WEBVTT

00:00:04.990 --> 00:00:07.250
In the previous video,
we introduced the concept

00:00:07.250 --> 00:00:09.700
of price-per-click and
click-through-rate.

00:00:09.700 --> 00:00:11.740
Once we know both
of these quantities,

00:00:11.740 --> 00:00:15.150
we can calculate the
average price per display.

00:00:15.150 --> 00:00:16.760
This is simply
the average amount

00:00:16.760 --> 00:00:20.300
that an advertiser pays when
a user is shown their ad.

00:00:20.300 --> 00:00:24.710
We can compute this by
multiplying the price-per-click

00:00:24.710 --> 00:00:27.250
with the click-through-rate.

00:00:27.250 --> 00:00:31.330
Let's go through an example
to see how this works.

00:00:31.330 --> 00:00:34.730
Suppose we have 10 users who
search for "best LTE network".

00:00:38.290 --> 00:00:41.220
Google decides to display
Verizon's ad to all of them.

00:00:46.310 --> 00:00:48.540
We know that the
click-through-rate for Verizon

00:00:48.540 --> 00:00:52.800
and for the "best LTE network"
query is 0.2, so only two users

00:00:52.800 --> 00:00:53.470
click on the ad.

00:00:58.950 --> 00:01:00.950
Verizon must now pay
the price-per-click

00:01:00.950 --> 00:01:02.860
for each of these users.

00:01:02.860 --> 00:01:06.770
Since there were two clicks
and each click costs $25,

00:01:06.770 --> 00:01:09.390
Verizon must pay a
total of $50 to Google.

00:01:11.920 --> 00:01:14.710
If we consider how much Verizon
paid to Google on average,

00:01:14.710 --> 00:01:18.400
per user, or equivalently how
much Verizon paid per display

00:01:18.400 --> 00:01:21.200
of the ad, we just
divide the total amount

00:01:21.200 --> 00:01:24.890
of $50 for the 10
users who saw the ad.

00:01:24.890 --> 00:01:27.380
Doing this, we see that the
average price per display

00:01:27.380 --> 00:01:27.980
was $5.

00:01:31.120 --> 00:01:33.780
We could have obtained this
amount in a simpler way.

00:01:33.780 --> 00:01:36.610
In particular, as we defined
in the previous slide,

00:01:36.610 --> 00:01:39.850
this turns out to be exactly
the same as the price-per-click

00:01:39.850 --> 00:01:41.400
multiplied by the
click-through-rate.

00:01:45.220 --> 00:01:48.460
For our data then, to obtain
the average price per display

00:01:48.460 --> 00:01:50.990
we simply need to multiply
the price-per-click table

00:01:50.990 --> 00:01:52.789
and the click-through-rate
tables together.

00:01:58.150 --> 00:01:59.710
The last piece of
data that we need

00:01:59.710 --> 00:02:01.650
before we can define
our problem is

00:02:01.650 --> 00:02:04.450
we need to know how
popular the queries are.

00:02:04.450 --> 00:02:06.790
Obviously, Google does
not control how many times

00:02:06.790 --> 00:02:09.460
a search query will be
searched because the users are

00:02:09.460 --> 00:02:11.240
the ones who submit the queries.

00:02:11.240 --> 00:02:13.360
However, Google does
have an estimate

00:02:13.360 --> 00:02:15.310
of the number of
times, on average,

00:02:15.310 --> 00:02:19.340
the query will be
requested over a given day.

00:02:19.340 --> 00:02:21.820
For the example that we
have been building so far,

00:02:21.820 --> 00:02:26.900
let's suppose that we expect
to see "4G LTE" 140 times,

00:02:26.900 --> 00:02:31.950
"largest LTE" 80 times, and
"best LTE network" 80 times,

00:02:31.950 --> 00:02:32.450
as well.

00:02:35.780 --> 00:02:38.590
We're now ready to start
modeling this problem.

00:02:38.590 --> 00:02:41.140
The problem that we
will consider is this.

00:02:41.140 --> 00:02:44.980
How many times should Google
display each ad for each query,

00:02:44.980 --> 00:02:47.970
so as to maximize
their total revenue?

00:02:47.970 --> 00:02:49.860
In the next video,
we will formulate

00:02:49.860 --> 00:02:53.090
this as a linear
optimization problem.