ShareAcross all digital marketing channels, you have the ability to implement Lookalike (LAL) audiences (as long as you have pixels installed), giving you access to endless research possibilities. Read on to discover what a lookalike audience is and how to scale them to test your campaigns.
What is a lookalike audience?
All LAL audiences start with a seed audience (a group of users performing an action you want, like buying a product), then the platform will use its machine learning technology to find a similar pool of users (known as a LAL audience), aiming to improve the performance of your ads.
Your seed audience could be anything from website visitors or converters to profile page data, ad engagers or email lists – the more metrics you can test, the better.
Generally speaking, broader seed audiences aren’t necessarily a better audience. The ad platform is relying on finding similarities between users in your seed audience. So, if you were to have tens of thousands of users within your seed audience, then the platform will give you a less accurate LAL audience – as there will be fewer attributes that match across the segment at scale.
How do lookalike audiences work?
Lookalikes typically work in a percentile format and the percentile refers to how homogeneous (similar) the audience is.
For example, 1% of LAL will choose the top 1% of users who share the most common traits as the seed audience, and 5% or 10% will have fewer traits in common and therefore become a less homogeneous audience. A 1% LAL in the UK would be 1% of the US users’ population that is most similar to your audience: this audience typically has a reach of about 2 million people (or 20 million for a 10% LAL).
How to scale lookalike audiences
When testing different lookalike percentiles, large audiences have a negligible overlap – so there is no need to exclude each LAL from one another. Similarly, when you are using lookalikes of different seed audiences, you don’t need to exclude one from the other – as the lookalike segments will be large enough.
Another way to scale your LAL is to create a ‘super’ LAL segment, this is where you combine multiple LAL into one blended audience. When using this method, it’s best to stick to low-percentile LAL (as you will be combining different seed audiences).
You can do this by creating a LAL for each of your best-performing seed audiences, and grouping them all into one ad set. You could also do this on a broader scale, by having 2 or 3 different ad sets that all target different LAL audience combinations.
You can then build on this further once you’ve found your best-performing super lookalike combination, then continue testing by extending the percentiles.
Alternatively, you can scale your lookalikes to test the recency of your data. When you create your custom audiences, you typically have the option to set a look-back window: like added to cart in the last 7 days or the last 180 days, and anything in between. These look-back windows will create very different segments, allowing you to experiment using seed audiences from the same event. If you don’t have as many events to test from, you can use shorter and longer lookback windows.
Or if you’re using CRM data, you can test two different approaches such as the highest lifetime value or highest cart value.
Explore your lookalike audiences
Ultimately, the possibilities are endless with lookalike audiences – there are different options to scale depending on the granularity of your pixel or CRM data.
Need some help? Get in touch with our friendly experts to test your LAL audiences.
