How to build an ad network (Facebook, Google)

A friend of mine and I were discussing a recent Valleywag article. In my latest response I finally "got" the article and it triggered some discussions.

Valleywag on Facebook Flyers & Ads: Facebook currently has all/most of their ads sold to Microsoft. They want more ad slots back that they can then sell via Facebook flyers (presumably they charge more to your for Facebook flyers than Microsoft pays since they bought like, oh, 7 gagiliion of them and you are buying 1 at a time). They are currently negotiating this deal and by upgrading their flyer system, they are tipping their hand that they want those slots back and if Microsoft knows that they want them, they'll want more money for them. Blah blah.

But the interesting discussion came out of: would people buy much on Facebook? The article makes a reference to current ad-powerhouse Google as though Google should be worried that Facebook will sells Flyers and in 6 months you'll launch marketing with Facebook flyers and some old crank in the back of the room (maybe me) will be like: "Don't you want to use AdWords too?" and the response from all the 23-year-old MBAs will be: "AdWhats? I53." (yes, they call bingo numbers to the old crank to keep him out of their hair and yes, I'm overstating the Google/Facebook thing).

I agree with two of the comments from the page: 1) a social network is a glorified marketing database and 2) let users vote with their wallet. Only time will tell, but that's just a weasel-way of saying "I'm a coward and I don't know and don't want to look silly by voicing an opinion."

Context vs. Pitch Ads on Facebook: to some extent, people click ads wherever they may be, whatever they may be doing, if the offer is right. You could be sitting at home, watching "Wag the Dog" in your PJs, having just eaten supper and if the right person calls and asks "What to go grab drinks?" then you'll go. Note: the right person. So the context was all wrong, but the pitch was right. So, even if the context is wrong in Facebook, if the pitch is right, it doesn't matter.

For this argument, let's call the "pitch" the product + timing + method. And the "right pitch" is the product that could be interesting at this time and the method = an ad that might work. On a computer, the pitch is devised partly by an algorithm (when to show it, which ad to show).

Another example: - really fun game, I play it regularly (stop laughing). It's "sponsored" by Logitech. They give out a coupon code to a discount for buying their stuff on it. They shill their stuff along the bottom. Do I even see the ads anymore? No. Do I own anything Logitech due to this game? No. (Do I own anything made by Logitech: only 1 device and that's because they are the only one that make that exact device: currently, there is no real substitute for it [aside: they should be hawking that louder]:

"Ahh, but they bought brand recognition with this game." Umm, no. Well, yes. Everytime it hangs (1 / 10 times) I curse their name. There are free flash games made by some guy in his basement that never hang. These guys were paid and are, presumably, running on good servers and they have it hang. Way Aside: I'd argue that comes down to lack of discipline and coding mechanisms - something I'm currently investing a lot of time to fix in my own work and personal programming.

The context is wrong: I visit the page to play a game and blow up other jelly people. The pitch is wrong: here is a static view of Logitech products - the only connection is that the "gameboard" for jellybattle is a keyboard (and a keyboard where little powerups and things stream down and each "turn" the keys change… umm not a "keyboard" then at all, it just looks like a keyboard).

Collaborative Filtering Collaborative filtering is "take a big database of stuff and let people vote on it in some way - then, by linking people together by similar votes, you can extrapolate future votes." Movies is a great example. If I like "Wag the Dog" and you like "Wag the Dog" and additionally, you also like "Glengarry Glen Ross" - does the chance of me liking "Glengarry Glen Ross" increase? Yes. Now, make that list of "liked movies" extend for many, many more movies and many, many more users.

With Facebook: you have a general profile of me and you have many other profiles of many other people (24 million now?). So take all that data and carve it up with SVD or some other algorithm.

Then, throw ads at people (collect data). When someone clicks an ad, figure out who else might be interested in that ad (because their profile is similar in some respect). As different people see that ad, you can figure out what part of their profile might be the causal factor for the interest in that ad.

Then, you do that over and over and over… and you have some idea of which ads are like by which people and why. And you can target your ads and you get better and better.

So, let's keep that movie thing going. I add "Wag the Dog" to the list of movies I like. "Martian Child" with John Cusak is coming out Nov 2 - so to promo it, the movie producers (or whoever does this) buys some Facebook flyers. Because I've listed "Wag the Dog" and because Facebook has learned (from other ads) that "people who like 'Wag the Dog' tend to like John Cusak [I'm just guessing here] and because John Cusak is in Martian Child…" they show me that ad. If the ad is just to get me to go to the movie, I might well click it to find out more info - watch the trailer, etc.. I am then more likely to "buy the movie" (go to it in the theatre). This goes double for me who never sees ads on TV and therefore if completely out of the loop on what movies are coming out (I had to visit a website to find this sample movie coming out soon).

That's a good pitch. The context may be wrong (I just signed in and am looking at the newsfeed of all my friends: no particular inspiration to watch a movie trailer).

Google AdWords is obviously matching a good context for buying with a good pitch. The Facebook pitch will be better (will? Yes, will, they can always steal away Googlers and they have more information on us). The opportunity would be the question. Right now, the opportunity is worse: you're going to Facebook to talk/chat/play/waste time.

Context If they change that into activities that are more apt to be complementary to buying online… then they have a proposition. This warrants further thought. I think there are a few activities that exist that Facebook can glean. I'll likely post on this further at some point.

One easy pick: when I add a movie, show me an ad for a movie. Clearly I am thinking about movies. If you show me the ad for "Martian Child" immediately following my addition of "Wag the Dog" I'm more likely to click on it: context is right: I'm thinking about movies I like - you're suggesting one I might like.

Tuesday, October 23, 2007, 12:00 AM

tagged: aaronhoos, algorithms, collaborativefiltering, flashgames, socialweb, webadvertising