• May 15, 17
  • Ronnie Andrade

Want to know how to get a digital marketer’s attention? Tell them you can take their most valued audience, and be able to grow it two, heck, three times its original size. This isn’t a magic trick, though its effects can certainly seem to be. The solution is quite simple actually; it’s called Lookalike modeling. 

 (image soruce: https://www.linkedin.com/pulse/grow-your-sales-lookalike-modeling-brett-parker) 

In the last few years, we have seen the introduction of countless new targeting tools, specifically with strides in mobile & cross device interactivity. Despite these advances in data and targeting, nothing beats what already has been proven; the insights we have seen from users who have taken a desirable action. A study by eXelate showed that 73% of US ad agencies and 64% of advertisers use Lookalike Modeling to enhance their targeting marketing focus. Additionally, of the agencies using the strategy, nearly 50% saw between 2x and 3x better performance of campaigns.While Lookalike modeling isn’t by any means a new tactic, the value it provides within a prospecting and performance basis warrants it to not be overlooked. 
In short, Lookalike modeling consists of taking existing audiences and creating similar expansions based off behavioral and demographic correlations. What highly affects the quality of a lookalike model depends on the respective algorithm at play, as well as the quality of the original “parent” audience. At Direct Agents, we have managed multiple digital campaigns with clients in various scenarios. The process has allowed us to gain some valuable insights to drive effective Lookalike testing.

Creating Effective Parent Audience Sources

A lookalike model is only as effective as the parent audience it is created from. For this reason, having clear audience segmentation is key to effective Lookalike testing. To get the best mileage from your data, we would recommend the following requisites for a strong parent audience.

  • Audience Size – When creating a lookalike model, a platform (such as Facebook) analyses the Parent data points to create an expanded audience. Models usually start at around 1 million users. Because of this, Parent audience size is important. As you may expect, a lookalike model of one million will not have much correlation with a parent audience size of 100. The more data points your audience contains, the more relevant your Lookalike audience will be. However, when you have too large of an audience, you run the risk of forcing the algorithms to make connections between irrelevant traits. At Direct Agents, we see an audience sweet spot of at least 5,000 entries. Audience size should also be coupled with relevancy, which brings us to our next point.
  •  Relevance Audience Segmentation -  Clear audience segmentation is key for effective Lookalike creation and testing. Ideally, we suggest marketers to use quality audiences that are valuable by outlining the KPI’s that are relevant to their business. Don’t stop at users who only purchase, but have repeatedly purchased, or have larger cart amounts. Don’t just look at App Installs, but post install App engagement instead.

Available Audience Source

For marketers of budding startups, obtaining extensive CRM lists of LTV users over the course of 12 months might not be an option, that information might not even exist. Luckily, you can still create solid lookalike models from various sources across Facebook and programmatic.

  • Site Traffic & Engagement – Create lookalike audiences from users who have visited your site. You can take this a step further and target users who have visited multiple pages. Additionally, you can target users who invest the most time on your site. 
  • Video Viewership – Create audience groups for users who view your organic and promoted videos. Segment users based on the percentage of video viewed to create models based off you most engaged users.
  • Top performing prospecting models – Leverage top performing prospecting strategies to create lookalike models for expanded reach.  

Test, Test and Test Some More...

Creating effective lookalike models from various sources should be compounded with proper testing and CRM maintenance.  

  • Relevance vs. Reach – It’s always recommended to test relevance and reach variants of the same audience source. At Direct Agents, we tend to see that relevance models provide the best performance and return. Reach models, in turn, tend to provide prolonged performance. As these audiences are larger in volume, audience exhaust does not occur as quickly as relevance models.
  • Parent Audience Refresh – Regularly updating your CRM lists and lookalike models ensures your target audience isn’t obsolete. We would recommend an audience refresh every 3-4 months.

Attack of the Clones
In a digital world in which your can easily get overwhelmed by the newest technological advances, our job as marketers remains the same: to provide the best solution to allow our brands to grow. As an agency focused on driving direct response campaigns, we gravitate towards strategies that provide the best tangible return for our clients. Time and time again, we see that Lookalike modeling provides some of the best performance returns on a strategic level, at times only topped by retargeting tactics. The ability to create models from various audiences sources allows the tactic to be nimble enough that brands, without the luxury of extensive CRM databases, can also see impressive results from the data they do have. For these reasons, Lookalike should always be a staple to any programmatic or social campaign strategy.

-Ronnie Andrade, Senior Programmatic & Social Analyst