Behavioural Targeting

I believe we have a lot of research to do to find the true value of inferring things from on-line behaviours. Inferences are either too simple or require a vast amount of data and analysis (which is expensive).

It’s Easter Monday (for my US friends, that’s a public holiday in here in the UK). What did you do this Easter?  I’ve been visiting family, and eating Easter eggs, in Shropshire and Wigan. I’ve been observing Roman ruins in Wroxeter (or ‘Viroconium’ as it was known when it was the fourth largest city in Roman Britain). I watched the film, Walk The Line, at Shrewsbury’s new(ish) Old Market Hall cinema. A lovely weekend.

So, if you were in the advertising business what adverts would you like to show to me now that I have shared this information with you? You know a little about my likes if not a lot about my dislikes.  Based on what I’ve just told you, would you promote a new film (Walk The Line was released in February so I am not sure it says I am big on new releases) or a new chocolate bar (I did extend my Cadbury’s love affair) or would you promote something historical? I am not sure there’s a movie based on Roman chocolatiers, but if there is then let me know!

Now, I know you’d actually profile me, determine that Walk The Line didn’t literally mean I loved country music but the fact that I bought Easter eggs and spent time with family firmly plants me into a demographic and the result of it all will be an ad for the new Volvo, or something similar. You get the idea. We can leap from A to C via M without much justification, can’t we?

This, however, is an issue I am struggling with at the moment at work. I know, I once promised not to write about work here but I’m not dwelling on specifics so go with me.

I’ve pretty much spent the last few weeks working with customers on how to implement behavioural targeting concepts for their web sites and if those concepts are right for their advertisers based on the customer’s behaviour patterns on the site.  Behavioural targeting isn’t actually as scary as I suspect some people think it is. It just suggests, anonymously, that if you have performed an on-line search looking for a new car then you might be looking for a new car. Now, if you do this on a car web site then it’s pretty much given you’re in the market. It’s not rocket science. Nobody knows who anybody is and I’m only working with technologies that do not share data between marketers. This way my behaviour on one site is not shared anywhere else. I’ve previously worked with technologies that tried to do this on a network level but it didn’t work then as the tech was too slow. DoubleClick moved away from network profiling in 2002 but, of course, technology has moved on and those working in that space should be able to be much more efficient.

The digital advertising industry is awash with people talking about behavioural targeting. I believe that reputable organisations will always exceed privacy legislation and protect their customers better than the law demands. After all, it’s in their interest. However, behavioural data requires so much analysis that I have to question if it is right for everybody. Certainly, publishers with a large user base who come back regularly might get some insight that would be useful to advertisers but we, as consumers, are complex beasts. Our likes and dislikes change based on many factors; advertising can help inform and change my mind and really don’t you want to advertise to all people interested in buying a car? Why just those who currently think it should be a new car? But what about my friends network or family? Aren’t they just as influential? We can’t look at their behaviours too but they are important in how I make decisions. Isn’t behavioural targeting too limiting then?

I believe it works in the right places. I believe that the technology I am working with is the best of breed for publishers. It’s anonymous and doesn’t enable any kind of network sharing so it’s good for consumers too. But finding the right use cases is the hard part.  How do I prove that this technology is working. I think we’ve some investigation to do yet.

I know that I am over simplifying arguments to make my point but I believe we have a lot of research to do in this space to find the true value of inferring things from my on-line behaviours. Inferences are either too simple or require a vast amount of data and analysis (which is expensive). I’ve got to spend the next few weeks reading a lot more data on this. If you’ve got anything that would be helpful then you know where I am.

Disclaimer: These are my view and not those of my employer. You have done the whole full disclosure thing, haven’t you?