Lookalike audiences are one of the most powerful audience-targeting tools. Lookalike audiences help advertisers use their customer data to target new potential customers with similar characteristics to existing ones. An e-commerce site with a list of email addresses can upload that list to Facebook or Google and they’ll automatically show your ads to new potential customers who are similar to (or “look like”) them.

The good news is Facebook and Google are highly effective at finding similar-looking audiences to the ones you provide due to the vast amount of information they have on their users and their AI and machine learning capabilities. The bad news is many advertisers don’t provide Facebook and Google with a list of their most valuable customers, and recent changes in Apple and Google’s privacy policies are impacting ad platforms’ ability to create effective lookalike audiences.

In this article, we aim to help companies understand how to create more effective lookalike audiences using predicted customer lifetime value that will increase the return on their ad spend (ROAS) on their digital marketing campaigns.

Not all customers are equal

If you’re like many advertisers, you build your lookalike audience by uploading a list of all customers that have visited your website, subscribed to your email, and/or made a purchase. Or perhaps you segment your customer list based on recency (e.g., 12-month buyers), frequency (e.g., two-time buyers), or monetary value (e.g., average order value) to create value-based lookalike audiences.

The issue with these approaches is not all customers are equal and have the same expected value to a company. Think about your business. Do all customers spend the same amount, buy the same thing, or buy with the same frequency? Does a customer purchase yesterday, indicate that they will return to purchase again in the future? Of course not. Some customers spend more and buy often, and are thus highly valuable. While others only buy when there is a sale, or buy once and never again, and are less valuable.

Customer lifetime value

Successful companies understand these differences in customer behavior and customer value. They recognize that some customers will generate a significant portion of their revenue and profits, and are highly valuable, and that other customers will buy sparingly or not at all, and are not as valuable. Importantly, they understand that past buying behavior may not represent future buying behavior and invest the time and resources to predict how valuable the customers are going forward to create more effective lookalike audiences.

The gold standard in marketing for predicting a customer’s future value is a metric called customer lifetime value (CLV). CLV is sometimes known by other names, including lifetime value (LTV), and has several definition variants. Most accurately defined, CLV represents the net present value of how much profit a company is expected to earn on a single customer throughout that customer’s relationship with a company.

While usually expressed as a single number, CLV involves predicting multiple elements of customer behavior including:

• Number of orders: How many purchases a customer will make

• Time period: How long a customer will do business with a company for

• Average order value: How much a customer will spend on each purchase

• Profit margin: How much profit a purchase will generate after product costs

• Customer acquisition cost: How much it costs to acquire a customer (CAC)

There are multiple modeling methods available to predict CLV, of which the details are beyond the scope of this article. Some of the more popular methods involve machine learning models such as deep neural network (DNN) models and probabilistic models such as the beta-geometric negative binomial distribution (BG/NBD) models. At SabinoDB our data scientists use a proprietary model trained on each client’s data to predict the CLV of each customer.

The value of predicting customer lifetime value

Creating a lookalike audience with a list of all customers that have visited your website, made a purchase, and/or a segment of your customers based on RFM can allow ad networks to target similar customers as your existing ones. Using Facebook’s out-of-the-box value-based lookalike audience which allows you to include historical transaction data can even improve these lookalike audiences. But if you want ad networks to target the best audiences possible, you want to create lookalike audiences based on which customers are expected to have the highest value going forward by using predicted CLV.

Predicted CLV is forward-looking and takes into account how much a customer is expected to purchase going in the future and how much those future transactions are worth to a company. Estimating the predicted CLV for each customer requires a little investment up front into predictive analytics, but is well worth it. Providing this data to ad networks like Google and Facebook can greatly improve their ability to create higher-value lookalike audiences that stand to earn advertisers a better ROI on their ad spend. For example, instead of creating a lookalike audience from all its customers or any customer that has made a purchase in the last 12 months, an e-commerce site can create a lookalike audience based on the top 1%, 5%, 10%, or 20% of their highest predicted CLV customers, depending on the size of their customer database. Facebook and Google’s AI tools will then use this new information on your customers to look for what high CLV customer have in common and then shows your ads to similar high value customers on their platforms.

CLV also has tremendous value outside of deciding which customers to target with advertising and creating lookalike audiences. For example, CLV can help advertisers decide how much to invest in acquiring a customer, as well as answer questions such as where your highest value customers are acquired from and what types of products your highest value customers buy first.

Summary

Targeting your ads to the “right” audience is essential. Online ad platforms such as Facebook and Google offer advertisers a powerful tool to target new customers called lookalike audiences. Many advertisers build lookalike audiences using a list of all customers that have visited their website, made a purchase, and/or a segment of customers based on recency, frequency, and monetary value. The issue with these approaches is they fail to distinguish between high-value and low-value customers. Successful companies invest their time and resources into understanding their customer behavior, predicting the CLV of each customer, and building lookalike audiences using CLV. Using CLV-based lookalike audiences can greatly increase Facebook and Google’s ability to help companies acquiring higher-value customers and in turn, lead to better ad results and increased sales and profits.

SabinoDB is an e-commerce and marketing solutions company committed to helping companies use big data, real-time analytics, AI, and machine learning to better understand their customers and grow their business. If you are interested in learning more about how SabinoDB can assist your business in calculating predicted customer lifetime value to create more effective lookalike audiences or how SabinoDB can help your company with its digital advertising, please reach out directly to Ryan Hammon (Email – ryan.hammon@sabinodb.com  / Phone – (415) 847-8103), or contact us below.

Sources

Meta Business Help Center. “About Lookalike Audiences.” Retrieved https://www.facebook.com/business/help/164749007013531?id=401668390442328

Facebook for Business. “Value-based lookalike audiences.” Retrieved https://www.facebook.com/business/m/signalshealth/target/value-based-lookalike-audiences

Google Ads Help. “About similar segments on the Display Network.” Retrieved https://support.google.com/google-ads/answer/2676774?hl=en

Stratechery. “Privacy Labels and Lookalike Audience.” Retrieved https://stratechery.com/2020/privacy-labels-and-lookalike-audiences/

Knowledge at Wharton. “How to Find Your Most Valuable Customers.” Retrieved https://knowledge.wharton.upenn.edu/article/160811b_kwradio_fader-mariychin-mp3-zodiac/

McCarthy, Daniel. “A general framework for customer lifetime value.” Retrieved https://www.dropbox.com/s/xjak7pezn6i9m06/CLV%20framework.pptx?dl=0   

Google Cloud Architecture Center. “Predicting Customer Lifetime Value With AI Platform: Introduction.” Retrieved https://cloud.google.com/architecture/clv-prediction-with-offline-training-intro