|
Although most online advertising is taking up more and more screen real estate, Google and a few other sites are heading in the opposite direction. They're implementing small text-based ads that work.
April 21, 2002
URL:http://www.drdobbs.com/web-development/when-small-is-better/184414520
Web advertising hasn't exactly been a growth industry of late, with customer apathy for obtrusive banners eating up click-through rates. Yet Web use is up—especially at search engines and portals. For instance, according to a study by Jupiter Media Metrix Google saw a jump in unique visitors per month during the second half of 2001. The number of unique visits increased from just under 12 million to over 22 million. So if traffic is good but click-throughs are down, then the question is how to advertise in a way that serves that market rather than alienating it.
This is a particularly thorny problem for Google, which has built its loyal customer base by delivering fast searches and relevant results unimpeded by banner ads or flashy graphics. While a strategy of building ever-more-intrusive ads to try to keep users clicking might work at some sites, Google believes it would drive users away.
Since its founding, Google's challenge has been to reconcile its search-centric philosophy with turning an actual profit. An important step in that direction was to offer premium sponsorships that appear above the search results (as shown in Figure 1). These sponsorships are in line with Google's minimalist attitude toward page design. Each ad is given two lines of text, and is the width of a regular banner ad. Sold on a straight CPM (cost per thousand impressions) basis, the ads are guaranteed a certain number of impressions for a fixed period of time.
The ads are tied to specific keywords in searches, so that an ad for, say, an automobile manufacturer will only appear when a user is searching for a relevant topic like "car parts." Salar Kamangar, Google's director of product management for advertising products, explains, "Premium sponsorships are successful because they focus attention on the advertisers whose information is most useful to users." Testing showed that if the ads were not consistently related to searches, users soon learned to ignore them.
A maximum of two premium sponsorship ads appear with each search. Google deemed any more than that to be too distracting. The revenue from these premium sponsorships, along with income from software licensing, helped Google become one of the very few dot-coms that achieved profitability in 2001.
So where could Google go from there? How could a search engine provider generate
additional ad revenue in an era of shrunken marketing budgets? And how could
it keep its highly usable pages from becoming
cluttered with flashy advertising promos?
In mid-2000, Google came up with the idea of selling space for small, highly targeted text ads that would be positioned to the right of search results. These ads would be inconspicuous and far enough away from the results so that they would not irritate those who had no interest in them. Yet, they would be within reach for those who did want to see them (see Figure 2).
The new ad format was named AdWords, and was launched in October 2000. At 25 characters for the ad title, and 35 characters each for two text lines and one link, AdWords insertions have a negligible effect on search times and have been accepted by users with little fuss. Google also displays no more than eight of these small ads on the results page of any search.
When placing a text ad, advertisers choose the keywords that will trigger the appearance of their ads. Typically, the more specific the keywords, the more successful the ad in terms of click-through rate. So a Canadian hosting company is likely to attract more interested viewers with an ad that is triggered by "hosting Canada" as opposed to an ad that appears on any search for "hosting," for example. This encourages advertisers to create ads that are highly targeted to search results, and thus, more likely to provide value to people who use them. In turn, the relevance of the ads encourages users to look at them. Click-through rates reflect this; AdWords are running at four to five times the industry standard for banner ads.
The AdWords system is easy to manage. It's completely self-service. Advertisers purchase space and specify keywords online. There's no need for a sales representative to become involved in the transaction. After an ad has been entered into the system, advertisers can check their results—such as impressions, click-throughs, and ranking—online. All they need is a credit card, which is charged between $8 and $15 per 1,000 impressions.
Although Google is a privately held company and does not divulge financial information, Kamangar says that AdWords represents a "substantial and growing" part of Google's revenue stream.
Although Kamangar won't divulge proprietary details about the implementation of AdWords, he notes that the system took a small team of programmers only "a few months" to implement. It was a modest-size project in comparison with Google's total engineering effort. Today, that effort is focused on babysitting large clusters of off-the-shelf PCs. Google's servers numbered 12,000 in early 2002, according to Laura Ramos, research director at the Giga Information Group. (When AdWords was released, Google had "only" about 6,000 machines—an indicator of the impressive rate at which Google's computing power has been growing.) Google's server farms include machines configured for three main purposes: hosting Web servers, index servers, and document servers. The group has approximately 1,000 terabytes of total storage capacity, and manages a database that is multiple terabytes in size.
Like many young companies, Google is strongly oriented toward open-source software. The company relies on MySQL for its databases and Apache for its Web servers. The OS of choice is Linux 2.2 and 2.4. Traditionally, the group has coded most of its major applications in C++, but also works with PHP and Python on some projects.
The programming itself posed few challenges for Google's tech-savvy staff. Of about 275 employees, half are in engineering, including 50 PhDs in computer science. But when some companies tried to put a wrench in the system by setting up robots that would click on their competitors' ads, quickly "burning up" all of the impressions so that real potential customers never saw them, Google leapt into action. Developers wrote software to recognize patterns of robot-generated clicks, which they were then able to discount when charging advertisers.
The only other serious problem, says Kamangar, was the "phenomenal" growth in demand for the ads. The act of displaying the ads isn't so difficult. However, Google also has to store all information from past ads, and advertisers have to be able to log in and obtain reports based on click-through data. The initial sub-second response times for reports deteriorated as the number of reports being run increased.
This was particularly distressing for the company, because Google is planning future enhancements that will place even heavier loads on database servers. For instance, one feature will permit potential advertisers to obtain traffic forecasts by country. This information will let advertisers more accurately predict the likely response to ads targeted to any given combination of keywords. However, this feature requires analysis of large amounts of traffic data from the database.
Another planned enhancement is a more sophisticated keyword suggestion tool. This will be based on database searches that include all past queries. Finally, Google plans to adjust its pricing model based on which keywords advertisers continue to buy and which ones they abandon. This market mechanism for determining the value of keywords also requires detailed analysis of past ad campaigns.
To manage the increased load, not only for the reports but also in anticipation of future developments, Google focused on optimizing the database schema to bring response times back under a second while keeping machine costs down. Nevertheless, says Kamangar, the company is continuously installing new servers to handle the increasing demands of administration and reporting.
If it took Google just a few months to build support for text ads into its
massive, highly customized system, one would expect smaller sites to be able
to implement a text ad system in just weeks. Indeed, other sites have followed
Google's lead and often report incredibly short development cycles and
very few technical challenges.
One of the first small sites to implement text ads was MetaFilter (www.metafilter.com),
a community Weblog. Matthew Haughey, designer and developer of MetaFilter, recalls
that the idea came up over lunch with fellow designer, Derek Powazek. (Derek
wrote "User to User Support," in the
November 2001 issue of Web Techniques.) The initial discussion took place
on a Wednesday in mid-October 2001. On Friday, Haughey spent $200 to license
the source code for Ad Manager, a ColdFusion-based ad management package. Since
its inception, MetaFilter has run ColdFusion. Haughey has become proficient
in the language over the years, upgrading and adding features to the site. He
estimates that buying the Ad Manager source code knocked a full week off his
development time for this project.
Haughey installed the software on the MetaFilter server, a Pentium III 550MHz
PC hosted from a friend's apartment. In addition to running ColdFusion,
the server is loaded with Windows 2000 and a Microsoft SQL 2000 database. Some
heavy revisions were required to the Ad Manager code, which was designed to
serve banner ads. Haughey tweaked Ad Manager for text-only output and integrated
it with PayPal for payment.
He launched the TextAds service on Sunday night, October 21—less than
three days after he started coding—promoting it as "a non-invasive,
non-annoying, low-cost way to get your site in front of thousands of people."
There was an initial rush of enthusiasm for the text ads, and advertisers reported
astronomical click-through rates, some over 10 percent. Metafilter gets approximately
one million hits a month, and Haughey took in $1,600 dollars in only a few days
by selling ad positions at a $2 cost per thousand (CPM). (It took him a lot
longer than that to play out all of the ads, though, because he displays only
one ad at a time.) At one point he tried raising the price to a $4 CPM, but
demand went down, and he returned to the original price.
Then the novelty wore off. Other sites implemented similar ads, and MetaFilter's
click-throughs fell into the still-very-respectable 2 percent range. Haughey
dropped the price to a $1 CPM. He now takes in $1,000 per month, rotating 40
to 50 ads at any given time. Although the revenues are extremely modest when
compared to sites like Google, MetaFilter is considered by many to be the most
successful home-brewed text ad site.
Haughey notes that the one thing that could increase the value of ads on his
site is if he were to let advertisers connect ads to keywords. He cites the
case of an advertiser that was looking to sell software for the Palm OS. The
advertiser told him that it would have paid five times as much if the ads could
have been displayed alongside every thread that mentioned PDAs.
Nevertheless, Haughey is pleased with the success of the TextAds system on
MetaFilter—a success that he attributes to an active community with strong
common interests in Web development and Web logging. Many of the 250 or so ads
placed since the October launch are from members of the community. The relatively
high click-through rates show that the ads appeal to a broad cross-section of
people who access the site. Haughey doesn't see text ads as a world-changing
phenomenon, however, because many sites do not have this type of community or
common focus, and thus would not be able to target ads effectively.
No doubt, the prize for fastest implementation of text ads goes to Web Developer
Oliver Willis, who built a system for his site in about an hour. He used CGI-based
WebAdverts shareware from Affordable Web Space Design (awsd.com/scripts/webadverts/index.shtml).
Text ads are growing so popular that there is now an open source, PHP-based
package that site owners can use to get up and running quickly. The package,
which is also called TextAds, is available through Sourceforge (sourceforge.net/projects/textads/).
The package was contributed by Bill Rini, having developed it for his own site
(www.rini.org).
"I was disappointed with traditional ads, through DoubleClick or other
third parties—especially as the ad rates fell to $2 or $1.50, and they
continued to take a big percentage of that," says Rini. "There's
nothing hard about implementing text ads yourself," he adds. "So
why let some other company be a middleman between you and the people who come
to your site, taking 25 to 50 percent from an already ridiculously low $2 CPM, or
worse?"
The TextAds package is designed to work on any Linux-based server running PHP,
Apache, and MySQL. One planned feature is an option that allows pricing based
on either impressions or click-throughs. Most systems apply charges based only
on impressions.
It wouldn't be difficult, says Rini, to trigger ads based on user input
and keyword searches. Rini explains that those using the PostNuke content management
system, as he is, could easily key ads to the topics of the articles on their
sites. But he acknowledges that at Web logging sites like Blogger, where people
are uploading content via FTP, the problem would be more challenging.
It took Rini between 30 and 40 hours over a period of four weeks to write the
code for the software. When he released the package in early February of this
year, a brief mention on the MetaFilter site resulted in 100 downloads of the
package within the first weekend.
On his own site, Rini has been seeing click-throughs anywhere from one-half
to 3.5 percent, with the average being under 2 percent. His site receives around
3,000 page views a day. He charges $10 for 2,500 impressions, which is a high
rate when compared to some other sites. However, he has been renewing the ads
for free, making them a good value. The success is subjective, though. Rini
points out that his overall profit from the ads is "absolutely insignificant."
But Rini never expected his site to be ad- supported. "I'd be
thrilled if it covered the co-location bill every month," he says.
Site owners who don't want to customize a software package or build
their own text ads implementation can use an application service provider. Companies
like TextAds.biz and Adfarm.org will host your customers' ads and serve
them to your site. Prices can be steep, though. TextAds.biz charges 10 percent
of the revenue on your ad sales. Adfarm.org charges $10 per 4,000 impressions;
you get $7.50 of this for advertisers that you bring to the service.
Rini is skeptical about the usefulness of hosted services. "I'm
sure with some of the hosted solutions, they put more expense and time into
it than I did," he says. "But realistically, you should be able
to set up and run a text ad service for a relatively low cost."
If you are considering implementing text ads, probably the best predictor of
success is your ability to target ads to your users' interests.
If your site has some sort of search engine like Google, you can bank on your
users to provide you with keywords. If your site has an active community, like
MetaFilter, your users themselves might be ideal advertisers. Or if you have
your articles in a content management system, as Rini does at his site, you
can interpret a click on a particular link as implying an interest in a specific
topic.
Growing "banner blindness" shows that most Web users, most of
the time, will not be distracted from their primary pursuits. This contrasts
with television, for instance, where people watch primarily to be distracted.
If text ads are significantly outperforming banner ads, perhaps it is because
they are a better fit for their medium. That, plus ease of implementation and
the self-service sales and production process, make text ads an attractive and
low-risk proposition for sites ranging from giant search engines to personal
Weblogs.
Michael is a freelance writer based in East Sound, Washington. Contact him
at [email protected].
Terms of Service | Privacy Statement | Copyright © 2024 UBM Tech, All rights reserved.
Ads in Three Days
Alas, given the nature of the site—unstructured, anything-goes Web logging—Haughey
doesn't see any easy way to connect ads to keywords. Analyzing massive
amounts of "dirty," unstructured data to retrieve keywords moves
beyond daunting to nearly impossible, due to his infrastructure, he says.Off the Shelf
Hitting the Target
