For a major client in the construction sector, we were tasked with improving their Cost Per Action (CPA) campaigns. The objective was clear: generate form completions (leads) from real users interested in purchasing apartments, with a limited budget and a target CPA of $80 or lower.

The funnel was straightforward: buy impressions, convert them into sessions, encourage user interaction, and finally, convert sessions into completed forms. However, we needed to optimize the campaign to meet this CPA target, all while ensuring that the entire budget was spent efficiently.

As the Product Manager overseeing this test, I worked closely with the internal team to:

• Plan the budget distribution across various campaign stages.

• Configure line items based on different media types (mobile, desktop) and optimization strategies.

• Lead the manual optimization efforts to improve CPA results using internal proprietary solutions, balancing exploration and exploitation phases.

We implemented a strategy that split the budget into four key stages, with different allocations based on the learning and exploitation phases of the campaign:

• Learning Phase: Focused on exploring various traffic segments to identify which segments performed best.

• Exploitation Phase: Concentrated on maximizing conversions by directing budget towards the most promising traffic segments identified during the learning phase.

Throughout the test, we used internal optimization tools to manage traffic filtering and prioritization, ensuring that underperforming segments were identified and excluded, while the most promising segments were targeted for higher spending.

The test was a success in terms of the outlined goals:

• Budget Fully Utilized: We successfully spent the entire budget within the test period.

• CPA Goal Achieved: The CPA target of $80 or less was met on the exploitation line items, with the cost per conversion in some segments reaching as low as $55.97 for mobile conversions and $76.99 for desktop conversions.

• Improved Performance Over Time: As the test progressed, the performance of the line items steadily improved, confirming the effectiveness of the optimization tools and strategies.

Following the test, we identified several areas for improvement:

• Faster Learning: Better automation of optimization tools would accelerate the learning phase.

• Improved Exploration: More focus on expanding the whitelist strategy would enhance traffic exploration and identify new high-potential segments earlier.

• Native Ads: Native ad formats showed potential as an ideal match for performance-driven campaigns, suggesting they should be tested further in future iterations.