This is the subhead for the blog post
Lookalike targeting on Facebook is one of the best targeting options out there, but marketers using it without testing its nuances aren’t getting the most out of it. This article will show some of the insights we got from a client in the technology sector and hopefully give some testing ideas for your next campaign.
Audience generation methods – which to choose?
The first step to creating a quality lookalike audience is to find the best source. This can be done in many ways; you can choose to upload email addresses or IDFA directly to Facebook, or create a custom audience using Facebook Pixel or Facebook SDK.
There are pros and cons to both. Using static email lists means you have control over what Facebook is using, whether it be high-LTV users or top-performing users, but it also means that you need to keep the seed audience fresh by updating it every other month.
Using custom audiences built off of website traffic or SDK eliminates the need to constantly update the seed audience, and it allows Facebook to dynamically capture new audience and feed to the algorithm.
So which method provides better performance?
Based on the above performance, pixel-based lookalike audiences deliver the most volume, the strongest result rate, and the lowest CPA. There are a couple reasons for this:
- Pixel-based seed audiences automatically update on a daily basis; you are always capturing the most up-to-date user information and feeding it to Facebook’s algorithm.
- Static upload quality depends on the match rate; if your match rate is 50%, this means that only 50% of your list is used to make lookalike, thus reducing the effectiveness of the feature.
Now that we have picked the best seed audience, testing different permutations of the lookalike audience is important. You can create lookalike audiences from 1% to 10%, 1% being the smallest and most closely matched to your user base and 10% being the largest and least similar. We always recommend testing lower percentages as they deliver the best quality; for this client, we tested lookalike audience from 1% to 6% using the Nested Lookalike strategy.
As expected, 1% delivers the best performance. As we open up to larger lookalike audiences, the result rates decline and CPA increases. While we do see 1% delivering the best performance, it’s important to keep in mind that volume is limited with 1% lookalike, and that larger percentage is important for scale as well.
Layer Behaviors or Interests for Targeting Punch
In order to get additional scale, we tested behavior layering on top of Facebook’s lookalike feature. With larger percentage lookalikes, we usually get more scale but lower quality as well. Layering additional interests or behavior helps narrow and further qualify the lookalike audience. As expected, we saw 37% reduction in CPA, and with better performance, we were able to spend more on this audience.