CFP Fall 2013 Plenary Money Flows in the Internet Ecosystem, David Clark, MIT Dave Clark: I’m actually giving one long talk, but I’ll present it in two halves with a break in the middle. My goal with this talk is to start some conversations. It’s a talk about money, and the motivation for this is a conversation that has come up from time to time. It’s come up in conversations with with ISPs – more often European ISPs than American ISPs – which is that they look at all the content flowing through their networks, and they look at all the people attaching to their network and they say ‘Why is all the money coming from the consumer? Maybe the ecosystem should have a different structure, in which there should be money flowing in from some place else.’ The European proposal, which didn’t in fact get to the ITU, is basically to impose a senderpays regime on international interconnection. To put it crudely, it was motivated by the sense that, ‘There’s some money out there, and we’d like some.’ And so I asked, ‘How much money is out there?’ If you look at the ecosystem, aside from money that consumers spend on things like paid content and e-commerce and so forth, there’s really only one other source of money flowing into the system, and it’s advertising. I wonder how much advertising money there really is? I wonder what the role of advertising in the Internet ecosystem is? I persuaded a very good grad student (Israel Valentin) to go hunt some up for me. He did all this and I thank him very much, and if you ask me a question I’m going to point at him. But also, Israel is doing some industry interviews and talking to people about strategy and I would love to introduce him to a few of you because he’s had some conversations with European ISPs (he is from Spain) and he’d love to talk to some American ISPs as well. These numbers, unless otherwise stated, come from an organization called the Interactive Advertising Bureau, the IAB. In 2012 the interactive ad spend is $36.6 billion dollars, or $34.63 per month per household. I had this sort of bimodal reaction to this number: First, that’s a lot. I’m worth $34 a month to the collective advertisers in the world? I’m not worth that much.’ Second, where does that money go? It comes from the advertisers and its essentially paying for what you might call “All of the free internet experience” – every website out there has an ad on it or does behavioral profiling—that entire ecosystem is being paid for for $34 a month. So if in fact there was no advertising and you wanted to have the entire internet experience the way it was today and every website you went to was sort of a performance, where the fee was managed by some sort of performing rights organization or something like that, you’d have to pay $36 a month, and you could have the internet experience free without any ads. But in fact what we’ve leaned in some previous meetings is that there are an awful lot of people fighting over that ad revenue, which is actually causing cost-per-click to come down – it’s supply and demand. And so there may be some evidence, although I don’t have it in this talk, but I’m going to come back to it, that there isn’t as much advertising money as we really need to sustain this ecosystem. And we may go to other payment models, which could be a performance rights collective or something like that. The next question we asked is ‘how does it look from country to country?’ And so we rounded up data from Canada, the US, and various countries in Europe. Now this bears looking at for a while: notice that the US is not at the top of the list. There is something very interesting going on here. Norway spends more per household than we do. The UK is right up there with us, Sweden and Denmark are up there. Obviously Scandinavia is a hot place for online advertising. Canada is $22 a month. Then as you look at other places in Europe, it really trickles off: Germany at $18, then you get down to France at $13, Italy at $11.5, and Spain at $9, Russia… This is nothing; this is peanuts; this is noise. There is no ad revenue going into those countries at all. I found this fascinating. I’m going to come back to this in a minute. Whenever you talk about advertising you have to talk about Google. So I started to write there “The 900-pound gorilla” and then I said ‘no, about 600-pound gorilla.’ I finally conclude that I didn’t know what it was – they’re big in this business but it’s important to understand how big they are. This is all from their annual report. They make $31 billion a year (this is worldwide) from Google sites. They make $12 billion a year from ad partners, but they give back $10 billion to the partner share, which means that on these ads (AdSense, ads placed on others things) they only make about $1.5 billion. So this is not a big business. And the other thing that is in their annual report is that 46% of their revenues come from the US. I think 11% came from the UK and the rest was called global. I didn’t split out the UK but you could do it. Now I’m guess that this 46% sort of equally applies to both these numbers, if you make a few calculations. (By the time I’m done I’m going to multiply so many highly variable numbers that number’s not going to mean anything.) What you come up with is that they net out about $15 billion in the US or 41% of the total ad-spend. There’s a member of the faculty in the Sloan School, a woman named Catherine Tucker, who basically said that the conclusion from this is that only company who really knows how to make advertising work is Google. Banner ads and other kinds of stuff just really don’t work very well – that’s why people won’t pay any money for them; they are worth nothing. And so people who just plop a random ad on their page are living off breadcrumbs. So that’s interesting. Audience comment: Google introduced the banner ad. Dave Clark: We’ll see how that goes… One of the questions that Israel asked is ‘Can we explain this variation amongst European countries?’ One hypothesis was that in countries where there is a low GDP or lower consumer spend there is less advertising. Israel normalized these two numbers using an econometric measure of consumer consumption – so presumably, ability to spend. And this is interesting too. You’ll notice again the UK is way up there. The UK is obviously really into advertising relative to people-spend. Denmark and Sweden and Norway are up there. The US isn’t quite as high, and it’s not that far from Canada now. But if we look over here [to the right of the slide] there is still a lot of noise in the system. Spain down there, Germany’s up here. Italy’s down there, France is down there. So, this doesn’t really explain the variation. I had thought maybe consumer wealth was an explanation. And the answer is no. The faculty member I just mentioned, Catherine Tucker, said one possible explanation – and I’m not sure I buy this and I’ll tell you why in a minute – has to do with the degree to which the countries enforce their privacy policies. If you’re in a country that strictly enforces the European privacy directive, you have to get consumers to opt into things like behavioral profiling. I’ve come up with one number – I don’t know if its right, I’d like to find some corroboration – that if there is no behavioral profiling so that its just a random ad placement, the amount you can get for taking an ad goes down by a factor of 4. Just a random ad slotted on something is of very low value compared to an ad that’s been placed based on some behavioral or demographic information. On the other hand, the reason I say I’m not sure I believe that, is that one of the countries that is pretty rigorous about enforcing its privacy policies, Germany, is way up there. So I don’t think we’ve found the explanation here. But there are clearly people off studying this, and people in business schools like to study things like this. Let’s go back to the question, ‘where is the money in the ecosystem?’ These are US numbers. A typical monthly broadband spend by US consumers is about $42. This is per household. For wireless it’s $160 - this is total wireless revenue divided by households. There can be more than one account in a household. So it’s not as if your typical bill is $160 if more than one person lives in a household. This gets confused by family plans and sharing and everything like that. Voice averages $15 – I was surprised it was so low. Video, the cable-class product, and I have to be careful here: per broadband customer who subscribes to cable, it looks as though they spend about $83. But not every broadband customer buys cable or buys linear video. So in fact, if you ask ‘what does a typical broadband household spend on video?’, it’s a lower number, about $52, because not everyone buys it. What do we spend on e-commerce? Israel found a series of numbers. This is the lowest one I think, $210 a month. Just for visualization purposes, we plotted what the monthly number is on advertising – this is an advertising spend, not a consumer spend. The blue ones [bars on the graph] are consumer spends. Just in case anyone was thinking it was worthwhile being in the CDN business, it’s about $0.74 a month – that’s what those guys get per broadband household. It might be a profitable business, but it’s a tiny one. So, here are a few more numbers on e-commerce, because all of a sudden you look at ecommerce and you say ‘wow, maybe that’s where the money is.’ So this is some data from eMarketer.com. The black lines are 2012. This says that e-commerce was $343 billion – if you divide that out to get monthly per household spend its about $325. So that’s substantially higher than the $210. As I said, that was the most conservative of the numbers. Here’s some data from comScore that says in 2012 the total spend was $289 billion – about $260 dollars per household. What’s interesting about this is that they break out retail in blue and travel in orange. And travel is actually a very big part of what people are buying online, at least according to this data. I found that interesting. Since I was picking on Google, lets pick on Amazon. This data comes from that same source – eMarketer.com. In 2012, $32.5 billion and I think this is the interesting comment: if you include not only Amazon sales but the third-party sales that are fulfilled by Amazon, they are about 25% of US e-commerce sales. They are clearly the gorilla. This is a worldwide picture. The second guy on this list, interestingly, is Staples. You get Apple, Walmart and Dell. So the second guy on the list is a lot smaller than the first guy on the list. Now, to come back to what I said earlier, the reason we asked this question was is there money in the ecosystem, such that there could be an alternative frame in which some of the cost of access is paid by the people who are, either in a bit sense, delivering bits into the access network, or in complimentary sense, selling services. And I’m going to pick on my friends who aren’t here; I’m going to pick on my friends from Telecom Italia. I actually had this conversation with Telecom Italia. They said, 'we really have trouble bargaining with Google because we really believe that Google should pay us a substantial fee for attaching YouTube to Telecom Italia'. And so I was trying to figure out how much money is in the system. Well we sort of know because Google has advertising revenues and we know how much Google makes in Italy. We know how much advertising-spend there is in Italy; we don’t know what percentage it is Google, but if you take this wild leap and you say that it’s the same as in the US, which is about 40%, then you can do this arithmetic: the per household spend on advertising in Italy is about $11.57 per month. 41% of that is $4.75 per month. Imagine that Telecom Italia went and bargained really, really hard with Google and managed to extract 5% of the top line – they’d be making $0.23 per month. And my reaction is: what a waste of time! Why beat the hell out of your potential partners in the ecosystem to extract $0.23? I think there is an easier way to get $0.23 out of every household – just raise your cost by $0.23. Just send them a bill for $0.23 more. But if you think some countries in Europe are struggling, think about the developing world. One of the things going on at the ITU was an interest in the developing world, in trying to make back some of the revenues they lost in telephone settlements. When it used to cost a dollar a minute to place a phone call to the developing world, that money was flowing from the developed world into the developing world and it was hard cash. And its clear that some of the countries in the developing world are saying ‘the Internet screwed that source of money. Why can’t we re-impose a sender-pays model on the Internet?’ To which the answer is, in the case of the phone call from the US back to the developing world, it was because somebody’s cousin wanted to call home. They wanted to make the phone call. In the case of Google sending YouTube to the developing world, they don’t want to send YouTube to the developing world – they’re not a charity, they’re interested in advertising. They probably make a couple of pennies a month off of delivering YouTube into a country in Africa. There isn’t any money in the system to extract that is going to satisfy the problem that these countries have. And I think this a profound argument that says the sender-pays model is really going to not work. I wanted to say just a couple things about paid content. A lot of people have said that Netflix is dumping a third or a half of the traffic into the access network – why isn’t it paying something to have that content delivered. Well the money is coming out of the consumer’s pocket – either the consumer pays Comcast or the consumer pays Netflix and Netflix pays Comcast. The difference in those two stories is not how much money comes out of the pocket (although it is subtly different), but whether all consumers of Comcast pay equally and some of them have more traffic coming or whether the customers who have Netflix pay the money to Netflix then it flows back in, so it is only the Netflix customers paying for the Netflix content. But if you go read Netflix annual report – I know they just went crazy and the stock went up a lot – it’s a pretty low margin business. If you look at ESPN, they’re profitable and one could speculate they have somewhat higher margins. So I’ve actually had some ISPs say to me, not in the US, ‘the problem is we don’t know to figure out how to charge so we can extract some of this value.’ And all of a sudden this is not content delivery extraction this is content value extraction. That’s going into a hairy territory because you have a flow of bytes coming in and every flow of bytes has a different value – this does not work. I don’t think there is a regime in which ISPs are actually going to do value pricing in this space. I don’t think anyone has put that idea on the table. There are people who dream about it; I know that because they’ve told me they dream about it, but its not going to happen. But, the thing you have to worry about is how would you set that price for delivery of content into your network? If you set it high enough to extract a significant amount of value from somebody who might have some margins like ESPN you just simply drive Netflix out of the market. Once you start putting a price there, you are now putting a hurdle that everybody has to jump over. If they can’t generate that much value they can’t jump the hurdle and they’re excluded from the market. When the consumer pays, a very straightforward thing happens, which is that the consumer only watches those things that are of enough value to the consumer that they are willing to watch them. So the consumer is making a value assertion and the consumer is paying. There is one other thing floating in the system, which is sort of interesting when you look at the economics. There is something called zero-rating. The name comes from countries that have a value-added-tax, or a VAT, and there are certain things where the tax rate on the object is zero, that is to say there is no VAT. So they are called zero-rated objects and that was transferred over to the following situation: if you have a system with usage caps, especially a fairly low usage cap, and then there is some content coming towards you and the content doesn’t accumulate your monthly usage because the delivery has paid for it, the other side has paid for it, this is what is called zero-rated usage. There are other names for it, but that is the one I put here. If you think about this, it has two interesting implications. What is actually recreates is a sort of 800-number in the telephone industry. If you as an ISP set a price to someone delivering bytes, saying ‘pay me so much per byte’ – of course you bargain over what that amount would be, and bargain discounts, etc. – ‘and I will zero-rate the content to the consumer and you can put something on your webpage. I’ve seen this in Australia and I’ve seen this in New Zealand – big icons on content websites saying (they don’t call it zero-rated) ‘this won’t take your monthly usage, watch it for free.’ Now what you get is a situation where, if the delivery of the content is valuable to the content provider, either because the consumer already paid for it or because its full of such high-value ads that you’re willing to put it in front of the marginal consumer, then you can pay. But if you don’t want to pay, then the consumer can pay. If it’s valuable enough for them to watch it, and it ticks their monthly quota, then that’s fine. So you have a system in which there are alternative value equations. You can either think about the content side, the complimentary side, or the consumer side. I have made an argument, and I’m not going to elaborate on this, that if this were to emerge, it would raise tremendous regulatory concerns about abuse of market power. This scheme is happening in other parts of the world. It’s happening in Europe, it’s happening in Australia and New Zealand. So its interesting how different models emerge in different parts of the world and in some parts of the world this is considered extremely dangerous and other parts of the world its considered a normal business practice. But, that’s what I wanted to say about advertising revenues. This really started us off by saying ‘is there money in the system? How much is there?’ What was really motivating me was some of my European friends who were basically saying something that I thought was a little over-simplistic – which was ‘Google seems to have a lot of money, why can’t I have some?’ That’s a little snarky, so I said lets dig another level deeper. As I said, I’m giving two talks, but I’m going to stop and see if there is comment or questions or if anyone wants to throw a brick at me, and then I’ll go on and give the next talk. Audience question: Two things: why isn’t tiered service a good proxy for charge back for high bandwidth services like Netflix? And my second question is, the idea of content providers sort of giving money to the ISPs to carry their traffic – I heard you give a talk some years ago which was that’s what Akamai does. Their role is to route money between the service provider and the content provider. And they solved the many to many complexity problem. Does a model like this reintroduce it? Dave Clark: So as far as tiered pricing is concerned, tiered pricing is a way to associate higher consumer spend for higher usage. Absolutely. And in the next talk, I’ll put some numbers up about that. There is a question of whether it’s beneficial: whether it’s socially beneficial, whether it’s beneficial to the ISP. At some point we tried to do an estimate. This is a wire line estimate. We know what the distribution is of usage by heavy users, and median users, and mean users, and light users, and you can sort of estimate what is the extra amount that the median user of a wire line system today is spending because he is paying the same as the heavy user. And its about $0.50, so the point is that number might be very noisy, but its not $10 and its not $0.01. If it was $10, I think everyone would be doing tiered pricing. If it were a penny, obviously nobody would care. Fifty cents – it might not be worth collecting. Put enough complexity in, and it might raise everybody’s costs. In my next talk I have some numbers from the wireless world where the caps are much lower and the tiers are much further apart. I did say that about Akamai. What I said is that content delivery people weren’t interested in negotiating with every ISP, because every content person would have to negotiate with every ISP. If ISPs had independent CDNs, it would create a horrible problem for global content delivery. And what Akamai does, as you say, is if I’m a content producer I negotiate with one Akamai. Or maybe I negotiate with Akamai and Limelight or Level 3, and they have one negotiation with every ISP who is their customers. So they can do money forwarding, if that’s what we wanted to do. In fact, in the wireless world, that is happening today in other parts of the world – in parts of Asia, Akamai is now providing a service (and I’ll talk about this later) where for customers of Akamai, certain traffic being sent is prioritized. So this is not a quota, this is priority. And so the customer pays Akamai to implement this, and Akamai then implements this in the various markets where they have support. They’re doing this in the wireless world. Their technology partner is Ericcson. So Akamai gets a little of the money and Ericcson gets a little of the money and the ISP gets a little of the money. So yes, they are absolutely doing it. It’s something that they see as a business. Now in the US, the FCC sent a signal that this would be dubious behavior. But in other parts of the world it seems to be normal business practice. So again, you get this interesting variation in what you might call ‘perception of regulation acceptability.’ Dave Oran, Cisco: I just wanted to make a comment that there is this rumor on the street that Ericsson felt completely screwed by that arrangement and is backing out. I don’t know if that is true or not. Dave Clark, MIT: Oh, that’s interesting. Okay. Good rumor. Have to ask our friends at Akamai. Bill Lehr, MIT: The data we were looking at per household – that was per broadband household? Dave Clark: In every case we tried to do per broadband household. For example, we actually did – Israel did – take account of adoption rates in different countries in Europe. So we did actually try to find that data. Tony Tauber, Comcast: Just going to ask about the zero-rated idea, and how it might pertain to the things we’d been talking about in the mobile broadband group. You have tuning on your phone for example, that says ‘only do certain things when I’m on Wifi.’ So, is that a form of the same thing? Dave Clark, MIT: You could think of that as a form of zero-rating, yes, only do things when its on Wifi so I don’t have to pay for it. You could also imagine that perhaps your phone could actually have tools on it that says ‘only deliver content over the cellular system if its not counting against my quota’. But, yes, I think it’s a form of zero-rating. I hadn’t quite thought about it that way, but it could be.