8 Keys to Leveraging Lookalike Audiences

They’re a marketer’s best friend. If you use them correctly.
Mae Rice
October 6, 2020
Updated: October 9, 2020
Mae Rice
October 6, 2020
Updated: October 9, 2020

In 2013, Facebook launched a new feature for advertisers: lookalike audiences. These algorithmically generated audiences allowed advertisers to dramatically scale up their reach without sacrificing sophisticated targeting.

To start, advertisers selected a “seed audience,” a group of at least 100 customers or likely prospects. Then Facebook drew on its vast data ecosystem — pulled from its eponymous social platform, Instagram, WhatsApp, its tracking pixel and more — to algorithmically generate a bigger, holistically similar audience.

The underlying logic: Your best customers probably have certain shared interests and demographic traits; that’s why your product, and your specific brand, resonates with them.

This isn’t a new idea. Companies, especially in direct mail advertising, were buying offline data sets, and crafting offline lookalike audiences, “before Facebook was a glint in Mr. Zuckerberg’s eyes,” Brett Schnittlich, president of market research firm Lucid, told Built In.

Facebook was first to apply the concept in the social media space, though, and, soon, its competitors did too. Twitter launched lookalike audiences in 2014, and Snapchat followed suit in 2016. TikTok has also rolled out lookalike audiences since its stateside launch in 2018.

At this point, the concept has proliferated across the internet to data management platforms from LiveRamp to Oracle.

“There are probably thousands of companies in the lookalike modeling space,” Lindsay Fordham, senior director of audience products at Lucid, told Built In.

What Is a Lookalike Audience?

An audience of prospective new customers, generated (often algorithmically) from a “seed audience” of current customers or likely customers. This allows companies to dramatically scale up their advertising’s reach without sacrificing sophisticated targeting. 

Facebook remains the premium digital offering though, thanks to the ubiquity of its tracking pixel and the sheer global reach of its ads.

“If you go to a website,” Dara Denney told Built In, “99 percent of the time it has a Facebook pixel.”

She would know. Denney is director of paid social at ad agency Fetch & Funnel, and she uses her YouTube channel to vlog about the nuances of Facebook ads.

Though she’s worked in content and dabbled in other marketing channels, she keeps gravitating back to Facebook ads and Facebook lookalike audiences, which she and her team find especially handy for customer acquisition.

“It’s just a platform that I find more users engage on daily,” she said.

Still — for some brands, other platforms’ lookalike audiences may provide a higher ROI. It all depends on an advertiser’s product, the strategy, and the intended audience.

So, how can you make lookalike audiences work for you? Schnittlich, Fordham and Denney weighed in.

 

Build a Seed Audience Based on Strong, Firsthand Signals If You Can...

A good rule of thumb is that the bigger your seed audience, the higher-quality your lookalike — provided everyone in the seed audience has something concrete in common. Ideally, this will be a trait the advertiser observed firsthand.

Denney often builds seed audiences from a client’s highest-value customers. Maybe they’ve made two or more purchases, or bought something in the last 30 days.

This is the ultimate firsthand data, straight out of the client’s CRM.

More on MarketingHow to Test a New Digital Marketing Channel

 

... and If You Can’t, Consider Data Sourcing

If you’re building a seed based on something you haven’t observed directly, make sure that you’re still building it around a “strong signal,” Schnittlich advised.

If you want to advertise to people on the market for a car, for instance, it’s probably better to target people whose car leases just expired than people who have visited a car dealership website once in the past year.

The former is simply a stronger signal.

If you buy a seed audience of “car intenders” from a data broker, though, it can be hard to know what signals they’ve relied on. Some data companies buy from a lot of different sources, and their methods can be a black box.

It’s worth asking for details, though. “Models based on models are problematic,” Schnittlich said.

And if you’re not careful, you might use someone else’s lookalike audience as a seed for your lookalike audience — a recipe for inaccuracy.

 

Third-Party Validation Can Come in Handy

When in doubt, you can try a data set before you buy it with a third-party validation product.

Lucid, a market research firm that can administer digital surveys at scale, offers one such product called “Data Score.” Advertisers can use it to validate a data broker’s claims about seed audience data, or check a lookalike audience’s similarity to a seed audience (outside of walled gardens like Facebook).

Essentially, it works by administering digital surveys to a sample of people within a data set, asking them directly if they have the intention or trait the advertiser wants to target.

 

Never Stop Testing

Of course, most ad campaigns have many possible seed audiences. Which ones will produce the highest-performing lookalike audiences? Denney finds out through constant, small experiments.

“We have to do testing on anywhere from five to even 15 lookalike audiences to find the one that knocks it out,” she said.

Even when she finds a seed that performs well based on common performance metrics — like click-through rate, cost per click and ROI — it doesn’t work forever.

For the window in which it does, she shifts her focus to trying to optimize the ad’s imagery and copy.

“It’s not like testing is ever done,” she said.

 

Try Stacking Lookalikes

Often, Denney gets her best results from “stacking,” or combining, lookalike audiences.

She might, for instance, build one lookalike audience from a “seed” of people who bought a certain product within the last 180 days, and another from a “seed” of people who bought within the last 30 days.

Combining them into a single target audience “gives us a better chance of success,” Denney said, because it gives Facebook’s algorithm more room to choose who sees her client’s ad.

Lately, “leaning into the algorithm” on Facebook has been getting her better ad performance than hyper-precise targeting. In fact, Facebook’s algorithm has been working so well for her that she sometimes drops targeting from her client’s campaigns altogether.

 

If Your Audience Can Be Found Off Facebook, That’s a ‘Slam Dunk for ROI

If a company’s customers use multiple social platforms, it’s cheaper to advertise to them on a niche one. One of Denney’s clients sells electric skateboards that cost over $1,000, and she finds that they get double the ROI when advertising with Snapchat’s lookalikes that they do with Facebook’s.

The platform is just cheaper — Denney estimates the cost per thousand views (CPM) on Snapchat is about a fifth of what it is on Facebook — and denser with younger people who might want to invest in a skateboard.

(It’s not just for teens, though — this year, Snapchat reached 75 percent of Americans ages 13-34, the company reports.)

“If you have an audience that would be better suited to Snapchat, that’s a slam dunk,” she said. Same goes for other social platforms younger and less ubiquitous than Facebook.

 

Don’t Forget to Gut Check

In other words: Watch out for data brokers selling impossible audiences.

“If someone says, I have 300 million unique [Americans] who are interested in buying a car in the next six months,” Fordham said, “that’s probably a sign that the data segment is low quality.”

She estimates that, at any given time, about 12 percent of Americans are shopping for a car. If 300 million unique Americans were looking for one, that would mean nearly everyone in the United States was poised to buy a car — including children. 

 

Have Realistic Expectations

No matter how strong your seed audience, no lookalike audience is 100 percent packed with the advertiser’s target demographic. There’s always some algorithmic guesswork involved in lookalike modeling, Schnittlich noted, and perfection is an unrealistic standard.

Instead, advertisers should seek out lookalikes that are denser in their desired intention or trait than the population at large. If a lookalike doesn’t hit that threshold, it’s not worth its premium price tag, and “you should just be buying reach,” Fordham said.

More on MarketingDisabling Third-Party Cookies Won’t Improve Online Privacy

Great Companies Need Great People. That's Where We Come In.

Recruit With Us