Four flavors of ad targeting July 7, 2008
Posted by jeremyliew in ad networks, advertising, targeting.trackback
I recently had lunch with Iggy Fanlo, CEO of Adbrite. He is one of the most thoughtful people I know in the online ad business and I always enjoy our conversations. He related to me how he sees ad targeting falling into four flavors:
-
Geographic
Demographic
Contextual
Behavioral
He noted that the first two of these, geographic and demographic, are black and white, and focused on the user. You are in one and only one location. You have one and only one gender, age or income. In each of these cases, the key is to gather a broad dataset with which to target. As a result, the largest sites and networks, with the largest datasets, will tend to be best at these flavors of targeting.
Contextual targeting is also black and white, but it is not user centric. Rather it is focused on the page. You are looking at a page that is about some topic. Search is the easiest case, where the user tells you what the page is about. Vertical ad networks with endemic advertisers are also pretty easy to contextually target because they only include sites within their desired topic. But the general case is much harder. Now the winner isn’t necessarily the one with the largest dataset of users, but rather the one with the best algorithm for figuring out what the page is about.
Behavioral targeting is not black and white, but rather shades of gray. Furthermore, it is both user centric AND page centric because behavioral targeting is the accumulated sum of historical contextual targeting. It is based not on what page you’re looking at now, but rather on what pages you’ve looked at in the past.
In this case there are advantages to both having more information about users past behavior, AND better algorithms. In it’s simplest form, retargeting, a web user who had visited Ford.com in the past will be shown Ford banner ads while on other sites. But ad fatigue limits the frequency with which one can retarget based on a single datapoint. Good behavioral targeting systems need good historical data as well as good algorithms to best manage the portfolio of advertising opportunities to a single user. Companies are using many different sources of historical data, including search history, looking for a user on the ad network, watching a user at the ISP level and even watching offline behavior.
AdBrite recently launched an Open Targeting Exchange where it will let any company with a targeting algorithm bid to be used to target ads across their network. It is a very interesting idea, and I’ll certainly be watching closely to see how it works for them.
Lightspeed Venture Partners Blog 
I think one interesting piece that falls out of the targeting is how the inventory is sold and if the targeting choice can increase CPMs. The retargeted ad is great in low value categories where there is little contextual data or if the yields are very low for the contextually relevant ads. Like on a Madonna artist page where contextual ads of Madonna CDs are going to produce very low CPMs.
If you include broad vertical categories with contextual, then I think this has been the big winner thus far. It’s simply the most straight forward selling proposition for the person carrying the bag and the buyer. But, from most things I’ve seen, I think it’s the web application that determines the best type of targeting to apply to a site. I suspect that this will play out in AdBrite’s exchange. It would be very interesting to see data on what type of targeting performs best on the various sites in the exchange.
There is some interesting research coming out of MIT (professor Glen Urban) that I think will eventually be relevant to behavioral ad targeting. The research recognizes that different people, in different contexts have different cognitive styles.
Cognitive styles can be measured across several different dimensions. Some examples include: impulsive (makes decisions quickly) vs. deliberative (explores options in depth before making a decision), visual (prefers images) vs. verbal (prefers text and numbers), or analytic (wants all details) vs. holistic (just the bottom line).
As we are better able to determine what sort of customer is viewing an advertisement, we will not only be able to show them an ad of something they may be interested in (geographic, demographic, contextual), but we can present the advertisement in a way they are most receptive to it. For example, if it is determined that a customer has a verbal cognitive style, they will be more interested in an advertisement listing impressive specifications of a product than they would be to a picture of the product. This will ultimately lead to higher click through rates.
Additionally, we’ll see these ad’s take users to the advertisers website in more intelligent ways. For example, take an impulsive buyer to a purchase page, and a deliberate buyer to a product details page. This will ultimately lead to higher conversion rates.
We should be careful not to confuse the way that targeting gets done in web marketing today with the way ad buying should be done in general. These four categories of targeting are artifacts of the current state of web ad tech, and are not the absolute ideal of the means marketers should use to address their customers. Fundamentally, marketers want to address customers who have needs for what they are selling (and the ability to buy). Demographics, geography, placement context, and “behavior” are all only poor proxies for the audience’s real needs and interests. Sure, they are better than nothing, but let’s not lose sight of what the real goal is.
Most of the conversation about online marketing tends to revolve around the latest, coolest piece of web tech that gets at an approximation of the audience’s needs a little bit better than the last one. The conversation has gone this way for so long that I think many web marketers forget that what they want to get at is the territory, not just the next less bad map. A lot of online marketers who professionally grew up “digital” never got to know this in the first place.
What marketers really want (whether they remember it or not) is to know who they’re talking to, and to know they are talking to the right people with the right message. This is the promise of “behavioral” targeting, and its many whizzy variants and supplements like engagement mapping. These efforts are all fundamentally flawed. Doing fancy collaborative filtering, cross-site tracking and attribution, dimensional reductions, or what have you, on a set of clickstreams is going to get you some maybe interesting data. It might even make your CTRs go up a wee bit. But don’t confuse these results with really understanding your customers’ behavior. This isn’t real behavior, this is just what your customers have done within the very, very narrow and artificial context of what you (and maybe other members of your ad network) have given them to click around in. These constrained choices are not real indications of who your customers are, what they need, or what their real interests are. Fundamentally, “behavioral targeting” today is still destination-centric, not human-centric.
The real revolution in online ad placement will be straight out of Marketing 101. Ads will get bought based on what the audience’s real needs are, based on deep, high-dimensional knowledge of who’s consuming and participating in what, and what their real interests are.
In utilising Google more and more, you realise that information delievered through news, general searches and normal day to day activities begins to become more accurate.
For example, my personalised news now delivers the headlines and news articles I would previously search for.
[…] Liew has a post on the four types of ad targeting. One thing targeting ignores is the impact of an ad’s creative on performance. As […]
[…] venture partners adds Four flavors of ad targeting: Geographic, Demographic, Contextual, […]
The four categories squarely apply to “traditional” web properties. When it comes to social networks and, especially, rich media networks (e.g. YouTube, Dailymotion), I am not sure that the four approaches are going to perform as well.
Video consumption especially lends itself poorly to demographic and contextual targeting, mainly for three reasons: a) many pages of a typical video hosting service are sparse (setting aside audio track anaysis, which has yet to prove a valid contextual dataset), b) video hosting services have little and shallow demographic data about their viewers and c) a large share of consumption is syndicated on external properties via embeds.
An alternative approach is to determine what the audience of a certain likely looks like. This can be done by mine data about social network user implicit (e.g. view counts, interrupted streams) and explicit (e.g. favorites, playlists, embeds) preferences and then identify consumption patterns, potentially cross-referencing the data with the available demographic data.
This approach, call it “socially-aware contextual targeting” is both portable wherever the video is published and focused on the audience rather than the content.
Another characteristic of ad targeting that is not mentioned are pyschographics of an individual – what are their hot buttons, what do they crave. You could maybe argue that their past behavior expresses what they are into, but there is not necessarily a product or service that caters to all of their wants and desires. If you can determine what these desires are by finding out directly from the individual, your targeting capabilities become that much more relevant, efficient, and personalized… which leads to dynamic value. Look here, these excerpts explain well… http://www.readtheanswer.com/index.php?RTA=web2
In online advertising, ad targeting http://adtargeting.com/ refers to the practice of advertisers attempting to reach ( target ) a specific consumer audience.