Retargeting isn't the only custom audience approach that you can use. Let's talk about Lookalike Audiences next. So lookalike audiences are a little different, the process to get to use them as exactly the same. You have a database you upload that data base to the actual place where you're doing the advertising and you define the audience as that list of people that you uploaded. But lookalike audience is then begin to behave a little different, lookalike audiences enable the advertiser to find people that have similar characteristics and then advertise to those people automatically. At the end of the day we're essentially telling the platform that we're advertising on to figure out what it is that these people share in common. And then use those commonalities as targeting parameters to then decide whether new people, people outside of this list see an advertisement or not. So lookalike audiences are prospective, they're not introspective, instead they're actually saying I've like to find more people, people that aren't in this list that may possibly be interested. Instead you're using that base list as a seed, a seed that you hope to grow into a much broader audience. I guess one of the big things that you need to have before. You do lookalike audiences is a good customer database that you really feel as exhaustive. If you see the lookalike audience on a small number of people, it'll over fit. So the commonalities that that group of people may share may be due to chance. And if that is the case then lookalike audiences probably won't perform any better than a traditional targeting parameter approach. So once we know we have good lists of customers that are exhaustive and large enough to be representative in some way. Then we can start to use these to look and see what these people have in common. Now, of course, we could do this ourselves if we know a lot about our customers. It's easy to just you know, do some type of what we call cluster analysis in business and see what are the big clusters of people and what are their interests. But if you're like most small businesses, you probably don't have much data on your consumers. You may have their email address, their phone number, the amount of money they spent, and the products that they bought, but you don't know a lot about their interests more broadly. You don't know if they're interested in music or if they play sports, and if so what sports do they play, and how much money do they make, and so on and so forth. This is where the social media platforms have added value. They've essentially said yes, we're going to do that matching for you and we know tons about these consumers. So we're going to do all that work behind the scenes for you and we're going to pull out these commonalities automatically and we're going to build these audiences and they're going to work really great. So if we were going to do this cluster approach by scratch, this is kind of what it would look like, right? So here's a nice cluster in the top towards the middle of non-hispanic white/Caucasian people. And if we had data on our consumers and it looked just like this then we would actually be able to just simply build our own type custom audience. We would actually be able to segment our audience down and only upload that small list and just say we'd like to advertise to this list. Or we could even just by ourselves say, you know what? Most of our consumers are non Hispanic and White and we believe that inference on Facebook is pretty good. That is that Facebook has a good understanding of whether someone is Caucasian or White. So we'll just put that in as a targeting parameter. We don't even need to do look alike audiences and pay that extra cost associated with it. To put a complex story short, if you have a long list of emails and you feel like it's most of your customers, but you don't know a lot about them, and you'd like to reach more people like that, then look-like audiences was a great idea. So again to unpack what's going on here, it's simply as easy as uploading your database to Facebook. And letting Facebook figure out what are the commonalities these groups of people have. Are there a large swath of women that are aged 28 to 40? Or they're a large group of men that make over $100,000 a year? Are there a large group of women that make $250,000 a year? What is it about my audience segments that I can learn. Facebook is going to take this approach and do it behind the scenes and it's going to build an audience that should engage with your content. So let's think about it under the hood, right? So if we have a Thousand people and we want to know what TV show they're interested in. Facebook should have an idea about that, right? If we remember from the inferences that I showed from my Facebook account. It had a pretty good handle on some of the entertainment stuff that I was interested in. So here's a look under the hood at what Facebook is doing when they do a lookalike audience. So let's say I have a group of customers and it's a thousand people. I don't know anything else about them, but their email addresses and their phone numbers, I upload that database to Facebook. Facebook is going to start to look at all the interest that it has on these thousand people. It's going to look for the most common occurrences. If this was under the hood at Facebook it might actually say, wow, 875 of these 1,000 people have an interest in a particular type of light beer. Or 765 people actually are interested in this one TV show. It's going to learn that these are targeting parameters it should use when it's deciding to serve and add to somebody or not. Remember it's not going to actually be one of these thousand people but it's learning from these thousand people in making decisions on whether someone sees an ad or not. Twitter also has lookalike audiences and that's really important for me to call out, Facebook is kind of the gold standard, but of course, Twitter does this as well. One of the key advantages that Twitter has is that you don't actually need a list of consumers to get started with lookalike audience. Instead Twitter has built a lookalike approach that is based on Twitter accounts. So if you know that certain Twitter accounts mirror some of the people that you're interested in, then you're able to essentially retarget to them. And what's really great is you don't actually even have to own the account that you want to do the look-alike targeting on. So let's say I have a small coffee shop in Boulder, if there's competition in the area and I'd like to actually target to people that already like a particular type of gourmet coffee shop, I can do that. Similarly If I have a robust Twitter following myself, I can use my account as the primary influence and just say the people that follow me are the most likely to engage with me. So therefore I'll create lookalike parameters from my account only. It's important to know that Facebook allows you to upload that customer database just like Facebook and you can actually leverage it in the same way. It's matter of clicks once you have that database on your local hard drive. The Twitter approach is really simple. All you need to do is upload that custom audience list, create that custom audience in the actual platform and then check this expand your audience features button at the bottom. And Twitter will begin to serve your ad to people that are like your custom audience. So start to end, the social media platform that you're using is finding those commonalities, using those commonalities as targeting parameters, advertising to those people. And then ultimately expanding the list of people that see your ad in an intelligent way. This is intelligent, but it's also lazy, it's lazy in that we don't get to choose these parameters that Facebook uses instead it does it for us. As any type of automatic approach should this should worry you a little bit right? It's a little bit lazy and that we don't really know what commonalities it's grabbing onto, there's going to be some over fitting that occurs. And what I mean by that is there will be some commonalities that appear on paper that really are true to your target audience. As a result we need to be really critical that whatever is happening under the hood actually is driving results to a significant degree. This is Google search data, and you can see here that when people search for hamburgers they also search for beach events, and those two things could be related, right? However, you can see there are a lot of spurious relationships as well. So when Macallan searches are really popular so our parchment paper searches, it doesn't mean that those two things are related. So even though a large group of your consumers may have a particular interest or they may have a particular demographic or geographic, that doesn't necessarily mean that the two things are related. It just means that they exist together just like the search data.