October 8, 2005Association of National Advertisers
"Technology Is Making Marketing Accountable"
I wanted to come here because there are so many people in the audience who have worked with us for all the years that I've been at Google, if not before. And there are many people here who are considering using Google in this new advertising world.
So what I thought I would do is talk about how we think about this problem [of advertising] because it's a little different from lots of the other ways people think about things. And there are many ways of describing how we operate, all of which are slightly bizarre. One way to think about is that Google set itself out to work on problems that involved people – problems that mattered about information at scale – building products and services for the whole world. Now this audience is a worldwide audience, and the businesses that you represent touch more than just one country, [so] you care an awful lot about these problems.
I'm a computer scientist. I don't have the marketing background that you all do, and so I look at this as an opportunity for using technology to solve problems that have not been solved before. Occasionally, we can make money at it, we can solve interesting problems, but we can also solve problems for advertisers.
When I first showed up at Google, I started talking about this, and I discovered my whole view of advertisers was wrong, as an end user. Ads actually do have value, if you can figure out the right ones to show. It never occurred to me, but, in fact that mattered to me as a consumer. So, Google is now organized about finding the ad that has value to the end user, at the right time and the right place.
Now you all have seen variants of this, and it's useful to just set the premise here – that [advertising] is a very large industry. The global estimate is somewhere north of $500 billion. Current rough run rates would indicate that the online world is two, three, four, five percent of this total number, Google on its current revenue path is on the order of one percent of this total number. So there's a lot of room to grow.
Again, coming from a different industry some years ago, I was CEO of Novell. Novell had to do an advertising campaign. And nobody knew what we were doing, so we changed our strategy – all the normal problems companies get themselves into. We wanted to talk about interoperability, and so we hired a very talented advertising agency who came up with a set of ads that I'm sure you all remember involving fish. Basically they were pictures of bowls – very pretty pictures – and then the fish were intermixed. The [agency] people, who were incredibly knowledgeable about the different breeds of fish, thought this was the greatest ad in the world.
I couldn't quite tell if this ad worked, and I did a series of metrics and studies which were not very persuasive. But I can tell you the reason we did the ad was because my mother called, and she said she'd seen the ad. Now that is the level of accuracy I think all business should be running: if you're mother calls you, you're doing well. Not a bad principle, right? Obviously I'm being a little facetious here, but you get the point.
So I look at this not as a brilliant way of doing interesting messaging and content, which I think it is, but rather as a question of: how do we get the numbers right? How do we know that the cable TV numbers should be 16 billion? That the newspapers should be 50 billion, and the broadcast should be 46 billion, and the direct mail should be so large?How do we know that? How do we prove that?
So Google set out, among the many problems that we try to focus on, to help solve that problem. It's had a wonderful by-product that has turned out to be a great business for us. For example, we started out working on search, we also started working on product search. We looked at email and said email hasn't been changed in 20 years, so we came out with a new and innovative email program. And on, and on, and on.
The principle here is to try and solve the problem on a scale that has not been solved before. When you think about how big this is, in your companies, how big is the parts catalogue? Just a question.
So I was at Sun – it's easier to reason from your own experience – and one day I said: show me the product catalogue that our sales people carry to our customers (and this is before the Internet). It was this tall – hundreds of pages. No human understands what's in the parts catalogue, and no wonder nobody was buying any of that stuff.
To And now some people, including people in this room, come to Google with 20 million products that they want to advertise on.
Now how do they do that? Well, they took the parts catalogue that was in some document format that was proprietary, and they translated it into a feed which they could ship to Google's product, which automatically generates targeted ads on parts that no one has ever bought, : might purchase in the future, but haven't so far.
Now, how big is that business? Well, for many businesses, it is the whole company. And historically in our industry, the advertising industry, we have ignored that. So at a minimum, we can solve that problem. But we can solve some other problems too.
This is a picture of a village in India : a traditional, rural village. You go and visit the computer room; what's the most important thing there? They already have a 10-megabit Internet connection, and the router is provisioned for 100 megabits. This is a wonderful thing.
Have any of you, realistically, considered these young men and women from the village as a customer? We're clearly missing these people. They're going to be online; they're going to have the ability to create impressions, knowledge, and communications in English and in their own language. So one of the principles of the Internet is that today there are only one billion people using it. So if you take the premise that the Internet will eventually touch virtually all humans, as close to the six-plus-billion as we can, then as Vint Cerf likes to say, "We're not even near done."
Another example: how much information is there in the world? A study that was done last year indicated roughly five million terabytes. How much is indexable, searchable today? Current estimate: about 170 terabytes. So again we're back in that two or three percent of the indexed and searchable world.
A friend of mine had a liver problem, one of these bizarre medical things. So he looks it up on the web, gets all this information, goes to the doctor and the doctor says, " Let me give you the link to a doctor's website that will give you authentication to look at this information for the next 20 days, and then your link expires." That's the way he finally found the information the doctor was going to use to hopefully treat what may or may not be a rare disease. So when we can get all of that information, we'll have yet another phenomenon. And by the way, Google has set out to make all of the world's information available. Not everyone wants information to be all available; not everyone agrees that that's a good thing, so we're going to face those issues.
This chart is called a long-tail chart - in mathematics, it's called Zipf's Law, and it follows a curve around concentration. It's true of almost all networked economies. We looked at this and studied the way the Internet had evolved – and virtually all of the monetization was in the middle of this curve, the middle section. The very largest assets and brands did not have, in our judgment and it judgment of others, the right set of tools and services. They didn't have ways of determining which asset categories were appropriate for their significant amounts investments. They didn't have an ability to handle these very large sets of static, and eventually dynamic, products.
What about the low part? What's the smallest business unit in the world? It's a person in a third world country who makes something themselves, and carries it in a basket. We don't serve that person either, and if you go back to my previous argument about broadband in the village computer center, it's perfectly possible that person will eventually have a photo of their product to be able to sell it.
So, one of our objectives is to span from the very top to the very bottom. From your perspective, your objective should be to work with providers, hopefully including Google, that will give you tools to allow your products to be sold from the very top to the very bottom. And remember, these are different. They have different needs, different ways of buying, different ways of paying for things, and different ways of getting them into distribution.
So this accountability that I know is the theme of a lot of the conversations here, really needs be applied to both products and reach: literally how close you get to the products, how close you get to the people.
Part of Google's success has been that in addition to having a dynamic auction, literally the advertisers enter a bidding process for keywords and these little ads. When we actually rank and show the ads, the feedback, literally the clickthrough rate of the ad, is incorporated in the process.
Many, many, many people offer ads, and the ads are kept in the system until people stop clicking on them. So I asked for a set of representative ads, and I was going to show you all of this. But there is one missing, and I asked why, because I wanted to show you what we were doing . The ads team said, well, one of them dropped out of the system, and I asked, what was the ad? (By the way the ads are 95 characters, two lines.) The top line was: Atlanta Skydiving Center – and the tagline was: only one death a year .
Now to me this sounds like a great ad. So why did our computers drop it? No one clicked on it. Isn't that shocking? But even more importantly the Atlanta Skydiving Center didn't have to pay for their ad. From the Center's perspective, it was perfectly reasonable, and in my perspective, it was a humorous experiment.
When I showed up at Google four years ago, our revenue was something like $20 million a year. I sat down with our sales executives, who showed me AdWords ads, and I said, "You have got to be kidding. People actually click on this stuff?" And they do!
I learned this the hard way a year later when we made a database conversion. Of course at Google everything is run at full blast, and of course they didn't bother to build a backup system. So we converted, and the reporting of our systems stopped working for a time, although the ads were running. For days my phone was ringing: "I can't figure out what's going on in my business." So another thing I learned, that in this new accountable world, it's accountable in both directions. Maybe that's obvious to you, but it wasn't obvious with me – I learned it the hard way.
In fact, you can argue that there is an inversion going on, and the inversion is fundamental. If you go back to technology: the power is moving from us to the end user, and it's occurring naturally. It's occurring by virtue of technology, and the power of the personal computer and the power of the personal cell phone.
Twenty years ago, 30 years ago, we all thought, "Ha, we'll decide – we will make a decision about how this category or about what this group will do." That's not how the world works anymore; the world is inverted now. That person, that individual, makes the decisions.
This has a lot of implications. For example, we would hope that they are better educated, that they make better choices and so forth and so on, and all the things we talk about in society, but it's a fundamental inversion in the model. Now this accountability is throughout our product line. When people search, they obviously search, they show these little text ads and people click on them – and it's turned out to be quite the business.
We also came up with a product line called AdSense for content. Let me back up and say that authors are often much better writers than they are business people. What they really want to do is write, publish, communicate, and get people excited. Often they have no clue how to make any money out of it, and this is not a bad thing – they're good at one thing. How could we help them? We took our ad network, we took those ads and allowed those ads to show on the author's websites, and as people click on those ads—because people read their material and are excited about it – we send these authors a check.
The first one of these authors was a fellow who worked in Boston fixing roads for a living. But he was also a fly-fisherman of a particular kind, a specialist, he was one of the experts at it. He began writing this website. He made so much money, he could devote his entire life to his specialty of fly-fishing that he doesn't have to fix roads anymore. I don't know if that's a problem for Boston , but from his perspective it's a very good outcome. That story is repeated over and over again. It's a new way. I don't know the exact number, but many, many, many thousands people are now doing this. It's a huge business for us.
Another example is that we took the ads we show on our site, and we also allow partners – even competitors in the search business – to use them to help monetize their site and we also then allow them to keep the vast majority of the revenue. Why would we do that? In our case, it's a benefit to you. It means your ad shows to more people. It's also brings in more people, more competition, and brings in a more efficient network.
How is this going to play out? This is a traffic chart for search queries around a movie called Hustle and Flow that came out this year from Paramount Pictures. When you market a movie, you spend an awful lot of time and lots of money before the movie is released to generate buzz, to get people hyped. You do promotions, you get people excited, you do interviews, you do trailers and those kinds of things – and you can see the pattern goes up. Then after the movie debuts in what is very much a hits business, it's initially very popular, and then it falls as the next one comes along.
What's interesting about this chart is that it should be possible for you, as an advertiser, to follow the same pattern precisely. You should be able to make money whether you're an advertiser in the movie or an advertiser of things about the movie. Based what the movie's about, you should be able to also link to that. So you should be able to ride the curve up and ride the curve down. That's the level of targetability that you should have.
Now there's a separate question which I don't think has yet been solved. What tools you need – what do they look like – to allow you to anticipate this? I don't think anybody has figured that out yet, but it's pretty interesting.
Google is trying to take the world's largest index, the most relevant search results, and target them, and build all these new services. As part of our core strategy we have these five buckets, and we try very hard to link them together. For example, we use the same hardware infrastructure for all of this; we get cost levers as a business. We have all this dark fiber lying around which we link everything together with. With respect to software, we do the same thing: we reuse all of it. One of the secrets here is that we can use these things over and over again, so if we want to, we can take our ad network and then we can monetize something else.
Here's another example. I think folks here have seen that the growth rate in advertising is primarily in the Internet and to some degree also in cable television. There are a number of categories which have modest to no growth, particularly in newspapers, magazines. Now these are very, very important operations and organizations – things we've all grew up with and care a lot about. So we built a tool, which people are using now, where they can target their ads in print. What they do is they say: this is the kind of ad I want, and this is the kind of placement I want based on the context. Literally, I want to be on this type of page. We're doing this as a trial. This one happens to be in PC Magazine ; this particular advertiser wanted to be surrounded by this kind of content. Again, it's a much more targeted way of ad placement. I can assure you the average magazine you read today or newspaper you see here in the hotel, or in your travels, doesn't have this targeting. There's no relationship there because the tools were not set up to do this.
I keep thinking: when I get in the rental car, the rental car company knows who I am, so why can't they program the radio to run the kind of ads that will actually get me to buy something? I suppose I can have a button that says, "anonymize me, don't give me personalized ads, instead give me the generic ads for Phoenix. ". But it makes sense that the information is there, and that it can be linked with the users' permission. Some people want this, some people don't. If I'm going to listen to ads on the radio, they might as well ones that are useful to me. Most people would probably say, I prefer not to have ads. But given that advertising is a fundamental aspect of commerce, and one which has, in my view, great value, why don't we do it right? Why don't we target it in this new way?
There are all sorts of bizarre examples of traditional advertising, bizarre when you think about it: in Vanity Fair , which is a great magazine, you can spend a million dollars for an ad for a Prada bag. $100,000 for an ad, roughly a million subscribers, if we use round numbers, 10 cents an impression. I just checked: with Google you can spend roughly 18 to 20 cents to get an ad tied to your website to purchase this same bag.
If you were going to found a business today, the first salesperson you would hire is Google. The second salesperson you would hire is Yahoo. The third salesperson you would hire is a sales structure. The reason is – not because salespeople aren't nice, they're very nice, I love them. It's because the cost per revenue dollar of these online advertising systems is so much lower than any other way in which you can spend your money.
Another example: is it better to have the ads at the top of the page or the bottom of the page? Well, if you asked me, I'd say at the top of the page. . But we looked at this, and it turns out the bottom of the page is better – because the person who actually gets to the bottom of the page is much more likely to be serious about the informationI – it's s much more likely to be useful to them when they see it there. Not a result I would have predicted.
And do you think that prices only go up in the Christmas season? I would assume so. In fact, prices, at least in our network, vary based on how other people see the prices. So if you're a constant advertiser, your prices, your actual cost per advertising impression, goes up and down, not based so much on a perfect user value, but also on competitive dynamics. In fact you're better off to be advertising at a constant rate and accepting the average of that rate. Not something I would have predicted.
So to finish up and put this into some context, imagine a world where bandwidth is free and storage is infinite. I'm not suggesting that bandwidth is free, although it's getting so much cheaper so fast that it appears to be almost free. Storage is not free, but it's declining in cost so quickly that we can do amazing things now.
What would the world look like? Well the first and most obvious thing would be speed. How many seconds do things take? If you talk to people, in user testing over and over again, what they really want is instantaneous access. How long does it take for you to boot up your computer? How long does it take for your cell phone to power up? How long does it take you to get your Internet connection to work when you're traveling in a hotel? This is my profession; I have a degree in this. But sadly, the answer is an average of an hour. I usually end up talking to this poor technical person on the night shift, debugging the hotel router. Whatever it takes, right?
Why isn't it instantaneous? Why don't I have a ubiquitous connection wherever I go that just works? Why doesn't it work all of the time? Why doesn't it roam? The answer is: because we haven't built it yet – but you could, in my model.
We would also store everything; we would store all of your information. How many of you have more than one PC? I would guess everybody. You have one at home, or a Mac, you have one at home, one at the office, maybe you have a portable, maybe you have two. (I counted one day, I had like 15, but – OK, I'm not normal.) How many cell phones do you all have? One, plus you have a back up, and if you're a CDMA network you have the GSM one to work in the non-CDMA and so forth. How much information do you have trapped in these client devices? Why don't you put it all on your server? What server? The server that you'd have if networks were free and storage were infinite – it would all be there. The transparent personalization would come with it – the machines would immediately know you. I want to emphasize this is with your permission, of course. We would know enough about you to give you targeted information, the targeted news, the targeted advertising, to make instantaneous, and seamless, happen.
Some other examples: when we say all of the world's products and services, there are so many products and services that nobody knows about. The internet is an opportunity to reach and get that information out.
Going back to my example of trying to work in the hotel; you just want it to work. Only people like me enjoy debugging their computers. The first rule of being a computer scientist is that nobody is like you, they are normal, they just expect it to turn on, like an electric motor. If we go back to the history of the car industry: the car in 1920's was actually a serious production to turn on. All the guys would stand there and crank for a while and they had a great time.
We're in that period with respect to computers: why doesn't it just work? And of course, it being automatic would help: "it just works."
If Google could do all of this, the world would be a much better place. The ads would be even more targeted, even more useful. You would be able to reach each and every person along that curve, that I was talking about earlier. You would get the maximum penetration of your message, your product, your service, and you would get maximum accountability. So we've set out to do that.
end of session