How the Facebook ad auctions work - Facebook Tutorial

- [Narrator] Facebook's algorithms are a closely guarded secret, but we can piece together how they work in practice, based on publicly shared information and common sense. Let's peek inside the black box. Facebook is a corporation that needs to make money, and they make money by showing ads. So the amount of money they make is the number of users, multiplied by the CPM, which is how much the ads cost, multiplied by the number of ads that we're shown per user. The CPM is actually set by the advertiser. Facebook holds auctions for each individual ad impression, and then the advertisers bid on those impressions. The advertiser would set their CPM based on this following function, the conversion value which is how much a product is worth multiplied by the conversion rate, which is how many people would purchase once they click on an ad, multiplied by the click-through rate, which is how many people would click on an ad, given you've shown it to them. There's one more thing to consider, the amount of users on Facebook is a function of how many people they retain. So for example, if they only showed ads, then users would stop coming back. So they need to balance the amount of ads and the user experience, with what the users are really there for which is pictures of their friends and family. So the number of users would equal the number of people they retained from the previous period, plus any new users they attracted, which isn't that many considering Facebook is very mature in most markets, subtract that by the number of people who churned. So the people who are active in the previous period, but now have left. So to summarize, Facebook's goal is to show as many ads as possible, but without annoying users. The way they bounce these competing objectives, is by audience segmentation. So with this fictional example, we can see how this might work. If you have a handful of people browsing the platform, Facebook would predict how likely each one of those people is to buy your product. That's the likelihood to convert. So for example, Dave, as a 0.01% likelihood to convert, whereas Sally is twice as likely to convert with 0.02. Well, that's not all, actually each person has a different cost to reach. So some likes, interests and demographics are more attractive to advertisers than others. That could mean that some people be much more expensive to reach, because multiple advertisers are bidding in that auction. Say for example, Sally, which has a very high cost to reach of 40. She potentially could be the CEO of a company, so she has a lot of advertisers trying to reach her. If you saw all the people in the audience by likelihood to convert, you get a table like this, with Bob at the top, which has the highest likelihood to convert at 0.1%. Bob also has a very cheap cost to reach of 10 cents. So he's definitely someone that Facebook would target in this campaign. Daniel also has a pretty low cost to reach, and still quite a good likelihood to convert. Bill has a very high likelihood to convert, but his cost per reach is really, really high at $8. So therefore Facebook might potentially skip this person, as well as Sarah and Sally, because they're too expensive to reach, despite their high likelihood to convert. Dave and Michael would make the cut, whereas Sam, Anna and Jim potentially have too low of a likelihood to convert. Now, this is important because this dynamic means that, as you increase your spend, you're moving away from the Bobs and Daniels, to the people who are very likely to convert and very cheap to reach and more into the Sarah's and Sally's, people who are less likely to convert and more expensive to reach. That's why as you increase your Facebook ads budget, you tend to get worse performance. That's called diminishing returns. Of course that's simplified example isn't how it actually works in real life. Facebook has to optimize from multiple actions. So they calculate what's called an expected feed value. We know this to be true because this is from a presentation from a Facebook developer. As you can see, they assign different points, based on different probabilities of conversion, and the expected value of those different events. So the more engagement that your ads get, the more points they get. And that means Facebook is much more confident in showing them to the users, and knowing that the users won't be put off by your ad. If someone hides your ad because of spam, then that could lead to a negative points total, and that would hide your ad in the auction, and would mean that you would get much worse results. The type of auction that Facebook uses is a Vickrey-Clarke-Groves Auction, and you don't need to know much about the maths of this, but one special feature is that the winner actually pays the value of the second place bid. So if you bid $10 and then someone else bids $5, then if you win, you would only pay $5. What this means in practice is that there is no incentive to cheat. Your incentive is actually just to bid the true value of your willingness to pay. That's also true of the way that Google's auction works. But one thing that's different with Vickrey-Clarke-Groves versus what Google uses, is that Facebook accounts for the loss from sharing an ad. Because they have multiple competing pieces of content they can put them the timeline, they need to account for the loss of inserting an ad in that timeline. So it's really important to know that if you don't have a high enough point score, your ad might not be shown at all. What are some conclusions we can draw from knowing how Facebook's Ad Auction works? One is that historical performance matters. If you score badly on points, it might take a long time before that campaign recovers, you might need to test a lot of new creative in order to improve it. And creative testing really is key, because the more engagement your ads get, the more conversions, the more likely Facebook is to show you in more auctions, and then for cheap amounts as well. And finally, don't try and game the system. Facebook's already worked out the math so there's no incentive to cheat. Therefore, all you should do is bid the true value of your conversion event and then focus a lot more in creative testing, to improve your performance. You might not know the exact details of the algorithm, but now that we have a mental model of how Facebook ads work, you can make much better decisions when optimizing your campaigns.