Why do I see Amazon ads on Facebook? — The Tech behind Targeted Ads

How does Google make billions without charging users a penny? The simple answer: targeted ads. Let’s break down what both of these words mean. As you’ve probably seen, apps and websites have long used ads to make money. They charge advertisers a small fee to show their ads on the app or website. But how exactly do apps and websites know how much to charge for ads? There are two main approaches:

Firstly, apps and websites can charge advertisers a small fee every time someone views an ad, a strategy called Pay-Per-Impression or PPI. Since so many people view ads, apps and websites usually charge in increments of 1,000 views; that is, the pricing for an ad campaign could be $5 for 1,000 “impressions”. Because advertisers often pay per thousand views, Pay-Per-Impression is more often called Cost-Per-Mille, or CPM. (Mille comes from the prefix milli, as in millimeter.) Alternatively, apps and websites can charge advertisers whenever someone actually clicks an ad, which is called Cost-Per-Click, or CPC. CPC is less frequently known as Pay-Per-Click or PPC.

Google and Facebook offer both Cost-Per-Mille (CPM) and Cost-per-click (CPC) advertising to earn money from ads. An advertiser who wants to place an ad on Google or Facebook products, like Google Search or Facebook’s news feed, specifies their “bid,” or how much they’d be willing to pay per click or view. Every time a visitor loads a page, all the advertisers face off in an instant “auction,” and the winner’s ad gets shown. Having a higher bid makes your ad more likely to show up, but the highest bidder doesn’t necessarily win. Google and Facebook consider a few other criteria, like how relevant the ad is, to decide which ad to show. Why? More relevant ads will probably get clicked more, so they might make more money than a less relevant ad with a higher bid. Think about it, if you were Google, would you rather show a $5 ad that gets clicked 10 times or a $2 ad that gets clicked 100 times? Ads are how Google and Facebook make money, but the reason they make so much money is because of a technique called targeting.

This “ad targeting” strategy is how Google and Facebook really set themselves apart. Because you do so much on Google’s and Facebook’s apps and websites, these companies know a lot about what you like. For example, if Google notices that you search for “guide for choosing a watch” or “cost of a cheap watch,” Google could infer that you’re shopping for a wristwatch. Then they could show you ads for watches when you’re doing future searches. Since these hyper-targeted ads are more relevant to you, you’re more likely to click on them than non-targeted ones.

Ad Targeting

Targeted Advertising

Targeted advertising in the modern sense implies a high-resolution view of the audience.

Creating this view requires several components, broadly speaking:

1. A means of uniquely identifying a person that might view our ad. This could be a device ID, an email address, or a phone number. When we connect these different IDs together, we make what’s called an identity graph. (The term “graph” here is being used in the sense of the computer science concept).
2. A data model of a person built around aspects that might make them more or less amenable to certain types of ads. Basic demographic information such as gender, age, and geographical location, is implied here as well as potentially more specific data, like previous purchases.
3. A data collection system or systems, such that this data model can be continuously updated as the person’s characteristics for the purposes of advertising change.

Indeed, if we take a look at the landscape of ad-tech and industry-adjacent companies out there, they all fall within at least one of these categories.

Data that goes into the data model may include browser cookie data (which can connect browsing history across web domains), search history (Google uses search history to power its advertising business), social media data (e.g., age, sex, geographical location), smartphone location data, and purchase history.

From the data model, we can decide if a person fits into an advertising category such as, “Males 18 to 35.” Sorting into these categories is called segmentation.

There isn’t a single data model, of course. Different advertisers have different data, and they don’t always work together. For example, Amazon has a different “view” of me when compared to Google.

Compared to traditional ads, advertisers in this world of targeted digital advertising are no longer satisfied with inferring your habits and interests from the publications you subscribe to. They are now actively involved in trying to learn things about you from third parties, called data brokers, who buy and sell personal data in order to enhance their data model.

On the other side of that exchange are a multitude of applications that surreptitiously collect data about your location and usage patterns which are then sold to data brokers to generate revenue for the application developer. The data brokers function as the intermediary in this exchange.

These can be broken down into four broad categories:

1. Web Tracking: When a user visits a website, the website loads built-in web tracking technology. The most common of these technologies is the web cookie, which can be “first-party” or “third-party.” First-party cookies are developed and placed on a website by the website owner. This enables the website to track a user’s movements and activities between web pages on the website. Third-party cookies are developed and placed on a website by a third-party entity, in partnership with the website. This enables the third party to monitor and track a user’s activity across the website, as well as across every website a user visits that have the third-party cookie code embedded in it. Cookies can enable the tracking party to accrue data that can help them infer the interests, preferences, behaviors, and routines of a user based on their behavior across a network of third-party sites. This information can then be used to target these users with specific and relevant ads. Although cookies are a highly pervasive form of web tracking, users can manage how cookies monitor and collect information on them by clearing cookies in their browser settings. Users can also deploy tools such as the “Do Not Track” feature available on some browsers, which sends a request to websites a user visits to disable its cross-site user tracking, which includes cookies. However, respecting Do Not Track is a decision made by websites, and some advertisers can actually use the fact that users have opted to use Do Not Track use as a signal in browser “fingerprinting,” discussed later in this section.
2. Location Tracking: Granular location data is an integral component of the digital advertising ecosystem, as it provides a significant amount of information about the interests, preferences, behaviors, and routines of a user. For example, a user’s location information can provide insight into where a user lives, where they work, what stores or businesses they regularly visit, and where they spend their free time. This information can be used to determine which ads a user should be targeted with. Smartphone application makers routinely use GPS signals, cellular network triangulation, Wi-Fi SSIDS, and Bluetooth connectivity to collect such location information.
3. Cross-Device Tracking: Consumers often access the internet from multiple devices. However, advertisers generally want to avoid delivering duplicate advertisements to consumers across multiple devices. In order to control when and where the ads are delivered, the digital advertising industry has developed cross-device tracking technologies that monitor user activity across devices. Once cross-device inferences are made with high enough confidence, many companies will associate a unique identifier with a user. This identifier becomes a central anchor of user data that is collected across multiple applications, internet platforms, and devices.
4. Browser Fingerprinting: A browser fingerprint is a compilation of data about the setup of a user’s browser or operating system. This information can include a user’s browser provider, browser version, operating system, preferred language, browser plugins (software that is typically third-party that adds functionality to a browser when installed), tracking settings, ad blocker settings, and time zone. Because users are able to customize their browser settings and preferences, their browser fingerprint can be used to identify them across the internet.

Data brokers such as Acxiom, Experian, and Oracle also play an important role in data collection and user targeting. These companies combine user records from a range of sources, including retail purchases and census data, in order to provide advertising platforms such as Facebook with hundreds of unique data points that they can use to enhance their profile database. Internet platforms, advertisers, and third-party data brokers can also use data modeling techniques to make further inferences and predictions about consumer traits and behaviors. Data modeling enables these parties to use the information on observed actions and self-reported preferences and interests to supplement and fill in profile information. For example, a data broker could use a user’s zip code and name to infer their ethnicity based on census or other data showing ethnic breakdowns by zip code. Data modeling can also be used to categorize users based on factors such as creditworthiness or interest in certain topics.

Shubham Chavan
IIM Calcutta (2019–21)

References:

1. Swipe to Unlock- A Primer on Technology and Business Strategy by Aditya Agashe, Neel Mehta, and Parth Detroja
2. The Role of Data In the Targeted Advertising Industry by New America Org
3. A Brief Primer on Targeted Advertising by Robert Quinlivan