Customer Interface: An Ad of One’s Own

Infoseek’s cutting-edge technology delivers ads to the most receptive eyeballs.

Wow. It’s been a while since I felt my hair (what’s left of it) stand on end. No, I didn’t have my hands on a Tesla Coil. I was sitting in a micro-office in the middle of Infoseek’s booth at Internet World this spring, listening with rapt attention to Peter Rip*, vice president and general manager of Infoseek Network. He was talking about the very near future of advertising on the Internet.

Advertising has come of age on the Web. We’ve progressed from getting flamed for launching commercial Web sites to getting bombarded with offers to buy banner space. And while there are many effective ways to get your message out besides banners, banners are starting to get smart.

Yes, you can go to DoubleClick and buy space across a whole network of Web sites to target men in the 18-to-24 age range by ZIP code who are surfing in from .edu domains and sell them condoms. You can also buy a banner on Women’s Wire to promote home pregnancy test kits. If you do, your click-through response will go from the usual 1.5 percent up to 3 or even 5 percent.

You can also buy keywords on search engines. You want to advertise a marathon on Maui? Buy the words “marathon,” “Hawaii,” “Maui” and so on. Doing so will increase click throughs significantly.

But what if the ads themselves were smart? What if a banner ad knew who might want to see it? In that case, you’d have a bit of a content/context revolution.

Infoseek Gets Ultra Smart

Rip talks with the urgency associated with a major discovery tempered with the strain of having to explain it–again. In a nutshell, Infoseek can now know what users like and what kind of ads they are most likely to click. That raises the same question that was asked of Napoleon when he described a thermos that could keep hot things hot and cold things cold: How does it know?

Infoseek’s new technology–called Ultramatch and based on the commercially available SelectCast system from Aptex–starts off with 300 Micro-Interests spread across 22 Behavior Segments. A Behavior Segment is a broad area of interest such as computers, entertainment, health, the outdoors or real estate. A Micro-Interest is a subset of those categories, such as soap operas or lacrosse. Users are categorized according to which subjects they seek on the Web. Their interests are identified by actions rather than by declaration. Infoseek watches while they surf.

Say a user goes to Infoseek and searches for “air fare.” The engine then gives her a cookie that identifies her as somebody interested in travel. She proceeds to one of Infoseek’s partner sites–to which the company delivers ad banners–and clicks on an ad for a Hawaiian vacation. Later, she searches for information on sporting events. At each step, her behavior pattern is crunched into a digital profile that Infoseek slips back into her cookie. The next time she goes to an Infoseek partner page, Infoseek reads her profile and selects a banner to show her–an ad for a marathon on Maui.

Before privacy hormones start coursing through your veins, rest assured that Infoseek is tracking the user’s profile–not the user. Infoseek knows what a user is interested in, but not who she is. It doesn’t know her name or e-mail address. It knows what sites she went to, but that information is added to her profile and then discarded. Yet because it knows her likes, it can give her a banner that she is likely to click.

As a result, the user sees more ads that speak to her interests. Users should like going to the Infoseek site because it’s sponsored by companies offering stuff that is right up their alleys.

The result for the advertiser is less waste. You buy 1 million ad impressions and you get 1 million people who are likely to be interested in your products and services.

Really Smart Ads

But Infoseek doesn’t stop there. Using automatic collaborative filtering, the technique pioneered by Firefly Network, it can actually imbue an ad with intelligence.

Firefly recommends things such as movies by comparing a user’s stated likes and dislikes with the stated likes and dislikes of a database of others. Infoseek’s innovation was to ask: Why not do the same with the unstated–but exhibited–behavior of others? Lots of others?

A banner ad is thrown into the profiling machine. Based on the ad’s content, Ultramatch makes some educated guesses about who might want to view it. The software then starts showing the ad around. Over a surprisingly short period of time, it builds a database of the profiles of people who have clicked on that banner. The result? Ultramatch knows what types of behavioral profiles represent people who are most likely to click on that ad. Not just that type of ad, ads about that subject or ads with that type of offer. That specific ad.

And once it knows, it takes action. When the next person with that profile calls up a page on the Infoseek network of sites, Ultramatch reads the cookie, matches the profile to the ad and ships it off. In milliseconds.

The software that runs Utlramatch was adapted from the code that checks someone’s credit card history when he makes an unusual purchase.

Value Ads

The bottom line is that exhibited behavior is a far more reliable gauge of future behavior than anything an individual might say on a profile form. Giving users an ad that people similar to them have clicked on dramatically improves the chance that they will click on it too.

How dramatically? In a six-month trial on its site, Infoseek measured a split run. Users exposed to the usual rotation of banners clicked the usual number of times (1.5 percent for random ads; 3 percent for keyword-selected ads). Those targeted by Ultramatch had a 25 percent higher click-through rate. When Ultramatch was coupled with the use of keywords, the results were even more impressive. In some campaigns, Infoseek experienced a click-through rate of 9 percent–better than anything on the Web this side of “Download naked pictures here!”

There are two problems with the Ultramatch approach. First, people who don’t like cookies can blow them away. But that is a small threat to marketers: Recent studies show that only 25 percent of the browsing public has ever even heard of cookies. (And once enough people do know about them, it won’t take Madison Avenue long to kick off a “Got Cookies?”campaign.)

The other concern is legislative. Some people worry that aggregated behavioral information could potentially invade the privacy of registered voters. Heaven help us as the government embarks on the long process of legislating technology that changes on a weekly basis. That cat, however, has already escaped its bag. Databases have been used by marketers for decades, and now those marketers will be able to talk to each other.

There is every expectation that Ultramatch technology will be applied to content as well as to ads. The ability of a search engine to learn about users’ tastes and interests has interesting implications for finding things on the Net.

*Peter Rip is now Managing Director of Knight-Ridder Ventures.
KRV is the venture capital arm of Knight-Ridder and will invest
in new media companies.