Contextual, Behavioral, and AI Targeting

Contextual (CT) and behavioral targeting (BT) represents the current buzzwords for online advertising. Similar to artificial intelligence (AI), they suffer the problems of scalability. Let's take a look at these simple concepts, how they work, and what impact they will have on advertising.

What is AI?

AI studies human thinking and tries to create programs that simulate the process. AI started at MIT and Carnegie Mellon in the early 70's. Rule engines allow practitioners to analyze the thought process, enter rules and knowledge objects, and hope that intelligent decisions emerge.


Not surprisingly, AI had limited success in narrow areas. The human thinking process is too complicated to automate with sets of rules. 30 years later, we've made evolutionary progress toward understanding the human mind.

I can't forecast revolutionary progress for AI. Millions of rules might be needed for simple everyday decisions.

What is Contextual Targeting?

CT uses programs to analyze the content of a page and deduce information. Google, Alta Vista, and other robots scan the Internet for pages, analyze them for content, and present relevant content to simplify finding answers on the Internet.

Google has developed the most advanced system for CT. Users find relevant content and view top of page ads (SEM) that are modern versions of Yellow Page advertising. Similar to AI, Google has developed enough rules such that relevant ads appear for the most requested keywords. They further optimize these ads to earn the maximum revenues based on CTR (click through rate) and CPC (cost per click) bids. Google keeps 100% of these CPC revenues.

Google also supplies text ads on affiliate sites. For most publishers, Google has not optimized the contexual relevancy to deliver higher eCPM's. Most publishers see ads on their pages and question the relevancy of those ads to their readers. Not surprisingly, millions of affiliates gain eCPM from Google of $0.25 to $0.50 per thousand page views.

Google earns less when users click on ads on affiliate pages. Thus, they have less motivation to optimize for affiliates.

Client-side CT

Competitors have stepped in to supply client-side CT ad serving. It's trivial.
Rather than using a robot to scrap the Internet for pages, Javascript supplied by the ad network analyzes the DOM (Document Object Model) of the page. The program deduces the keywords on the page and communicates with their ad server to present relevant ads.

Every CT provider have claimed success with increased eCPM for publishers. However, like AI and Google, their solution may not scale. Solutions are specific to certain types of pages; and there are infinite variations on the world wide web. Thus, the specific CT implementation represents a specific solution that works for narrow cases; and may not scale for every publisher.

Behavioral Targeting

BT uses information gathered about users in a network and infers interest to present relevant advertising across their network. For example, Facebook notes that a user plays golf from his Facebook profile. Facebook then presents golf ads when the same user appears at a sports affiliate site.

BT has some technical issues. Cookie deletion disconnects the user at an affiliate site from the user's profile at Facebook. Users need to revisit Facebook, get assigned a new cookie, and thus reconnect with the profile before learning about relevant ads at affiliate sites.

BT also faces privacy regulations. The FTC and IAB seek regulations to define the transparency that allows Internet users to protect privacy rights. Similar to DoubleClick in the early stages of Web 1.0, Web 2.0 BT companies may be constrained from innovating and monetizing at will.

Finally, BT has the problem of scalability. If I state golf, how do they know if it's true? Why only present golf ads at a sport site? Why not present at a dining site? Or any time? Do I have other interests that are immediate? How do you determine what those interests may be? What rules do we define to optimize relevant ad delivery.

Facebook reported my son playing games at an affiliate site on my profile to my circle of friends. Will I be plagued with game ads? Will my circle be followed with game ads?

Like AI and CT, BT can show success with limited cases. The scalability can be even more challenging than Google's continued efforts to optimize revenues for it's own SEM keywords.

Conclusion

Using BT or CT to optimize targeting is useful. However, like AI, the issue is scalability. Every network claims robust solutions. AI never found the killer app. If Google has not achieved ubiquity after spending billions, the progress for CT and BT may also be evolutionary.

(c) Dash Chang, 2008

Artificial intelligence - Wikipedia, the free encyclopedia

The modern definition of artificial intelligence (or AI) is "the study and design of intelligent agents" where an intelligent agent is a system that ...

iMedia Connection: Contextual vs. Behavioral Targeting

Mar 31, 2006 ... A new study conducted by Synovate for Vendare asks consumers how they like to be targeted.

Contextual and Behavioral Targeting: Worth a Try - ClickZ

Refined technology, glowing case studies -- media buyers owe it to themselves and their clients to really home in on their target audiences.

Behavioral Targeting and Contextual Advertising

Sep 1, 2004 ... Behavioral Targeting and Contextual Advertising Is online advertising undergoing a rebirth? Some might argue it is, first spurred on by ...

The Coming Ad Revolution Wall Street Journal 

This approach (called behavioral targeting and already in service by ad networks that track users through so-called tracking cookies) undercuts traditional..., by Esther Dyson