Advertising on a real-world platform: Facebook advertising
We have learned a lot in class about auctions, including those for advertisements, which may include different advertisement slots that may have different values to advertisers (typically some slots may be easier to see and pop out more, which makes them worth more than other slots that may be harder to see). In these examples, we saw advertisers bid based on a singular value: value to them per click, which could also be translated to value for a slot (when we took the expected number of clicks into account). Each advertiser would pay the value that the advertiser below them paid (for the generalized second price auction) or would pay based on harm done (based on VCG auction prices). Since there was only one factor for each advertiser, it was easy to determine who won and to find the matching market prices as a result of it (since the advertiser who is willing to pay the most wins the best slot). However, in real-world platforms, bidding isn’t as simple, due to the fact that advertisers and advertising platforms have a lot more information to work with.
As an example, Facebook has a lot of information on users, such as demographics (profile information), interests (liked pages and communities, comments on interesting posts), and responses to advertisements (completely ignores them, or possibly clicks on ads often). This allows ads to be more targeted and increases the value per impression (and possibly per click), which is better both for Facebook and for advertisers. Since Facebook makes most of its money from advertisements, this is a core component of its business model.
The first thing that differentiates Facebook ads from the ads we discussed in class is two more variables. In addition to a bid, ads have a relevance score (which is determined by taking into account all positive interactions users have with your ad) and an estimated action rate (which estimates the number of actions that would be taken on an ad), both of which may give an ad more priority. Ads can also be optimized for specific metrics, depending on use cases. For example, one advertiser may want to prioritize impressions (if trying to get more people aware of something), while another may want to prioritize clicks (if trying to sell something from a website). Ads can be optimized on eight different metrics: conversions, link clicks, impressions, daily unique reach, post engagement, brand awareness, leads, and landing page views. Each of these metrics may appeal to different advertisers, who would prioritize differently based on their use case. There are also three bidding strategies: lowest cost, lowest cost with a bid cap, and target cost. The lowest cost is designed to get an ad in as many places as possible, at the lowest possible cost, and setting a bid cap prevents Facebook from bidding any amount higher than the bid cap. Target cost is designed to maintain an average cost per conversion, which is ideal if trying to get as many conversions as possible. It also is ideal for scaling, since advertisers can limit average cost per conversion.
In modern advertising platforms, auctions are way more complex and have a lot more information to work with, as well as options for both advertising platforms and advertisers to choose from.
Information about Facebook ad bidding: https://blog.hunchads.com/facebook-ads-bidding