What is the Value of Demographic Data?
The group I am working in at Microsoft has to do with business intelligence and advertising displays–this is a very interesting space to be in at the moment as it is central to the monetization strategy for the MSN division (now Windows Live). It’s something that has been forced upon all Internet players.
In the pre-Google era, advertising-based services were a failed model–to use the then oft-quoted quip “there is no such thing as a free lunch”. That changed when Google proved that ad-based services can make money and they did this through state-of-the-art datamining and simple two line text-based ads which marked a complete shift in the conventional thinking. It was very innovative and it undercut premium services like Hotmail which eventually followed suit. It is the driving force behind Google’s billions (2.7 billion in revenue for their most recent quarter).
Recently I got a an internal newsletter inside which I stumbled upon a few Microsoft bloggers in this space and spent a bit of time reading through some interesting posts. What prompted me to write this post was Mark Jacobson’s point regarding user profiling and how Google lacks demographic data on its users. You have to ask yourself, why is it that Google is not interested in this information and how much of a disadvantage do they have? Do they know something we don’t?
If you sign up for Hotmail or any of the Passport services, one of the forms you encounter will require your birth date, gender and location. Google on the other hand doesn’t ask you for any of this information–their form is extremely minimalistic in that regard.
Why Demographic Data
Most conventional marketing and research is based on targeting demographic profiles. If you are a 29 year old male living in the 90210 zip code (Beverly Hills) it would indicate that you are in the 1MM+ income group. That helps in targeting ads to you. Your very first ad could be for a new Porsche or a Versace suit. This is traditional marketing and while I don’t know how optimized those heuristics are, but on the web you can–and should–leverage a lot of the newer metrics it has to offer.
As an aside, ads on television networks also have a lot of room for improvement in the ad relevance arena as we step into the digital age, however, there are technical hurdles (harder to datamine videos; harder to show different ads to different people).
Context is Critical
So I am a 19 year old whose zip code is in the university district and you can show me ads about a Dave Matthews concert in the area but how likely am I to click it based on my age and zip code? The probability that I am a student is pretty high because I’m the right age and in the right zip code but that also assumes you have prior research to deduce user profiles for that demographic data. And now you still need to know if I like rock, heavy metal or house music. You also don’t know if at that very moment I am very busy researching symptoms of an illness or researching my next big investment move in a particular stock and don’t give a hoot about some concert.
On the web you are what you read. It’s about contextualization. What should matter to advertisers is what I am looking at, at any particular moment because that is what is most relevant to my interests at that very moment. With demographic data you are targeting most 21 year old students for most of the concerts most of the time (up until the concert is sold out or the marketing campaign ends). All the while you are hoping to cast the biggest net possible to grab a tiny audience of concert-goers in that region. There is a lot of wasted effort. Most of the people don’t care about the concert. The clickthrough rates will be very low. In fact, this model was so bad that nearly all services I can think of that were built around ads failed with this business model around the dot com glory days.
If you shift your focus to contextualization for achieving higher clickthrough, you get a lot more out. I am on a page looking at the discography of U2 and you show me an advertisement to purchase the U2 iPod Nano or a U2 CD. Those ads are agnostic to demographic data. This means that you cannot show me an ad for a U2 concert in my state because you don’t know my zip code. However, as an advertising company you can only so show many ads on any page and the U2 concert ad is better left for the Seattle radio station website where locals visit. Local ads on localized content, adult ads on adults content, children’s ads on children’s content, rich ads on luxury content and so on.
When you have good contextualization, the value of demographic data falls considerably. That said, demographic data obviously has its uses. A page like msn.com which is generic does not represent the mood of the reader since it’s dynamic and very general. On a page like that, knowing if your user is male or female helps promote the right kind of ads.