Optimizing Your Media Targeting with Better Lookalike Modeling
You put a lot of time and effort into your ad campaigns, but all that’s wasted if the right people don’t see them. That’s why effective and efficient targeting is so important. With lookalike modeling, the ad industry has taken a giant step forward from simply targeting by demographics. But that’s not the end of the story.
Lookalike models are only as good as the data that feeds them. Your marketing goals probably include acquiring higher quality customers, improving customer lifetime value, and improving the overall ROI of your acquisition media. The key to achieving all of these goals is use your own 1st party customer data (data that your brand owns) to power the best lookalike models possible. It’s not hard, but you do need access to customer data from a few difference sources. In this post, we walk you through the steps to optimize your ROAS with lookalike modeling done right.
The Status Quo
Advertising on the web isn’t new, and there are a variety of common targeting strategies. Here we walk through the two most common, but suboptimal, approaches, followed by our recommendations for how best to optimize your media targeting strategy.
Okay: Demographic and Interest Targeting
This approach is “okay” because any targeting is better than “spraying and praying” with no targeting at all. Most marketers will opt for broad, demographic targets like women; ages 30-50; interested in a certain product or category (e.g. “running” for a sportswear company or “sun destinations” for an airline). Unfortunately, the quality of the data used in this approach is relatively poor. This is because it’s based on inferences made using browser and device behaviors across many different websites. If you haven’t already, take a look at your own anonymous profile based on your browser behaviors. It’s comical how wrong many of these attributes can be.
As a result, purchasing generic 3rd party audiences is often an inefficient way to find and target high value audiences and generally yields lower returns.
Better: Pixel-Based Lookalike Modeling
Another strategy involves placing a pixel on your eCommerce site and building lookalike models based on your own site visitors. This can be a better approach because it leverages a portion of your own 1st party data which will always be higher quality and more relevant than externally sourced data. But there are limitations.
First, it requires cookie matching from your site to a 3rd-party data provider’s anonymous dataset, which is an inherently lossy process.
Second, many of the people identified by the pixel will be same ones you identified above – external audiences of a mediocre quality. Therefore you are creating a self-fulfilling cycle of buying generic 3rd party audiences, driving them to your site, tagging them, and re-targeting these same poor quality audiences.
Finally, eCommerce sites typically have low conversion rates, topping out at 3-5% on average. This means that the vast majority of your site traffic is not made up of your best customers, but of “window shoppers” instead.
Best: High-Value Customer Lookalike Modeling
As we saw above, lookalike modeling is the better approach, but models fed on low quality data don’t produce great results. Instead they need current, accurate, and rich 1st party data to perform at their best.
The key is to build target audiences using your highest value customers across channels, not from everyone who visits your site. You can get an accurate picture of your best customers by bringing together online and offline transaction data and loyalty data, and using it to calculate lifetime value. Through this process, most brands find that a very small group of customers (usually about 10%) drive 30-50% of their annual sales.
You can then use segments that have been narrowed down to only your highest-value customers to fuel your lookalike models, for much better results. This is because your model is founded off of narrow, precise, high quality data, and uses powerful signals from smaller segments to discover and target larger audiences with similar qualities. Many lookalike modeling tools also provide dials to trade off the reach of a campaign against the similarity of the lookalike audience, empowering you as the marketer to balance accuracy with reach based on your campaign goals.
Step by Step Process
Here we outline the critical steps to lookalike modeling done right:
- Create an omni-channel view** of your customers and their transactions. Data sources should include, at minimum, eCommerce and POS transactions, site behavior data, and loyalty membership data.
- Calculate the lifetime value for all of your customers, potentially including attributes such as purchase frequency and customer profitability in addition to top-line revenue.
- Create segments for your highest value customers overall and highest-value customers by product category.
- Feed these rich, 1st party data segments into lookalike models to produce your target audiences. This can be done in two ways:
- The quickest path, and one that produces strong results with no additional effort from your team, is to load your segments into pre-built modeling tools within your media buying platforms (e.g. Facebook).
- The other path is useful for brands who want even more control over their models, to ensure that very specific types of customers are targeted. In this case, brands send data into their data warehouse, run models themselves, and then export those models to media platforms for targeting. In general this is a more advanced technique and not a starting place we recommend.
- Advertise to the resulting precise, high-quality audiences with relevant offers and products.
To learn about how one Fortune 500 brand followed these steps to test and optimize their media targeting, read the case study: Better Lookalike Modeling for Media Targeting
**What to do if you lack an omni-channel view
For many brands, building the prerequisite omni-channel view of customers is not so easy. This is where a technology like a Customer Data Platform (CDP) comes in. CDPs like Amperity bring together online and offline data from all your 1st party data sources to form rich customer profiles. Because it sits at the foundation of your marketing technology stack, a CDP gives you the ability to build detailed customer segments and power a wide variety of marketing and advertising use cases.
To learn more about building a great customer data foundation, read the white paper: The Marketer’s Guide to First-Party Customer Data Unification