A New Approach to Acquiring Loyal Customers on Meta

Introduction

Meta underperforms in acquiring high-LTV customers for apparel brands. In this paper we will discuss a new approach for deploying Meta customer acquisition campaigns to acquire higher quality and more profitable customers.

A chronic problem within business as usual is the acquisition of one-and-done customers. If the brand’s average ROAS is under 2.5, it’s likely the one-and-done customer is a drag on profits. PreciseTarget is a specialist AI company focused on improving customer acquisition in the soft goods retail sector. I lead a team of data scientists and engineers focused on delivering targeting audiences for the Meta and Google platforms. We’re unique in our focus on a particular problem: helping apparel and footwear brands acquire higher quality, higher profit, customers. We have an amplified focus on repeat purchasing, which has yielded deep experience in developing targeting audiences to acquire high LTV customers in the apparel, footwear, and beauty segments. We help brands acquire higher quality customers, as measured by repeat purchase rates through the use of sophisticated AI modeling. It’s likely you aren’t reaching these customers in your current digital marketing campaigns.

In our discussions with clients, we often encounter an assumption that Meta’s AI will eventually solve the customer quality problem. Meta’s marketing team has aggressively marketed their AI, leading many to prematurely conclude customer quality is on Meta’s roadmap. In this paper we will provide a hands-on data science perspective, explaining why it’s unlikely Meta will ever solve this problem. While Meta’s AI is impressive, the customer quality problem may be beyond their reach.

Retail Data Sparsity: A Challenge for Both Retailers and Meta

Data experts in the retail industry live in a world of sparse data. The challenges are numerous, including the difficulty of unifying a customer’s in-store and online purchases, the anonymity of wholesale purchases, and the fact that consumers have very little loyalty to brands. Consumers spread their purchases across countless sellers and often show indifference to whether they purchase in the wholesale channel or directly from the brand (DTC). This sparse data problem has historically plagued retail marketers and is now a significant challenge for Meta.

AI systems depend heavily on training data, and while Meta trains very effective models, they are limited by the types of data available for specific sectors like retail. At PreciseTarget, we believe our advantage lies in having more specialized data tailored specifically to the fashion industry, allowing us to deliver more targeted predictions.

Meta possesses high quality data about consumers in its walled garden, but lacks insight into purchasing behaviors that happen after the consumer leaves its ecosystem. Meta knows who clicks on an ad, but often lacks visibility into whether that click led to a purchase. Retail brands

experience this firsthand when they see the attribution data reported in Meta Ads Manager — many complain that it is inaccurate. This gap in attribution has fueled a plethora of third-party attribution vendors aiming to give brands a clearer picture of what’s working.

Why Meta's Sparse Data Limits AI Effectiveness

Imagine a data science team tasked with building a predictive AI model to accurately target potential buyers. They would want high-quality training data, ideally based on historical purchasing behavior in the apparel space. Training an AI model to predict future purchasers without such information is inherently flawed. Here’s why Meta doesn’t possess the necessary training data.

According to Capital One, the largest issuer of consumer credit cards, more than 80% of apparel and footwear purchases are now made using mobile phones. However, Apple’s release of iOS version 14.5 in 2021 initiated a new paradigm for protecting consumer privacy. This version of iOS enabled consumers to block apps from tracking them — and 96% of consumers elected to opt-out.

This inability to track consumers significantly restricts Meta’s ability to track and collect purchase behaviors. Privacy controls from Apple and Google continue to evolve, further hindering Meta’s ability to gather comprehensive post-click purchase information. This lack of data makes it difficult for Meta to accurately target high-value customers for apparel brands. We highlight the following examples which demonstrate structural blockages within the iOS data ecosystem. These limitations further amplify Meta’s sparse data challenge:

1. Cross-Browser Limitations: When a mobile user clicks on a Meta ad, unbeknownst to the user, the platform will launch a proprietary Meta browser. Rather than visiting the retailer’s site with Safari or Chrome, the consumer is visiting using the Meta browser. It’s common for the consumer not to complete the purchase in the initial session; the consumer likely will reflect/contemplate and make the purchase hours or days later. Let’s say the consumer ultimately decides to move forward and make the purchase on the brand’s ecommerce site the following day. This new site visit will likely be done using the Safari or Chrome browser. Apple’s do not track makes it impossible to connect the initial Meta proprietary browser event with the subsequent event visit using Safari or Chrome. The result: Meta is unable to identify that their user made a purchase. This creates two problems: Meta’s attribution data won’t be accurate, and Meta won’t be receiving training data for a targeting model.

2. Cross-Session Limitations: A significant proportion of conversions happen at the bottom of the funnel. While the shopping journey may have originated on the Meta ad click, the purchase often happens further down the funnel in search or as ‘unattributed’. Given that Apple prevents connecting events on iOS, Meta has no ability to tie purchases being made in the lower funnel to an ad click by a Meta user. This reality blinds Meta to most purchases being made by its platform users.

PreciseTarget’s Approach to Advancing the State of the Art

At PreciseTarget, we use a combination of your pseudonymized first-party data and a large fashion consumer dataset to build targeting models. The models create targeting audiences composed of high-value customers who are not yet in your customer file. Presently our dataset includes greater than 5 billion 1st party conversions, among the largest in the retail industry.

Our unique value proposition lies in our specialized approach to data integration and predictive modeling. Specifically:

• First-Party and Bespoke Data Integration: We integrate your pseudonymized first-party data with our proprietary fashion consumer data, allowing us to predict new customers who are likely to have a high lifetime value (LTV). Unlike Meta, which lacks post-click purchase visibility, we use robust historical purchasing patterns to train models designed to predict future high-value buyers.

• Detailed Consumer Insights: Our consumer insights go beyond simple ad click behavior. Our bespoke dataset includes specific individualized fashion preferences and transaction behaviors, enabling us to understand the consumer journey more comprehensively. This advantage allows us to predict the consumers who are most likely to engage repeatedly with your brand.

• Privacy-Focused Data Collection: With increasing consumer privacy measures from companies like Apple and Google, our approach to combining pseudonymized first-party data with pseudonymized bespoke datasets ensures regulatory compliance, including the CCPA, while still delivering actionable insights. Unlike Meta, which is increasingly limited by privacy controls, our privacy-first methodology has created a safe environment for customers to leverage their data assets.

• Proprietary Predictive Models: Our predictive models are built specifically for the fashion sector. Using advanced machine learning techniques, we combine your pseudonymized customer data with broader consumer behaviors to create targeting audiences. The resulting audiences are a cohort of consumers predicted to be your future high lifetime customers. Direct integration with the major ad platforms and retailer e-commerce systems enables rapid and seamless delivery. This includes targeting audiences delivered to ad platforms and customer enrichment data delivered to internal CRMs and CDPs.

• Higher Accuracy in Customer Targeting: By focusing on historical purchase behaviors and combining wholesale and retail data, our models achieve higher accuracy in predicting the best acquisition targets when compared to Meta's generalized audience models. This directly translates to lower customer acquisition costs (CAC), higher conversion rates, and greater yield on advertising spending.

Our approach fills the gaps where Meta's data limitations prevent optimal performance. Meta’s data sparsity restricts its ability to precisely identify consumers who will be future repeat buyers. Our novel approach combines high quality historical transaction history data with advanced predictive modeling methods. This enables the creation of target audiences that greatly expand a brand’s reach and higher yield on ad spending.

A Future of Symbiotic AI Systems

It’s natural for businesses to desire a universal, all-knowing AI platform. However, the reality is future systems will be composed of an ensemble of specialized AI systems working symbiotically.

Meta is unlikely to become the exclusive full-stack AI solution for advertisers. We advise our customers to expect to use a blend of AI systems within their digital marketing efforts. This ensemble approach may include AI-based attribution systems like Triple Whale and Northbeam, specialist AI audience providers like PreciseTarget, and campaign optimization vendors like Smartly.

Meta provides APIs and integration points that facilitate the use of third-party platforms. These integrations suggest Meta acknowledges the importance of third-party contributions. We believe most specialized targeting audiences will be developed within this broader ecosystem.