Autodoc Lifts ROAS by 135% with Predictive Audience Targeting

Overview

• Autodoc’s custom audiences made it difficult to maintain a stable Cost Per Purchase on Meta and TikTok.
• Adjusting the audiences’ manual rules took constant effort, with mixed results.
• Autodoc tested AppsFlyer’s AI-driven Predictive Audiences against its manual lists – and won on ROAS, conversion rate, and Cost Per Purchase on both iOS and Android.

What is Predictive Audiences

Predictive Audiences shows which users are likely to reach a goal, such as making a purchase. Instead of marketers manually defining rigid rules, the model finds the right users on its own by analyzing patterns in historical conversion and behavioral data.
Learn how to create your first Predictive Audience.

Background

Autodoc is one of the biggest online sellers of car parts and accessories in Europe. Meta and TikTok are the major channels driving  Autodoc’s mobile growth. Both channels deliver  new customers and help win back old ones across markets. As its business grew, audience quality started to matter more as better audiences meant lower costs and more effective remarketing.

Challenge 

Like many marketing teams, Autodoc built its audiences manually based on rules from past user behavior. For example, they targeted users who bought recently or visited a product page.

The team’s main goal was to lower and stabilize Cost per Purchase (CPP/CPA). As Vadim Histsev, VP of Digital Marketing at Autodoc, said: “The challenge was mostly unstable CPA, but in general, it’s the constant experimentation approach that actually got us on board.”

Autodoc wanted to know if a predictive approach could beat its strategy, and joined the AppsFlyer Predictive Audiences beta.

Solution

AppsFlyer invited Autodoc to join the beta as a co-creation partner. “Our collaboration with the AppsFlyer team during the Predictive Audiences beta gave us a chance to test how AI performs in real-world acquisition scenarios.” said Vadim Histsev.

The team ran A/B tests across Meta and TikTok. They compared campaigns using Predictive Audiences against campaigns using manually-built audiences. The team used Predictive Audiences to automatically and rapidly identify and target users who might convert. 

Results 

Predictive Audiences improved Autodoc’s primary KPI, Cost per Purchase. It also delivered strong gains in ROAS and conversion rate.

Meta: 135% higher ROAS, 38% lower Cost per Purchase, 54% higher Conversion Rate TikTok: 57% higher ROAS, 18% lower Cost per Purchase, 19% higher Conversion Rate

The Predictive Audiences were about 34% smaller than the manual ones used in the test. That’s proof the results came from precision, not from reaching more people.

“Working closely with AppsFlyer on the Predictive Audiences rollout allowed us to bridge the gap between AI theory and practical performance. By shifting to more precise, AI-driven audience segments — roughly 34% smaller than our manual lists — we achieved lower CPAs and improved conversion rates across Facebook and TikTok. “ said Histsev.

Beyond the campaign metrics, Autodoc reported less manual work. The team labeled the beta as a proof of concept for better remarketing audiences in the future.

The company also plans to evolve its segmentation strategy, shifting from behavior-based lists to objective-based audiences. This change aligns audience selection with campaign goals.