6.04 – The Smarter Way To Build Audiences With Artificial Intelligence
Facebook has the ability to “model” any Custom Audience you create on their system. This is called a Lookalike Audience.
When you create a Lookalike Audience, it engages Facebook’s Artificial Intelligence (AI) to help you find people similar to your model audience. Facebook calls this a Lookalike Source Audience. Lookalike Audiences are the preferred Audience to use for prospecting on Facebook. Simply put, a Lookalike Audience is a collection of Facebook users that the AI determines is most like your model audience.
Your Model Audience should contain somewhere between 500 to 5,000 people. If you give it less than 500 users, there aren’t enough data points for the AI to see what is unique about those users. When you start to get above 5,000, the AI gets overwhelmed with too many data points and your Lookalike Audience will not be very targeted.
The smallest Lookalike Audience you can create is 1 percent of a country’s Facebook users. In the U.S., it’s about 2.2 million people. In Canada, it’s about 250,000. The largest you can create is 10 percent.
One percent and 2 percent Audiences are usually enough for most businesses. Most businesses are advertising to drive sales, so it makes sense to model your customer list when creating Lookalikes.
However, if your goal is inexpensive traffic, you might choose recent website visitors as your model.
If you want people to download your app, you can create a Lookalike Audience from recent App installs.
How to Create the Best Lookalike Audiences
When most new advertisers think about Lookalike Audiences, they think the best Model Audience is their existing customer list. While this seems like a good idea, with one small extra step you can make a Lookalike Model that’s made up of your very best customers and find more like them. If you upload your entire customer list, you’re telling Facebook that your best customer who might have spent thousands or tens of thousands of dollars is equally as valuable as someone who has only spent $10 and then asked for a refund! Using the segment of your customer list that represents only your best customers is a better way to create a model. While you probably know who some of your best customers are by name, the way you create a “best customer” model is through the RFM process described earlier in this chapter.
Here is a way to approximate your best buyers using RFM segmentation:
- Recency: Export a list of your orders and keep the newest 20 percent. Don’t go back more than about 18 months.
- Frequency: Sort and count by customer. Take the top 20 percent of the most frequent customers.
- Monetary: Export a list of customers with their total spending and take the top 20 percent.
- Combine the lists and you’ll see that many customers will overlap. The total number of customers on the list should be between 500 and 5,000 to build a good model. If you have more than that, use the top 10 percent or 15 percent rather than 20 percent.
- Upload this list as a Customer File Audience and call it Customer RFM Model or Best Customers.
You could also choose to model a file of prospects or any other outside file of people. Just remember, it should have about 500 to 5,000 members to be the optimum size for modelling.
To start, choose the customer file you want to model. Then choose Actions -> Create Lookalike.
You’ll be presented with an option of Value-Based Sources or Other Sources as shown in Figure 5-10 on page 61. Value-Based Sources include sources with sales transaction data like a Website Visitor (on the checkout page) or Customer File Upload (with an LTV column of data), and they are used if you are prospecting for more customers. Under Other Sources, you’ll see the rest of your Audiences.
— Lookalike Audience Sources Screenshot —
Once you’ve chosen your Lookalike Source Audience, choose the country you want to advertise to. Then, create one Lookalike Audience at 1 percent size. In the U.S., it will give you an Audience of a little over 2 million people as you can see in Figure 5-11 on page 62.
Some larger advertisers with larger budgets exceeding $1,000 per day choose 2 percent as their Audience these days and get good results. This gives them about 4.4 million people in the U.S. while giving the Optimizer more data points to choose from.
Your Lookalike Audience will be fully populated in 15 minutes to several hours. One feature of these audiences is that they are dynamic. The Facebook AI is constantly working and this Audience will refresh with new users at a rate of about 30 percent a month.
Lookalike Audiences are very powerful. Back in Austin, Facebook stressed their ability to find the right target for our ads.
After hundreds of tests, we have found that to be true. At least 85 percent of the time, in our estimation, a Lookalike Audience will achieve lower costs and better cost-per-action than an Audience we try to build using Interest or Detailed Targeting.
— Lookalike Audience Builder Screenshot —
This innovation has made life better for advertisers. By essentially using powerful AI technology to automate the bidding process and select the very best audiences for you, all your time can be spent on the thing they haven’t figured out how to automate yet, Creative. Creative is covered extensively in Chapters 12 through 17.
USING DETAILED TARGETING TO NARROW YOUR AUDIENCE
Oftentimes, Lookalikes on their own aren’t quite enough for you to build a perfect target audience. You need to be able to narrow down your audience by age, gender, location, or language.
You still have some ability to target people based on the categories and interests of Facebook users, but since the privacy scandal, many of the most useful categories and criteria have been removed. Many of the remaining options are much too broad to be effective for targeting the right audiences.
You will learn about this in Chapters 9 to 11 where everything about Ad Sets is covered.
