It's time to clear up a big Facebook lookalike audience misunderstanding
Many people don’t fundamentally understand how Facebook lookalike audiences work and are, unfortunately, backing themselves into a corner at a time when it’s imperative to have at least a theoretical knowledge of what’s going on behind the scenes. As the co-founder of a paid-acquisition-focused digital agency, I’ve been running ads since 2013 and have overseen millions spent on this platform for both large and small brands.
Most people who run ads, even “Facebook marketers,” don’t truly understand how ads work -- they only know that they do work. To be fair, the logic makes sense: As long as your ads are yielding results, knowing how the sausage is made may seem unimportant. However, if you’re lacking knowledge about how the Facebook ad platform functions, you’re essentially just going to be pulling levers and pushing buttons in the hopes of success.
Facebook’s ad platform is a machine learning model -- a functional machine learning model, not just an alluring buzzword. Facebook built its advertising platform to be incredibly user-friendly, so your average advertiser doesn’t need to dive into any code to run a successful campaign.
Think of Facebook’s machine learning as a big computer that you feed various data points. In theory, the more that's introduced about a person (i.e., someone who bought something from your store), the better Facebook will be able to target more people like them. It’s important to note that all of the data Facebook collects is anonymized, a key issue in response to Facebook’s ongoing data privacy battle with regulators and the media.
Facebook takes X number of data points, such as name, age, birthday, pages liked, hometown, current town and other useful demographic data about an individual and tries to find some sort of pattern with the millions (billions, trillions?) of data points in the background that the marketer hasn’t explicitly entered (e.g., maybe a correlation between people who live in a specific hometown and their higher affinity for purchasing dog-related products).
Here’s the magic of Facebook many people don’t realize: Thousands of websites are using a powerful, yet small string of code called the Facebook pixel, which tracks everything users do on your site and compiles it along with anonymized behavioral and demographic data from each individual user, provided they are logged into their Facebook account. With over 2.38 billion Facebook users as of earlier this year, chances are most people on the internet are logged on.
This pixel also tracks user-generated events, such as sites visited, products added to cart, checkouts initiated and purchases made. So much as submitting a lead form on a website with a pixel gets thrown into the mix.
Since Facebook’s pixel is on a great chunk of revenue-generating sites, we get a bunch of extremely valuable connections between data points that help alchemize demographic and personality-related information into revenue. For example, if you go and buy a donut from a donut website, Facebook will recognize that someone with your data profile has a purchase value associated with donuts (or sweets, food, etc.).
There are hundreds of billions of data points and connections Facebook works with, and machine learning makes it feasible at a scale that’s never been possible before. This helps make Facebook one of the most powerful advertising tools ever.
Facebook is able to break out all the data it gathers across its web of pixel-integrated sites, while also quantifying and qualifying any sort of correlation and purchase value. The more input, the better your output will be. Providing the model more and higher-quality data will churn out a better result (more sales).
Supercharging Facebook’s Machine Learning With Lookalike Audiences
Now, lookalike audiences are basically giving Facebook’s machine learning a huge push in the right direction. A lookalike removes many questions, ultimately saving you time and ad spend.
Facebook’s ad platform wasn’t always as sophisticated as it is now. Back in the day, making a Facebook campaign required telling Facebook things like “our ideal customer likes TechCrunch and startups, lives in San Francisco, etc.” Facebook would then do its best job to target a wide range, albeit more specific than no demographic information, of people. Companies were still able to make sales but would end up wasting a lot of ad spend to target masses of people to hopefully find their paying customers.
Today, you have a machine that takes all the explicit data points you give it, locks them in and does most of the work of finding implicit connections with behavior that would result in sales for you. Now there are thousands -- if not millions -- of data points that work in your favor to find potential customers who will convert.
The best way to use lookalike audiences is to give Facebook information about people who will actually buy -- or, better yet, have bought -- something from your store. Many successful marketing campaigns are built on email lists containing individuals who have taken key actions already, such as adding an item to their cart or purchasing an item.
The better your list, the better your advertising results. An email list of 10,000 people with a vague interest in your products is worth bupkis compared to an email list of 1,000 people who have already bought something from you.
Alternatively, if you’re starting a brand-new site from scratch without an email list, you’ll have to embed the pixel and wait for it to collect data from the people going to your site. This is where prospection, or discovery, campaigns come in handy. By targeting broader ranges of people you think will respond positively on your site, you’ll help feed the hungry pixel much-needed data that will later help you build lookalike audiences.
Generally, the colder you go with your targeting, the more expensive it will be. However, a successful prospection campaign can help you reach the goal of turning the pixel into a high-performing machine with a rapid feedback loop that gets better and better over time.