How do targeted social media/web ads work?

14 views
0

How do targeted social media/web ads work?

In: 1

Someone searches for an item or a word on the app and then advertisers show ads related to the things people are searching / looking at

From the ad firm’s perspective: harvest a bunch of personal and demographic data from users (based on search terms, emails, messages, comments, sites visited, behaviour etc) and feed it into mathematical algorithms (called “machine learning”) to infer specific traits about that person, and predict their future behaviour and how best to manipulate that.

From the advertising company’s perspective: ask the ad firm to show ads for your company to people with certain characteristics. How they do that is basically a black box. Pay them based on how many people view or interact with the ad.

From the user’s persective: use an app or website and consent to the service provider monitoring your usage patterns and collecting personal details. See ads in the app/web page based on your interests, your demographics, your personality, the content you’re viewing, or just in general whatever is most likely to convince you into giving them your money.

If you need more specific information you will have to clarify, per Rule 2 “whole topic overviews” are not permitted on ELI5.

Imagine you want to sell targeted advertising.

Maybe you want to advertise your… *wedding venue*. Who are you going to advertise to? Well. Maybe you know that most of your customers so far, have been brides. Women, aged 20-35,
relatively affluent. Let’s start with that.

The “woman” part is easy, chances are, most of your users input an accurate gender and year of birth when they signed up on Facebook. Great. So instead of showing your wedding venue ad to EVERYONE on the site, just show it to the women aged 20-35. Affluence is trickier, but maybe you can narrow it down by area, at least. People in your city, specifically in the few areas of your city that most of your customers are from.

At this point, you have a decent campaign running. You’re showing the ads to the right people, more or less, and are probably getting a lot more bang for your buck than if you were to show the ad to random uninterested strangers. But it can be better.

You start narrowing it down to be really personal. What do people who plan weddings do? Join wedding planning groups on Facebook. So, let’s make sure to prioritise people who are women, ages 20-35, from your specified area, and are part of groups tagged “wedding”. Well, what if they’re not part of those groups? No matter. You can include people who just read about weddings online (and Facebook does have some insight into your browsing patterns). You can compile a list of wedding-related keywords, then ask Facebook to sift through the users’ private and public conversations looking for those keywords. Hell, maybe you can even plug into voice recognition on their phones and pick up on them *saying* those keywords.

At this point, you likely have a VERY nice campaign running. But you might also start to realise its shortcomings. You’re completely missing all men who are organising the weddings instead of their fiancees (because you’re specifically targeting women). You’re also missing all people who are acting in ways you hadn’t predicted. And what if you’re wrong about who might be interested? What if, hypothetically, your venue could be *just perfect* for gay men getting married? What if *that* demographic could potentially bring you more business than the women you think you should be targeting? You’d never reach the men with your current campaign because you’re not targeting them.

So you can also… Let go a little. Let AI / machine learning take over. Let an algorithm show the ad to a random-ish set of people, then just tell the algorithm what the results were. Facebook Users 1, 2, 3, 4, and 5 never clicked your ad. Facebook user 6 (a 20yo woman about to get married) and Facebook user 7 (a 37yo man about to get married) DID click your ad. So you just tell the system to show the ad to users who are more like Facebook users 6 and 7, and let the machine figure out what that means. What characteristics might predict if someone’s interested in your product.

Even targeted advertising isn’t perfect. Your ad may still be bad (maybe you’re showing the ad to the right people, but the ad has zero appeal). Your system may be unable to deal with unexpected, unusual cases (a 15yo kid doing research for his 60yo grandma’s wedding to her second husband). You may waste money on people who are interested in the TOPIC of weddings but not in your product. You may face backlash if your ad targeting took a slightly unexpected direction (and your wedding ads showing up in unexpected or inappropriate contexts). But overall, from the perspective of a business, it’s usually better than screaming into the void.

internet users are frequently targeted with ads based on their web-browsing history. Advertisers use information about the websites that users have visited to show them relevant ads. This is done by placing a cookie, which is a small piece of data, on the user’s computer. Cookies can track what sites a user visits and how long they spend on each site. They can also be used to store information such as login details or preferences. Based on this data, advertisers can target specific groups of users with particular ads