that's pretty much the biggest line item, obviously excluding fulfillment expenses in the retail P&L, and you think that 15% to 20% of that line item is satellites and various other kinds of lines, those rows below the where the profit is made.

Yes; they're new inventions that Amazon is undertaking, in order to create greater efficiency in the fulfillment part of the business. In order to create a greater top line and improve the online customer experience of business. You could think intelligent product recommendations and so on, and the AWS infrastructural pieces. All of these investments can be monetized through AWS. You can go to AWS and get their intelligent recommendations technology that's used across Amazon.

They're building innovations in order to monetize those innovations, so that they benefit both from the cost advantages of those innovations, the top line growth variable in their own business, and then thirdly monetizing them externally. The Go store is a very good example of this, so that technology also falls in that bucket.

How do you look at G&A?

It's a huge expense. G&A at is held corporate level and the two million headcounts that supports the fulfillment networks is held in the operations lines as those costs.

What’s the split between AWS and retail?

On the corporate side, maybe 65% AWS, and on the operations field fixed, and operations field corp and tech side, 90%, and first line workers on the consumer side of business are more than 90%.

The consolidated numbers are $8.6 billion total for the group what do you reckon? Would you split that evenly between AWS and consumer?

I think so. The dynamic is changing. The first line worker headcount is stabilizing and the AWS head count continues to grow.

Are the first line workers the people that work in the fulfillment centers?

Yes, exactly.

Isn't that in the fulfillment expense line though?

Principally, yes. If you exclude that again, you'll get just the corporate G&A side, as I mentioned before, maybe 65% on the AWS side, 35% on the consumer side. The AWS average headcount is more expensive, given software development engineers are commanding high fees. You can expect the cost of a single software development engineer for the $400K. That's a loaded cost.

I just assumed that the tech and content line included engineers for AWS, given that it was R&D or an expense for running AWS. Is that not the case then? That it’s actually just pure labor cost in G&A for AWS, those quite highly expensive engineers?

I think the latter. When you think about the tech, you have the Go stores, you have Rivian; you have very big investment. Machines that are secret that is being made to improve fulfillment center, picking, packing shipping experience, and trying to automate that, which requires massive investments in technology.

Let's just go back to the revenue lines. Let's just include all the retail revenue, minus the COGS, and you get a gross profit margin of about 20%, which seems about fine. Then you minus fulfillment expenses, you can net out Whole Foods, based on a 30% gross margin and a 3% to 4% EBIT margin. You get a contribution margin 8.5%, by my numbers, in 2015 down to zero or minus 2% last year.

On a contribution level, that seems very low. If you net out those costs based on what we just said, like marketing expense, advertising costs, retail tech and content – let's just say it's 60% to 70 % of retail tech and content is in consumer, which seems a bit conservative – and 35% of G&A, that is minus 20% EBIT margin, which excludes ads and subscriptions, but on a pure unit level, it's highly unprofitable.

I think it's unprofitable because it will include many of the emerging economies in which very large investments in fulfillment networks are being made. For example, if you take India and you think $10 billion there, and South America a few more billion dollars there, and then you say, what's the operating rhythm of your business and what does it look like? When these markets mature what you're moving towards because all of these are large investments. They're investing toward the future and they're being very aggressive recently in investing for the future. They want to be one of the top retailers in India; 1.3 billion, 1.4 billion customers.

At the consolidated level, we've got a 20% gross margin for retail. What would you estimate it is in North America?

25% and there's very good control over the operations costs driving toward your operating income.

Now, at a consolidated level again, you’ve got a 20%, 22% average in the last five years. The fulfillment expense, as a percentage of retail sales, has been between 18% to 22% in the last three-four years; what would that be for North America?

18% to 22%, did I hear you right?

Yes; 22% last year, and 18.6% in 2018. What would North America be, given how established it is?

8%.

It would be that low?

8% to 10%. They are a machine; they're so efficient. I've only worked in a fulfillment center for five days, as part of training, and it was extremely well run.

Let's just say for the North American P&L, you’ve got a roughly 25% gross – and these are estimates – and then, say, 8% to 10% is the fulfillment expense. You’ve got a 12% to 15% contribution margin for the US, and then you got marketing expense, retail tech and content and G&A, which obviously eliminates all that margin and you plus back. When you plus back subscriptions and advertising, do you treat those differently in how do you add them back?

No; they're just single-line items internally. When you're saying, S-team review, they want a simple 13 or 14 line view of the entire consumer business, as an example. There will be a single line for operation, a single line for marketing, a single line for Prime subscription revenue and so on. India was being held separately, as a single line, because it was such a material investment area.

How much do they spend roughly in India?

I think it might be available online, but I can't pull that off top my head. Maybe $10 billion a year or more.

That much?

Yes; it's really expensive to set up, and it's an enormous market.

There's just so much spending on these line items that are not actually for the US ecommerce.

Yes, exactly. But the future is not necessarily in the US for that business.

What do you think is a stabilized EBIT margin for the ecommerce in the US, for Amazon?

All up?

Yes.

Maybe three years from now, it's 5%.

Including ads and including subs?

Including everything. Maybe it's 1% right now?

Yes, I get 1.2%.

On a quarterly basis, last quarter, probably negative; fourth quarter, positive.

Yes, but it's amazing, because people question the profitability of this business. I don’t get it. Investors really question it; they think it's a shitty business. They think retail ecommerce is a crappy business; they're spending so much.

It is one of the very best-run businesses that I know of. It’s working backward from customer need and, if anything, on behalf of the customer, with deep data-driven research, and then driving really efficient internal innovation. If you were going to bet on a startup, that's the world's best startup. I think Mark Hewitt said that as well, so I agree with Mark. Although he never worked at Amazon, so what does he know?

What about working capital? How does the networking capital profile work for consumers and AWS?

For the consumer businesses, especially, there's always been an obsession with free cash flow. I believe what was once called the something to the effect of a negative operating rhythm, where you take in money, but it takes you longer to pay it out than it takes to take it in and so you're sitting on this free cash that is meant to be accumulated and reinvested before it's paid out again. Remember that Bezos was an ex-finance guy, and so he was all about ramping up free cash flow and redeploying that capital effectively, in the meanwhile. When you think about gross margins, the AWS business is a completely different profile from the retail business; massive margins. When you think about operating income five years from now, I'm betting that 75% to 80% of operating income from the all-up business will be from AWS.

You can't grow at 40% year on year, with a $40 billion run rate, and not be colossal in five years.

Just so I’m clear, if I look at the networking capital, like you said, of course, the retail business generates cash because, with the Prime subscriptions, you pay suppliers later. What's interesting is, if you look at the cash conversion cycle, which is 100 days of cash they've got on hand, effectively, in 2001, it was 30, 40 days. Now it’s half of that; 17 days.

They are becoming more seller-obsessed as well. They have begun measuring the health of the third-party business based on seller net proceeds, meaning the money the sellers put in their pockets after paying fees. They give it to him quicker, they try to give them more. Yes, they're increasing the fees but they're trying to apply a lot of internal pressure to stop those fees from increasing.

How is the growth of AWS affecting the working capital and how much free cash do they have, based on paying suppliers, given that AWS sell on-demand instances; they don't actually receive the cash.

They kind of pay themselves. There’s been a big focus on shifting the business toward AWS obviously. If there weren't, JAW would have been assigned CEO instead of Jassy. Jeff Wilke was the CEO of consumer business and Andy Jassy was the CEO of AWS business. Andy Jassy is now the CEO of the entire system. There's a big bias toward that, and that's the cash cow and it will be in the future.

Let's say you got the EBITDA margin, to get to free cash flow, how did the change in working capital work? Do they eat cash as they grow or do they earn more cash as they grow? For AWS, it seems as if that could be requiring a bit more cash to grow investment, before you receive it from customers, given they might be paying on different term structures than Prime and suppliers and sellers.

They do. You have a less predictable cycle for AWS because these contracts are negotiated as enterprise agreements or as a customer success partner agreement, where you have a third-party partner who is maybe reselling or building on top of your platform and maybe you're making a marketplace to build on top of your platform for these companies. There is a big emphasis on spending capital on sustained innovation to remain competitive, in the AWS business.

Just for this working capital bit, one of the things I love about the retail business is that, let's just say for just for argument’s sake, if you’re breakeven on an EBIT basis, for retail, you add back some depreciation, amortization and let’s say it’s 1% or 2%. Then you've got the working capital inflow that you get as cash, from Prime, from not paying supplies quick enough. You actually get 1%, 2%, 3% FCF margin. Over the last ten years, before they had AWS, my numbers show the change in networking capital on the cash flow statement. It’s been 3%, on average, positive. Basically, this is the cash that they get that adds to free cash flow. As AWS has grown, it has been far more unpredictable and actually lower, at 0% to 1%.

You also have different arrangements on those enterprise agreements. The same is true with competitors of AWS. It is very common, in the industry, to offer consumption commitments, where customers promise to pay X over X period of time. Basically, if they don’t spend it all at the end of the period, AWS gets paid, and the customer loses out on that usage that they paid for. Sometimes you have those and then you have monetary commitments as well; we will spend this. But you don’t necessarily receive the capital; you receive a commitment to that spend, down the line. But you have already put up a very high cost of sale, upfront.

The working capital eats a bit of cash, AWS versus retail, but obviously much higher margin so it’s hugely profitable anyway.

Very expensive to make a sale in the business and to build a technology in order to close the sale in first place. All the engineering teams and upfront expense to launch a product then slow ramp for the product. That’s software sales. If you have 80% at a 50%, you are doing okay. You are doing just fine.

What are your estimates, roughly, for long-run gross margins for AWS?

Probably 80% and a 50% discount from their list price.

What do you mean 50% discount?

When you negotiate those enterprise agreements, you give huge discounts. Because you negotiate with enterprises; they fully expect it. The salesforce can discount up to 80% or 90% off the list price. The point is to land the sale and get the customer into the ecosystem; land and expand. You get an infrastructure sale; they’re on your data models, on your servers; your service is sticky all of a sudden. Now they have started adopting your SaaS application, your partners’ SaaS application, your new infrastructure applications that you launched. Often times there's a willingness to deeply discount to land the initial contracts, and then you build out from there.

What do Jassy and Brian look at on the AWS business? Going back to the revenue line, on the consumer business you had first party, third party, FBA and MFN. What is the equivalent for AWS?

They're going to break it out based on infrastructure services, software services, and platform services. Sometimes those are hybridized into a single product. You might have a broad group where you will have a set of AWS services, in terms of the AWS categories, and you can reflect Amazon, overall, in terms of cloud products, for example. You have the bucket for that. You can have your AWS marketplace, a high-level bucket, as well. You can still have the first part and the third party breakdown of your business. But on your first party side, you are going to emphasize categories such as compute storage, the actual database side, that includes data in egress and transfers across customer tenants that maybe in different countries. Networking, content delivery, which is expensive, analytic services which are abundant, machine learning and then the security piece. These businesses are making tons of money on their security businesses.

Would that be around 10 or 15 line items in AWS that they are looking?

Yes, and then you can break it down to another 30 line items. All of a sudden, you have blockchain, you will have IoT, you will have management and governance inside of your platforms. You have satellites, robotics, quantum technologies. But those are the high-level buckets; the compute store, database, analytics, machine learning, security and networking.

What’s, roughly, the largest in that?

I don’t know. It is likely compute. Compute is expensive. You are processing and transforming data. Storage is a big one, but storage is table stakes.

What about databases?

Yes; that’s huge as well.

How do they look at the cost structure? Obviously, in the consolidated accounts, they just provide the EBIT which obviously is 30% margin.

I am unable to speak to that.

The long-run EBIT margins, how do you think they are going to evolve, five years out?

They will look much the same when you specifically look at AWS. There will be a product mix shift but the compute storage database pieces will still be the highest revenue contributing. If let’s say, hypothetically, they went big into satellite then your operating income may decrease but it depends on how they are going to leverage those satellite technologies and how quickly they can monetize them and spread the cost, but that is largely cloud, compute and networking.

Would the satellites spend be in AWS or in the tech and content side?

Part of it is an AWS and part of it is in tech and content.

Roughly, those three segments, EC2, S3 and the databases, do you expect those to be 80% to 90% of the AWS revenue, in the long run?

It will continue to diminish, as a percentage. The mix shift will go more into AI, ML, quantum products and security.

How does it impact margin?

It increases margin. You are going from something that does not have a differentiated value proposition to something that does. It’s all about their ability to invent successfully and redeploy those tech expenses.

What do you think is the biggest challenge for AWS margins? Because people argue that it’s going to be commoditized in the long run.

You’re right; part of the business will be commoditized in the long run. They always plan to deprecate business units in the long run. This is why they keep doing this investment that we are talking about. Why are they spending all this money? If they stop spending this money, the investors would be right, over and over again about this business. They are not going to be profitable in the long run; it’s going to get commoditized. Yes, absolutely it will, which is why they are making the satellite and blockchain investments.

Do you think they will ever change?

No; it is the core ethos of the business. That is why it will continue to grow.

It is almost like the they purposely spend. They would rather not show profit, then actually show profit, sometimes.

They would rather surprise and delight.

But they can show a little bit here and there.

Yes, they could flex up their profitability easily. But they are not so interested in Wall Street, which is so perplexing about the entire business. Investment management organizations reach out to me to talk and they say, what are they doing? Why are they spending money like this; it’s crazy. It is the same thing over and over again and it has been for 20 years with this business. But they continue to operate as amazing entrepreneurs.

What are Brian and Jassy like? Do they really not care about Wall Street?

they care; they care very much. But they are relentless, and they are focused on the efficient operation of entrepreneurial organizations. You get measured in every way that you can possibly conceive of and then ones that don’t exist yet, when you are proposing a project. If your project proposal is really good, the document gets reviewed the next year and the next year, and there are only a few tiers between. Between tier level four and tier level 10, that is basically all the tiers you will see if you work at Amazon. And you will go through these doc reviews and then you are launched with your measurement mechanism and then you will be relentlessly measured against them.

Is it the memo kind of thing? A six-page memo?

Right; they are so diligent around the investigation, launch and the successful and efficient execution of these proposed business models that come internally. That’s why people burn out.

One thing we found is a very high turnover at Amazon. People last two or three years.

All the other companies I have ever partnered with or worked for, or provided services to, none of them operate with this kind of intensity. It is not just that you put in long hours and that you work hard, but you worked at a super human intensity, at Amazon. If you don’t, you are performing poorly.

The other thing is about the server lifetime. They keep increasing the server lifetime at AWS, which I think is interesting, pushing up the margin. How long do you think they could last? Now, it is five or six years. Do you think it could be seven or eight years?

That is a tricky question for one reason, really. The utilization of those servers is going to change in the next five years. You are going to have AI or ML services that need high power compute and so you are going to have to make significant outlays for your new servers in order to handle HPC, high power compute. That said, you can charge pretty handsomely for that. It is seen as a more differentiated service than core compute. With that said, you also have to create new types of servers to accommodate quantum and then, down the road, quantum learning, which is the intersection of deep learning and quantum computing, as those new algorithms and hardware emerge. They are going to have to make very substantial server investments and they will be able to monetize those handsomely, as well.

How much of the AWS revenue do you estimate is from the Amazon retail stores?

I don’t know about and, even if I did, I don’t know if I can disclose any kind of internal transfer pricing arrangements that they have. However, I would assume 5% to 10% of AWS’s total revenue is attributable to the consumer business. Everything is run on it; all the servers. AWS was built by the consumer business because it needed those services.

Maybe we can assume that is charged at cost.

I would probably would not assume that it is charged at cost.

They wouldn’t do that from a cultural perspective?

Yes, I think so, because of frugality. There is relentless focus on the business intelligence engineers to efficiently manage the Redshift clusters. Almost every team has got a BIE, who is focused on making sure the clusters are running efficiently.

Even when you got your own internal compute stack, you cannot even get it at cost? You’ve got to pay full price, at Amazon?

It’s all about applying pressure. But at the same time, it is so beautiful because you can really learn how to run a business.

How old are these guys? Jassy is 60, odd. Are they actually this relentless? Are they this intense?

No; they are more intense. The intensity comes top down. This high turnover, you build a culture of intensity by your top people. They talk the empathy talk, etc. but the people they put in charge are most fearsome. When your leaders have a penchant for sending question marks and that is it to groups of people; you would send an email and it would just be a question mark.

it would just set off a fire storm. You have that kind of behavior informing a lot of the cultures which some people could describe as toxic. I noticed this and I would like to better understand what underpins this scenario. Because a lot of things happen for a reason, that superficially look bad. But sending question marks is not an appropriate way to treat your people. You should be curious before you are furious and you should treat them with empathy and respect. My current employer and my companies do that beautifully and they are able to retain people for longer than two years, as a result.

Looking at the long-run margin, you mentioned that 5% for the retail business, all in, with ads and subscriptions. What could be the biggest risk to that, in your mind? If we look back, in five years, and it is still 0%, why would that be?

I am actually thinking about governments, government policy and domestic protectionism around technology. I have made some very fine investments over the years. In Amazon’s investments, they liquidated everything they did in China, basically, from the consumer business. If India goes in a similar route, the very large capital outlays they have made to be successful in the consumer business, over the long term, would be in very much jeopardy and could be devastating to consumer business. The same is true, and perhaps even more true, for data residency and cross-border data transfer which can bravely impinge AWS’s ability to scale up their operation globally and effectively serve multinational corporations, which is the fourth customer they are targeting. Really, it is a technology piece where you have protectionism over the marketplace, meaning the consumer business, and protectionism over the desire to form domestic technology companies that are able to serve the needs of customers in these more innovative areas where the AWS runway is. You know it is not in store and so on; it is in the AI/ML services and a lot of those we instantiated where the customer is.

What do you think investors most misunderstand about Amazon’s P&L and unit economics?

They see what the P&L is today. They see what it has been historically, but they don’t see and they can't see – it is purposely hidden – the investment areas. They don’t know what is necessarily going to happen with Rivian or Project Kuiper. They don’t know that Amazon is attempting to cure a common cold which may or may not have been one of the grand challenges in the past. Amazon really does try to tackle very big problems.

It could be anything they are doing, like healthcare stuff, cancer or colds?

Yes, and they have a gift for it. These micro teams are full of entrepreneurs. You have to look at Amazon as this company that is investing and its P&L today is a reflection of the capital outlays it’s making to dominate the next seven to 10 years and what the business will look like then. It is very hard to get on board with a business that is investing in the future, as opposed to solely stabilizing its current operations.

It is almost like they are trying to find the next AWS.

That is exactly what they are doing and that is exactly why I shared my contributions and investments analogy. They are trying to find the next AWS. They have always realized that complacency leads to death. There is a nice Bezos quote on it about what day two looks like; you should look it up. That is the ethos of the company; it is that fear of day two.

Does that means you have to go and cure cancer, because it is day one?

Exactly; you have to do something. You have to invent and that is the pressure that is put on the employees to make them breakthrough to surprise and delight customers.