Lookalike Audience Gridding: Making the Most of Facebook Lookalikes
by Kendall Beyeler | August 16
Too much of a good thing doesn’t sound like much of a problem, especially when that “good thing” is a fresh, new Facebook audience that has a lot in common with your existing customers. Facebook’s Lookalike Audience feature was designed to do just that: give you access to statistical clones or “lookalikes” of your best targets.
Generating a huge Lookalike Audience is easy to do. If you send a seed list of hundreds or thousands of names, your Lookalike Audience can return millions of similar people. Finding the best leads in an audience that size is a needle-in-a-haystack proposition. A very big haystack. And since there are many different types of lookalikes you can create, it’s actually like trying to find a needle in 30 haystacks!
Here’s a tip for getting the most out of your Lookalike Audience: Use Lookalike Gridding to segment an undifferentiated audience into meaningful sub-groups. By segmenting, you can derive much more value, much more quickly.
How? In many ways. First, gridding can help you identify expensive customer groups so you can direct spending to better-performing segments. Gridding can also give you better insight into what drives Lookalike performance. For example, if you layer demographic age targeting on top of your Lookalikes, you can see the effect age groups have on the model.
In an undifferentiated audience, factors like these are totally obscured.
The diagram above (Fig. 1) shows the Lookalike Audience Gridding process of a hypothetical e-commerce company. The “Y” axis is a typical checkout funnel: landed on site, viewed product, placed product in cart, checked out. The “X” axis is divided into three groups: conversion pixel, email, and telephone phone number. Each cell is a discrete audience, and now there are 15 of them. To create even more variation among the Lookalikes, you could use recency. By Gridding, you can subdivide (for example) Lookalikes who added items to their cart via the pixel in the last 7, 14, 30 or 60 days, and so on. Each of variation gives you a distinct slice of the audience. Using your own criteria, you might easily produce a list of 100 different audiences from the master list of Lookalikes – and you’d know they were the audiences you want.
How does this play out in the real world? One of our clients was running ads to a handful of audiences. Using Lookalike Gridding, we identified 45 subgroups in that handful and ran ads to them simultaneously. As a result, their cost per lead dropped 75% in two weeks.
We’ve found that it rarely makes sense to restrict the Lookalike Audience search to existing customers without testing other combinations first. If you try to build an accurate statistical model, more data is better. Typically, “existing customers” is a far smaller group than all those who hit your site. While intuition says that existing customers would be more indicative of traits of future customers, reality might contradict intuition. It is impossible to know until you test.
We’ve seen clients create new huge business streams from a variety of counter-intuitive sources: email newsletter, pixel, add to cart and more. By segmenting the Lookalike Audience, and testing for the differences among those segments, you can allocate more precisely and know that you are not missing hidden opportunities.
What’s your experience been with Lookalike Audiences? Have you tried Gridding or something like it? Get in touch with us to learn more about Lookalikes and audience opportunities.
PS – we can’t take complete credit for this approach. We are friends with the guys at Acquisition Labs, who initially suggested segmenting in this way. We’ve iterated on the approach since then, but our hats off to them for identifying this strategy.