et tu, Microsoft?

Microsoft recently announced the release of two new features for IE8 that could potentially have a huge impact on the functionality of ad networks at the end-user level. The first feature, called InPrivate Browsing, automates a process that can easily be done manually; remove cookies, delete history, removed cached files from any websites you’ve visited. The target audience for this feature is obvious, and it’s impact on the ad networks is nominal.

The second feature, however, should be of concern. This is called InPrivate Blocking, and is essentially single source domain scripting, the best known example being Google Analytics. Rather than focusing on single domain interactions (which is what cookies do), domain scripting embeds a script (e.g. javascript) from a single source that runs across multiple domains that form part of an ad network, and that’s where the problem starts.

InPrivate Blocking specifically blocks these scripts, so network advertisers are unable to track what sites you’ve visited, what products you’ve looked at, and in general your overall interests. This is a huge win for privacy advocates, and a huge loss for the ad networks. Going forward, this means users of IE8 with InPrivate Blocking will no longer be served targeted ads; relevancy and context in advertising will no longer be possible. Users will still get ads, in fact probably more than before, and what they get will be much more spammy, because the ad networks no longer know what you’re interested in, and will be forced to deliver a one size fits all model. From a business side this will also have a significant impact on ad network revenue, click through rates are likely to drop as ads are forced to become less relevant.

The whole Behavioral Targeting space needs to do some serious navel gazing, they are clearly losing the battle for the hearts and minds of consumers; first NebuAd gets a public ass-kicking and sullies an entire industry in the process, now Microsoft tightens the screws even further. More on this topic soon.

The data homunculus

One of the constant challenges for anyone developing an analytic application is how to simplify something that is inherently complex from the start. This requirement rides above the underlying complexities inherent in any enterprise grade solution, regardless of architecture, process flow, or data sources. Once you’ve figured out how to build and integrate an application that provides depth and breadth of intelligence for your business application, how do you make it understandable to a non-technical audience? One of the technologies I’ve been looking into involves creating models using hypercubes, which lets you map multi-dimensional data and keep it updated in real time. Rather than looking at two or three dimensions, a really effective model needs to include far more dimensions than most people are comfortable with from a visual perspective. The challenge is then capturing the data, making sure it’s clean, then integrating across an n-dimensional framework and making it compelling enough for a business person to want to use it. I’m trying to apply this framework to a consumer model across multiple dimensions, essentially creating the data equivalent of a homunculus (a term used to describe the distorted human figure drawn to reflect the relative space our body parts occupy on the somatosensory cortex, and the motor cortex). It would represent a scalable, temporally-driven framework for describing why different consumers interact the way they do in an on-line environment, which would allow people targeting the consumer to build out a site/offer/etc. that would map precisely to that consumers needs. I’ll comment more on this later as I get the idea more fleshed out.

Behavioral Targeting gets a needed shot

So the uproar over NebuAd continues, and appears to be expanding to include not just NebuAd but the whole Behavioral Targeting space. There have been a variety of posts wailing about whether NebuAd has ruined the market for Behavioral Targeting applications by triggering such an intense level of congressional scrutiny that regulatory oversight that could develop as a result. My guess is no, for a couple of reasons. One, implementing any sort of regulatory framework while we’re in the throes of an election year seems incredibly unlikely. Two, there are some very big fish in this pond (Google, Yahoo, others) who have a broad presence backed by deep pockets and can push back with a lot of force. I’m not saying these companies are in the same boat as NebuAd, but politicians, who are renowned for their lack of familiarity with the business of technology, are likely to use as broad a brush as possible in the name of protecting their constituents from unwanted advertising. Three, rather than ruining the Behavioral Targeting space, what NebuAd has done is (at a minimum) show other companies in this space when to duck, and when to jump. Heck, these guys have done a great job of triggering traps the rest of us can now avoid.

Jumping on the grenade

The folks at NebuAd had an interesting week. It’s not often that a relatively small start-up is called to the Capitol Hill woodshed for a public spanking, but that’s pretty much how it played out. Although it appears NebuAd clearly underestimated the impact their technology would have on privacy advocates, the problem is that politicians are famous for not understanding technology (even slightly), and in general have a very thin grasp of business—and yet they find themselves in a position to dictate terms to folks in the business of technology. I think Scott McNealy captured it best when he said “You have no privacy, Get over it.” Organizations who are trying to protect our “privacy” such as the Center for Democracy and Technology have good intentions, but that genie is way out of the bottle, and no amount of chest beating is going to put it back in. The bottom line with companies like NebuAd is their technology produces results, and big companies are more than willing to spend big bucks to get that level of targeting accuracy.

Personal taxonomies?

One of the core requirements that would need to be in place in order to nail the connection between the individual and the merchant trying to reach them is communicating in terms that (as I’ve mentioned before) truly resonate. This is not just a matter of creating an end-users behavioral profile (no small feat in itself), but actually creating copy that is specific to the individual. This touches on a number of issues, one of which, component content management, has been commented on in previous posts. The other aspect that defines the content experience is not focused on how the information is managed, but on how it’s created. This is where the idea of a “personal taxonomy” comes into play. Everyone uses their native language a little differently, everyone has pet terms and expressions they like to use—I know this because I have a five year old who is beginning to sound just like me; while it’s usually cute, it has occasionally blown up in my face.

People in general are more comfortable with people like themselves, and a big part of that is how you communicate with each other. Creating a personal taxonomy that is specific to the targeted individual would be a huge step forward towards personalization, the question of course, is how do you do it? And there is an answer to that as well, which I will address during the coming weeks.

Clusters of one

How do you deliver against a core motivation? The first step would be to define the specific motivation in the context of a delivery framework. There are a lot of companies out there that provide demographic segmentation services, grouping consumers into cute-sounding clusters for purposes of selling lists to retailers who can then target them as a whole. This is an adequate solution, if your smallest acceptable level of granularity is measured in the thousands, but even with this level of analysis and filtration, most retailers are lucky to get a response rate in excess of 2% (or as I prefer, a 98% failure rate). Why does this continue to happen, and why are retailers willing to accept such a dismal response rate? One option is to limit your initiatives to the 2% that you know are going to buy, which, if you can sort them out of the larger cluster, jacks your response rate up into the 90+ percentile. You actually end up with the same new/existing customers, but you haven’t wasted time and effort trying to get the attention of people who do not and are unlikely to care. It’s actually possible to have a cluster of one (I’ve worked with start-ups who’ve achieved this level of granularity), the problem-at that time-was the lack of technology to create a message for that one person cost effectively. A few years later, while working at another start-up, we developed the ability to create content at a granular enough level that micro-messages could be crafted on the assumption that the end destination was a cluster of one. So how do you link these two concepts? That’s one of the things I am working on now, and so far the results look very interesting. More on this later.

Making a genuine connection

Predictive analytics is pretty much what it appears to be; it addresses the question of what is someone likely to do, given previous behavior. Psychometric profiling, on the other hand, while addressing the same trajectory on a broad basis, is approaching the consumer from a motivation perspective; why are they doing what they do, as opposed to what have they done before and are therefore likely to do again? When both applications are combined in a consumer’s profile, what you end up with is a model that not only tells you what they’re likely to do, but why? This aspect is huge; knowing why lets you drill in far deeper than a traditional branding approach (Just do it? Why should I just do it?) and allows the advertiser to really connect with the consumer. This enables interaction on a much deeper level, and takes the concept of resonance out of the realm of recognition/familiarity into core motivations.

What vs. Why

Buyer motivation varies hugely. Often we have no choice (e.g. gasoline, basic food, etc.), often we have some latitude of choice (do I buy a $500 suit or a $2000 suit? And what is the opportunity cost of the more expensive option?). There are multiple, very mature technologies that are being applied to this issue, the most obvious examples being predictive analytics and collaborative filtering. Another area that is starting to develop is behavioral profiling. Behavioral engines (sometimes called psychometric profiling) are most commonly used in education and human resource applications, and will occasionally dip their toes in the retail sector as a targeting mechanism. While the delimited between predictive analytics and behavioral engines can often be gray, there are some clear distinction between them, and they can often be very complementary. How exactly? I’ll talk to that in my next post.

Getting below the surface

So if small, concise messages work well in the new mobile ad paradigm, and the content management systems have evolved to the point where micro-ads are a viable payload for delivery on a wireless network, the next question becomes how to get the right message to the right person? For any given situation a broad statement like “just do it” is likely to resonate, but resonance is not the true objective, the real purpose is to get the consumer to take action (safe to assume Nike’s subliminal interpretation of “it” is “go out and buy our shoes”. So in order for this model to be truly effective, the ad has to go below the surface, resonance is useful for branding, but advertisers are ultimately interested in follow-through on a purchase. So how do you go from surface stickiness to below the surface drilling?

Practically small

So what are the practical limits of going ‘small’? If your objective is to communicate a concept clearly and unambiguously, a good frame of reference would be any bumper sticker or billboard advertisement. When Nike says ”Just do it” everyone knows exactly what they mean. An even pithier version would be “Got Milk?” Simple, to the point copy that is catchy and sticky. The point with these examples is that consumers are already acclimated to the micro-message, in fact, it works better for them because it uses less of their time and is more likely to break through the overall ad static that is an ongoing part of their lives. This type of approach can be extended even further by creating a rich media version of an ad snippet; a short, animated gif file can convey paragraphs of meaning in a matter of seconds, and can tie in very tightly to a branding strategy. The fact that it’s rich media actually works well with most component content management systems, particularly with those that are object-oriented. This also translates very cleanly into a mobile advertising environment, where the footprint is small, the attention span is short, and distractions are rampant.