Etgar Shpivak, CEO, Fixel.ai
Lookalike audiences. Today’s guest post comes to us from Etgar Shpivak, the CEO of Fixel.ai, who explains the ins and outs of how to scale this valuable and popular type of audience targeting.
Lookalike Audiences are one of the most powerful tools in the modern marketer’s arsenal. Often used to expand existing audiences to new locations and products, their true potential is broader and can help scale virtually anything.
First introduced by Facebook in 2013, Lookalike audiences are now an industry staple, available on many ad platforms including Facebook, Google Ads, Outbrain and Quora to name a few. The unifying premise is that given a group of users, the ad platform’s algorithm can assemble a second group that resembles the original one. Facebook even added a degree of similarity ranging from 1% difference to 20% difference sample size. Let’s break it down and dive a little deeper into what lookalike audiences are and what you need to know to consistently build effective lookalikes.
All lookalikes start with a seed. This is the ‘first group’ that we mentioned earlier and it’s the basis of the audience. Much like its plant-based analogy, not all seeds are created equal and picking the right one is a bit of an art form, but we’ll get to that shortly.
Generally speaking the ideal size for a lookalike audience is around two thousand users. Contrary to intuition a bigger seed does not mean a better audience, in fact, the opposite is true. If your lookalike seed is made up of tens of thousands of users, the ad platform will have a difficult time finding users that match with everyone in the group, resulting in a less accurate lookalike audience.
Lookalike audiences take the top percentile of visitors that share the most common traits as the audience’s seed. Choosing the top 1% will result in a smaller but more homogeneous audience. Choosing a larger percentile, e.g. 5% or 10%, will result in a larger audience that is less homogeneous. It’s a classic balancing act of scale vs effectiveness.
There are many ‘best practices’ for building lookalike seeds. These are often effective for a specific use case and less so for another. In reality, there isn’t a single clear, cookie-cutter solution that works for everyone, rather it’s important to match the lookalike seed to the available data, metrics and business goals. To that end, we can offer a few guiding principles.
There are many metrics that can be used as the basis for a lookalike seed. Here are a few practical examples:
CRM Data refers to information that’s typically stored in a CRM and has a broader historical context. These metrics tend to be more strategic and long term. For example:
Conversion Data refers to the ‘Hard’ actions taken by a visitor, actions that signal intent and are fairly simple to measure. For example:
Engagement Data refers to ‘Soft’ actions taken by a user that are often more complex to measure and understand intent. For example:
It’s important to note that many of these metrics are simpler or lower grade versions of each other. For example, a Past Purchase Data can be substituted by Add to Cart Audiences which in turn can be substituted by Product Viewers who were highly engaged.
Metric substitution allows you to create consistent lookalike seeds, which is very important.
One of the key factors to keep in mind when building a lookalike seed is consistency. The basic rule being, the more consistent the lookalike seed is, the easier it is for Facebook’s algorithm to find users that match it.
To achieve consistency, it’s a better idea to use data from a specific category, even if the quality of the actions is not as good. For example, using add to cart data from a specific category should be preferred to overall purchase data across all categories. To explain the reasoning, consider a convenience store that caters to many different consumer types. Using purchase data without a category context would result in a seed that has both mothers who purchased diapers, men who purchased razors and aftershave next to athletes who purchased muscle balm. The lookalike seed in this example would be very ineffective because the sample groups do not share the same purchase intent.
To keep with the principle of consistency it is important to look beyond raw metrics and into the user journey, understanding the unifying motivations and context that drove the specific behavior you’d like to replicate in the lookalike audience.
To explain this principle, consider Valentine’s Day for the same CVS styled store. A great lookalike seed could be built based last years purchase data in the period leading up to and immediately after the holiday. The logic being that purchases made in the perfume department at this time frame are gifts and the same type of people would be interested in buying them again come next holiday season.
Value-based lookalikes are the most commonly used lookalike seeds. When done correctly they deliver some of the most effective lookalike audiences available.
It is, however, important to remember that consistency is key and it’s better to have a consistent audience seed of users who Added a Product to Cart over an inconsistent audience based on lifetime value or purchases.
The holiday season provides massive sales and growth opportunities for eCommerce businesses. However, with media prices reaching an all-time high, it’s important to focus on the right audiences.
An effective tactic is to create 1% lookalikes based on newly acquired customers from the past years season, these audiences will represent people that found your products or services useful within the context of the holiday season. This tactic also works for any kind of recurring event or enterprise. From conferences to live events or seasonal services.
For a newly formed agency offering several services or an eCommerce business offering several distinct products, waiting until each product or service reaches enough sales or deals to make for a solid lookalike seed might be impractical or simply take too long. Many in this case sacrifice granularity and use the general ‘Purchases’ metric as a lookalike seed. This is a common mistake. It’s better to use engagement metrics to keep your lookalike seeds consistent. Ultimately, engagement-based lookalikes allow for faster and more granular scaling.
When launching a new product it’s often difficult to know who to target your product at. An interesting solution involves targeting a broad audience and then using click conversions as a lookalike seed, trying to find a unifying factor in the audience.
Despite the obvious advantages lookalike audiences have, they are not all-powerful and even Ad Platform algorithms have limitations. If there’s one takeaway that sticks with you from this article, let it be the importance of seed consistency and granularity. Ultimately, building a great lookalike audience requires the balanced use of metrics coupled with the ability to see beyond them into the motivations of your users, while making sure your lookalike seeds are consistent and granular.
This approach will help you use this powerful feature in new and specific ways that are relevant to you, effectively growing and scaling anything.
Request a Demo of Kenshoo