How to succeed using look-a-like targeting
Look-a-likes have been the bedrock of digital marketing since their inception.
There are two ways to improve a look-a-like’s performance.
Improve the quality of the source audience
Calculate the optimal size of the audience
What is a look-a-like?
They allow you to reach people who are similar to your existing customers. Increasing the probability of affinity for your product increases conversion and revenue.
There are two inputs when building a look-a-like:
A source audience
The desired final audience size
The ad network uses the source audience to create a new, broader audience to expand your reach to the desired size, between 1-10% of a country or region's total population.
How do they get made?
An algorithm scores the network’s users based on their similarity to the source audience. The look-a-like is made of the highest-scoring users. The number of users is based on the size you request, 1-10%.
How do you properly seed the custom audience?
Many advertisers use dynamic audiences, where the network tracks who visits and purchases and then updates the source audience in real-time. They then regenerate the look-a-like based on this dynamic audience.
While fast and easy. There are two massive issues with this approach.
Not all visitors and customers are made equal. Some are worth far more.
Any skew within the data will produce a skewed output that will underperform. This problem commonly arises when a company expands from its initial audience to the broader population. Imagine you start on the West Coast but are expanding into Middle America and the East Coast. There will be a considerable divergence between your current audience’s interests and the target for expansion. The geographic skew will create a look-a-like heavily skewed toward the West Coast as a high degree of shared regional interests will skew the output.
Thankfully, you can adjust the source audience and minimize these issues.
Compensate for skew within the data by sampling your user base to mimic the general population. For example, if your userbase is heavily concentrated in a single geography—20% of the overall population but 50% of your userbase—limit users from that region to 20% of the source audience.
If there is a large difference in behavior, segment the user base into separate source audiences. Begin by focusing the sample on high-value customers who generate the highest gross profit.
Include a more significant proportion of recent customers in the source audience. While you don’t want the source audience to be entirely new customers, recent customers should share more traits with future customers.
What’s the optimal look-a-like size?
I’ve seen two approaches to selecting the size of a look-a-like. Either the company randomly chooses a size and see if it performs. Or they produce 2-3 different-sized look-a-like to compare performance.
Neither is right.
The correct approach. The Donut Hole. Create three different sized look-a-likes: 2%, 5%, 8%. Then, you build 3 ad sets within the same campaign with the targeting like this:
2% LAL
5% LAL and excluding the 2% LAL
8% LAL and excluding the 5% LAL
You are cutting out the center of the look-a-like by excluding the smaller one. This format calculates the marginal performance change of expanding the look-a-like. Without the exclusion, you’re measuring the overall look-a-like performance, not the marginal performance of the expansion.