Schedule aside, there’s an idea that has been on my mind for quite some time. Actually, since I saw the What happens in the Facebook stays in the Facebook Flash presentation, and after my advertising teacher told us something like : “Advertising on the web is great, because you can track the viewers behavior, including pre and post viewing”.

AdServer conspiracy

Well, Facebook knows a lot about you, and in advertising, the more you know about your viewer, the more relevant the ads you present him can be. Now, Facebook has an internal advertising platform, Facebook Social Ads, based on this. Nothing external, such as AdSense, yet.

But lets imagine that Facebook launches an external AdServer and builds a big network of small advertisers, like Google AdSense, or a small network of large advertisers.

First, as Facebook knows a lot about you, it simply has to set a cookie in your browser when you go on your profile, and reuse the cookie when displaying a banner on newyorktimes.com. The AdServer would then know that I’m a 27 years old male in a relationship with a university degree.

The force

This could lead to very interesting possibilities:

• Segmented Ad Performance Calculation : Does you ad under-performs except with single gay females with a college degree? You can’t know now, but a Facebook AdServer could tell you.
• Segmented Ad Presentation: Only present the ad to your target segments. So you could have a Monster ad presented only to unemployed viewers and viewers who didn’t update their work profile for the past 2 years.
• Segmented Ad Selection : Present an ad more specific to the viewer’s segment. Creating various versions of the same banner is cheap, so if you can present a version more specific to a segment. So you could have an Obama ad targeted for:
  • Afro-Americans
  • Hispanic-Americans
  • The bunch

The ethics

Sure, this raises many ethical questions, as it could be misused by the wrong people (you know, the bad guys in 24 and in Michael Moore’s films).

I remember (or I think I do remember — please let me know if I’m misquoting) that, while I was working at ZeroKnowledge Systems way back, Adam Shostack’s moto was: “Someday, soon enough, all databases will be interconnected”.

That’s why ZeroKnowledge developed their Freedom platform, allowing you to browse anonymously. But nobody cared at the time: querying a 500MB database took a few seconds on cluster, so no one could realistically store so much information in a usable way. That’s why ZK’s great idea and platform never really worked, business-wise. Too much, too soon.

But in the 2008 world were Google and Yahoo crawl — and even cache — a huge portion of the web, where 25% of adults in Quebec have a Facebook profile, and where interconnecting platforms is easier than ever with all the APIs around, this could be up and running tomorrow morning. Too little, too late?

Big brother is watching us.

I came across a old video podcast of Sage Lewis yesterday talking about Google Trends, and an idea came to my mind (again).

If you don’t know about Google Trends, this nifty evil Google tool tries to detect trends in user searches. Some sort of Digg, but on another level:

• Digg shows content people found (either by pull or push) that interested them enough and that they wanted to share.
• Google Trends shows topics people were searching for (pull).

My experimental idea is: building a fully automatic opportunistic platform that will exploit these trends in order to generate traffic.

The three main components are :

1. Be able to “sense” these trends on time
2. Be able to give the users the content they are searching for
3. Be able to have the search engines reference you for those topics rapidly

Sensing is easy in itself : Google Trends provides the information. However, more data must be brought in since Trends is only updated a few times daily. If we take American Idol for example, user searches only make it into Google Trends past a certain point, after a certain time. So we are definitely not going to catch the trend at it’s beginning. The goal is to catch it before its apogee.

Content acquisition could be achieved by using Yahoo or Google’s API for searching, or by having an human editor (but that’s kind of defeats the fully automatic opportunistic platform). Or it could be achieved by preselecting web sites for specific topics (such as a lyrics site, a sports news site, etc.), and matching them against trends.

Reference is perhaps the hardest part. Since we aim for most of the content to be automatically acquired, and the experiment being new, it’ll be hard to be rapidly considered for keywords by Google.

One of the paths I’m looking at is automatic content submission to social engines. There are very good chances that trended content we’ve acquired might be “Digged” soon or later. So why not tell Digg about it right now? The subject is in the air for the majority of the population, so it might be relevant to the Digg population (nb.: Digg != the world, my sister never goes digg.com, yet she’s part of the world).

The second path I’m looking at is search engine advertising. People will type in “that song lyrics” into Google. We know that they will, it’s a trend. “So why not automatically buy those keywords through the Google Adwords API?” say I with an evil grin on my face. This could work pretty well to get on the trend (image below, dart 1), then more traffic could be brought through sharing features such as “send to a friend”, digg, del.icio.us, etc. (dart 2) before search engines finally catch up (dart 3).

So these are my thoughts about how to try to exploit Google Trends information. Everything is still to be done.