Getting Closer to the Perfect Ad
Two credit-card companies, specifically Master Card and Visa, are planning to store data for targeted advertising. This has yet to actually begin, so some of the details are not definite yet, but it may not be accepted by all customers. Fortunately, Master Card plans on including an opt-out option, in case people don’t want their privacy invaded too much. It is so far unknown whether or not Visa will also have an opt-out option. In either case, it will be interesting to find out exactly how effective this turns out as a business strategy, and how many users will be accepting of this.
Some data mining can usually be beneficial to the company, and sometimes it can even be beneficial to the consumer. Naturally, companies want ads to be as targeted as possible, in order to maximize the click through rate. they want to show the user ads that appeal to them. In most ads, they wish for the user to spend money on something, and what better as a predictor of what someone will want to buy than a history of past purchases. If someone eats at a restaurant every week, odds are they will be interested in ads for that restaurant. Better ads can be mutually beneficial to both the company and the consumer. When ads are appropriate, they will interest the consumer much more than random ads. An interested consumer would click these ads much more frequently, so the company hosting the ad also gets more money from this, in theory.
We have talked about ads in terms of matching markets before, but this is a bit different. In class, we focused on cases when everyone would want the same ad slot, and the whole system is that different ads get different slots. Imagine multiple ads sharing the same slot, but with different ads shown to different users. Say we have two users, x and y. x is interested in product A, and y is interested in product B, and it’s possible to tell this from their purchase history. The company could then easily show x an ad for product A, and y an ad for product B. The users are happy – they get ads that they would like. The companies behind the ads should be relatively happy – they get clicks from interested users. However, will the company hosting the ad be in an optimal situation?
Choosing what to charge the holders of the ads in this situation is nontrivial. The simplest way is to charge a fixed price per-click to all ads. In this scheme, they will make more money than the same scheme with just a single ad, as better adds => more clicks => more revenue. However, with just a single ad, instead of a fixed price, using a second-price auction is the best choice – it’s charging a certain amount per-click, but it’s based on what companies vote, so it should be a better fit than just a fixed amount. Is there any sort of auction that could be used with multiple ads? First, the company has to find a way to reward those that pay more over those that pay less, otherwise everyone would bid 0. For one ad slot, very little could be done other than showing the cheaper ad less. Unfortunately, this is somewhat contradictory with the scheme to only show the most appropriate ads. This may work in the much larger case when there are multiple appropriate ads to choose out of hundreds or thousands, and multiple ad slots. However, in the smallest example, unless they can get a very appropriate price to charge per click, this scheme could actually earn less than just ignoring the data and doing a second-price auction.
So, let’s quickly take a look at a larger example with still a single location for an ad. It could be modeled as different interests as different ad slots. Say customer x is interested in ordering pizza, and customer y is interested in going to a restaurant to eat. There could be many companies that sell pizza interested in showing an ad to x, and many restaurants interested in showing an ad to y. A second-price auction could be held for both types of customers individually, and most companies would want to spend extra because the ad is very targeted – much more so than without having any information about the consumer. Using credit-card data can become very similar to Google’s ads based on search terms, in that they are both targeted, but it could become even more powerful because it holds information on what the customer is actually wiling to spend money on.
http://yro.slashdot.org/story/11/10/25/203213/mastercard-visa-to-help-target-ads