Tip:
Highlight text to annotate it
X
Hal Varian: So, thank you very much.
It's a pleasure to be here.
My job, at this point in the afternoon when everybody's blood sugar is low,
we're all feeling a little drowsy,
I'm supposed to wake you all up by talking about economics.
[laughter]
So that's a little bit of a challenge,
but we'll see what I can do.
And in fact, it's really great to follow Kevin
'cause two weeks ago, Kevin sent me an email.
He said, "Hey, what's the value of search?
What's the value of the web for finding answers to questions?"
And I said, "Kevin, you're really in luck.
That's what I'm gonna talk about at this,
before this group in a couple weeks."
So, let me launch into it.
Now of course, when I say the economic value of Google,
and I'm not thinking just about Google,
it's not.
It's if I said the economic value of libraries,
I'd be thinking about the libraries and the books and the circulation,
the readers, and all that stuff.
So, I'm thinking about the value of the whole ecosystem.
If you want, think of this as a value of search in general,
or the value of search on the internet.
But there's a couple different ways to look at that.
And I'm not gonna try to say,
"What would the world look like without search engines?"
Or, "What alternative histories would be like."
What I'm gonna do is I'm gonna try and quantify the value of search,
or the value of Google, in two dimensions.
One is, what you bring to advertisers and what you bring to users.
And the way to quantify that is to try to look at the value
that advertisers are getting from search engine advertising
and look at the value that users are getting
by using search engines to answer their question.
And of course, this is inherently a back of the envelope exercise.
I'm not gonna do things to two or three decimal points.
So, if it comes within a few billion, I'll be happy.
So, I'll give you a little model of a value of Google to advertisers.
I gotta warn you, there is an equation here on the next slide.
When I wrote a book once, my editor said,
"You know? Every equation you put in a book cuts its sales in half."
So, I'm kind of worried that when I get to the equation,
half the audience is gonna get up and walk out.
But let's see how it goes.
So, the general idea is the advertisers, they have a value per click.
And that pretty much has to do with the profit they make on the sales they have.
There's a number of clicks.
There's the cost of the clicks.
And they wanna maximize profit.
The value of the clicks times the number of clicks minus the cost of the clicks.
And that value can include all sorts of other things
that I'm not going to go into now.
And suppose you got an advertiser.
He's getting a certain number of clicks.
He's spending a certain amount of money.
Well, he could cut his bid.
He could cut his bid down to x-hat.
He'd spend less money.
And if he's making the right choice now,
then the profit he's getting now would be better than the profit
he would get by cutting its bid.
And as soon as you write down that statement,
that the value of the clicks minus the cost of the clicks
is at least as big as the alternative,
then you get this nice, little bound that says,
"The value of clicks has to be at least as big as the incremental cost."
And the logic is really just Econ One.
I could cut my bid and move down.
That would save me some money,
but I'd lose some clicks.
And if I don't wanna move down,
then the value of those clicks that I lose
must have a higher value than the money I would save.
So, it's just looking at that comparison.
I could do something different.
If I don't wanna do that,
then I've gotta have some inequality on the value of clicks.
But how do you know how many clicks you'd get if you bid less in the auction?
Well, you could experiment.
Or, Google's got this tool called a Bid Simulator.
And what the Bid Simulator does,
it basically reruns the auction from last week
and it says what you would have gotten last week
if you did something differently,
just by going over and simulating
what the outcome of the auction would be.
So, if you decrease your bid, you move down in the rankings.
We can estimate how many clicks you'd get from being in that lower position.
And then, we could see how much you'd have to pay based on rules of the auction.
And that gives you a pretty good estimate of what the relationship
is between clicks and cost.
So, once you get that lower bound in clicks,
then you can plug back into that value formula
the value of clicks times the number of clicks minus the cost of the clicks.
And you can calculate the value/cost ratio.
So, that's like your ROI, your return on investment.
And if you go through all that stuff, we have a few computers at Google,
so we're able to do this calculation, and
it turns out the value over cost ends up being two.
Somewhere around two.
And if you think about that,
it says the ROI on that advertising is a 100%.
Well, that's crazy.
I mean, how could you have an ROI that is a 100%?
That's huge.
But the answer is, go back and look at those auctions
because if an auction is oversold,
that's when you have a whole page full of ads,
which on Google is 11 ads,
then the competition is really intense to get on that page.
And you have to bid close to your value in order to get your ad shown at all.
But if you have an auction that's undersold,
that has two or three ads on it,
well then what happens is the competition is much diminished
and you end up getting a real bargain
because in an extreme case
where there's only one ad on the page,
you don't have to compete for being in that page at all.
You just have to pay the reserve price
and the reserve price ends up being quite a small amount
compared to the value.
So, it turns out that in fact, this ROI is pretty reasonable
because only about 1/3 of the pages have ads
and only, when you go to those pages,
the average number of ads is around four.
So, for most of the pages on Google, the competition isn't very intense.
If you're selling an idiosyncratic product,
the long tail sort of thing,
then you get tremendous value out of it.
And when you go advertising over at all those auctions that are very competitive,
and all those auctions that are not very competitive,
well this ROI is not implausible.
In fact, what happens is almost every advertiser
would like to have more clicks at the price they're currently paying.
So, in general, you can.
The constraint is in the number of queries,
the number of things that people are asking.
Not so much on anything else.
So now, we've got an idea of how to get to the value of clicks,
but that's not the end of the story
because you also get all those free clicks.
You get all those free clicks from the organic search results.
So, if you look at that it's roughly,
advertisers get about five times as many free clicks as they do paid clicks.
And then, you're getting back roughly seven times what you spend.
If you look at the ad clicks, which you're paying for,
plus your organic clicks.
And what else do you want to add into this calculation?
Well, the publishers are getting revenue from Google,
and that's the AdSense revenue.
And then we have a non-profit program
where non-profits get the value of search services provided to them.
So, if you add up the total value to the advertisers,
the total value to the publishers,
the total value to the non-profits in the US,
you get about $54 billion.
So, it's a pretty big number and that's just the commercial value
or the advertising plus publisher value alone.
So, if you wanna follow this up, you can Google it.
We released all these calculations.
We look at a state by state basis.
And you can look at those numbers yourself if you want to pursue it.
Now, the second component I talked about
was looking at the value to the users.
And then you wanna ask,
how much is search worth to users?
Well, you might say, "How much would you pay to give it up?"
Right? If I went out and said,
"Hey, I'm gonna pay you.
You can't use a search engine for a year.
How much would you require that I pay you to give up search?"
So, there's a nice little paper from some faculty members
at University of Michigan.
It's called "A Day Without a Search Engine."
And what they did is we gave them some queries
that people type into Google.
And the Professor Chen took teams of students
and divided them up into two groups.
One group went to the library and tried to answer those questions.
And the other group went to a search engine.
In this case, Google, and tried to answer those questions.
And then she got the answers to the question.
She had a board evaluate how accurate they were
and tried to measure the time savings
from using a search engine versus conventional technology.
And notice that these students started the exercise in the library.
So, they didn't have to get in the car,
drive to the library and do anything like that.
They could just start right there and try to answer the questions.
Well, it turns out that without the search engine
it took about 22 minutes to answer the average question.
And with a search engine, it took about seven minutes.
So, it was saving people roughly 15 minutes to do it.
Now, there are little subtleties there and I wanna bring those out
because not everything that people type into Google
is really comprehensible.
You don't know what it is.
So, she divided the questions into those that were answerable.
So, there was the query, "Where in the world is swine flu?"
And she translated that into a specific question for the students,
"Is there a map where I can see where swine flu has been diagnosed?"
And the same thing for these other questions I listed up there.
And then there were some things that were just not answerable,
TV shows on the internet, TechNet, Teacher Day MySpace Comments.
Couldn't figure out what that meant,
so that fell into the unanswerable list.
And there are also lots of questions on the web
that people ask that are self-referential.
They only refer to things that show up on the web.
So, those were ruled out as well
because you couldn't really plausibly imagine
you could answer those using the library.
Now, there were about 2,500 searches
and there were about half of them that were answerable using a library.
And if you eliminate all the duplicates,
you get about 350 searches.
And they were classified into these different kinds of searches,
factual, source, web, and other.
And then they went through and tried to convert all these to questions;
questions that people could actually answer.
And when they went to the library,
they could use the reference room or the library stacks.
They could consult the reference librarian.
And then, once the questions came up,
they were rated by raters
and then they took an average of the quality of the questions
that were answered.
And so, 99% of them were answered in the web treatment,
about 90% in the library treatment.
And by the way, it is important to recognize that
these were questions that people ask on the web.
So, they were expecting to get answers.
You'd also try to look up questions
that people would ask to a reference librarian
who were expecting to get library answers
and see how many times she could answer those on the web.
And this is the number I gave you earlier.
It took about seven minutes to answer questions on the web.
Twenty-two minutes on the library.
And you got almost the same quality of answers.
And hey, guess what?
The students liked the web better than the library.
What a surprise that was.
So, now the question is, once we know how much time you save,
that 15 minutes a day,
then let's try to convert that to value terms.
So, this is gonna be a little bit of an economic challenge,
but hey, we'll see how it goes.
So, say using the library was 22 minutes.
Time using the web was seven minutes.
So, if you ask how many times users do a search a day,
well, the rough answer is one.
And remember, the answerable questions,
the ones you could conceivably answer using the library,
that would be about half a question per day.
And think about what happened before the web was around.
Well, now people are asking billions of questions a day to that search engine.
And before, to all intents and purposes,
they were asking far, far fewer questions than billions.
So, back when it was really expensive to get answers,
you didn't ask so many questions
because it was a big deal to get up and drive to the library
and try to get those questions answered.
But now that answers are cheap,
we ask all sorts of questions, all sorts of questions.
Some are silly and frivolous, but some are really important.
Same kind of questions people were asking 20 years ago.
So, this is what we call demand curve for questions.
When it took you 22 minutes to get an answer,
you ask very, very few questions.
And now it takes just seven minutes to get an answer,
you ask a lot of questions.
And if it took you only three minutes to get an answer,
you'd probably even ask more questions.
So, we think of a little demand curve
where we have price in terms of time on the vertical axis
and we have questions on the horizontal axis.
And then what we try to do is we try to calculate
what's called the "consumer surplus,"
which is the value of answering those questions.
The time saved over what the technology was like in previous times.
So, it turns out if you do that
and of course I am gonna wave my hands a little bit here,
you try to convert that time savings into money savings.
So, the average hourly wage is about 22 bucks.
You're saving about 3.75 minutes a day.
That's a $1.37 a day.
And just as a little gut test, ask yourself,
would you pay $1 a day to use Google?
Well, yeah.
It's worth $1 a day, or $1.37 a day.
But this audience is an audience that asks a lot of questions, right?
So, it's pretty clear that for this audience the value
that you would have to using a search engine would be at least a dollar a day.
You add that up over 365 days a year, that's about 500 bucks.
And now, if we look at the number of users,
well, I used the wage rate.
So you could say, let's use the employed population.
That's about 130 million people,
or you could use the entire population.
That's 300 million people.
And you get somewhere between 65 billion, 150 billion,
depending on which of those numbers you actually use.
So, roughly speaking,
if you want to figure out, "How much is Google worth?"
Well, if you think it's worth about a dollar a day for the average user,
the typical user,
then that translates into those numbers that I described a minute ago.
Now there have been various other attempts to try to pin down a value for the internet.
Bob Litan and I did a study of a productivity impact of the internet.
McKinsey has done some work on this.
And the McKinsey numbers come up pretty close to what I just described,
namely around a dollar a day.
And they got that from basically survey evidence
rather than doing the time savings approach that I used here.
So in fact, this is a little picture from the McKinsey report
that shows you the demand curve in terms of the stated willingness to pay,
how much you would pay to have these services available.
So I think I've got one more slide here.
And if you look at the little summary,
the value to the advertisers plus publishers,
the commercial value, around $50 billion.
If you look at the value to users in terms of time saved,
around $65 billion.
And taken from the McKinsey study, I didn't talk about it here,
the value of all the ad supported applications is around $25 billion.
So, those are getting to be pretty big numbers.
Not like the national debt or anything like that,
but still, they're significant from an economic point of view.
Now, I've left out the cost of trips to the library.
I've left out the unanswerable searches.
I mean, celebrity searches and things like that.
I've left out the value to the non-employed people.
The value that the people are being able to get from finding just that right product
because of the product variety available on the internet,
the entertainment value.
All that stuff has been left out.
So, you should think about these numbers as an underestimate of the true value.
But no matter how you cut the cake,
the value of search, the value of the internet,
the value to be able to get answers to questions immediately,
that's a pretty big deal.
So, I'll leave it there.
Thank you.
[applause]