Why ad targeting is tough and what does it take?

Microsoft just recently stated that it was taking a whopping $6.2 billion writedown owing to the unsuccessful aQuantive acquisition. This bit of news, and the closer scrutiny of Facebook business model post their IPO rumblings, suggest that, in the realm of online advertising, it’s basically all about the targeting.

AdWords versus banner ads

As a Reuters analysis goes to explain, there is so much online ad space available that simply hosting billboards all across the Web is no longer a profitable proposition. Meanwhile, Google AdWords still remains successful, generating more than $36 billion in revenue in 2011. So what is the key difference here? Well, the answer is targeting! The company’s sophisticated and enhanced ad-targeting algorithms increase the user relevance, and hence also the likelihood of them clicking on an ad. Obviously, this is one factor that makes AdWords more effective and lucrative than banner ads.

Then what is it that preventing everyone from improving their targeting? Actually, it’s not really that simple! Ad targeting is a tough artificial intelligence problem. While all of you might not agree that it’s indeed a worthy one, the process does demand immense technical knowhow. Here’s why!

Nuances of a targeting algorithm

A targeting algorithm will take invariably everything that you know about the impression, including  search keywords, demographics, location, time of day, previous user activity, the ad’s previous CTR (clickthrough rate) and so on. It uses that to pick from among millions of ads the one to display. And it needs to do this all in a matter of fraction of a second – a tricky thing, in the least!

Ad targeting is a sort of relevance problem slightly similar to online search: given a vast repository of data and whatever we believe we know about the user’s requirements, fetch the most relevant information and provide it. While the algorithms are different, and indeed Google has two separate divisions tackling each problem, both for ethical and technical reasons, the difficulty level is similar.

Essence of ad targeting

Basically, to even start tackling ad targeting, you need to avail of hardcore, experienced data scientists with expertise in Information Retrieval or other allied AI domains. Even once you’ve an algorithm, it is of little use without data. Of course, the more you gather knowledge about the audience, the more precise algorithm can get. Machine Learning ones are so-called since they adapt via an iterative process, and they gradually increase their precision.

The kind of data that you can compile largely depends on the consumer service you offer: Google knows about your intent on basis of search keywords. Facebook does so via your social activity. Thus far, when it comes to proper ad targeting, intent seems to be much more valuable than context. However, the holy grail is to harp on both, something which explains partly Google+.

You still face the problem of applying both the algorithms and the data efficiently. You cannot make your user linger around even while you figure out which ads to display. It takes large and complex distributed systems for this. For example, Google’s SmartASS system is one of the best-engineered ones of its kind. The fact is: For a precise ad targeting, a mix of sophisticated systems, cutting-edge algorithms and relevant data, organized and monitored by a team of world-class engineering team.

The user, the advertiser and the ad publisher all are incentivized by proper targeting. Building this synergy instead of merely putting up banner ads, is the way to success in online advertising.