Machine learning for commerce media: The strategic separation of Organic and Ads ML

Commerce media, the advertising within e-commerce platforms, depends on two distinct machine learning (ML) types: Organic ML and Ads ML, each with different use cases and requirements.
Organic ML generates predictions for product ranking, such as predicted click-through rates (pCTR), to improve user engagement metrics like clicks.
Changes to Organic ML are typically assessed through simple A/B experiments by splitting traffic.
In contrast, Ads ML predictions are only one of several factors in ad ranking, often combined with bids to calculate expected revenue.
Inaccuracies in Ads ML predictions can lead to suboptimal ad placement and revenue loss, especially in second-price auctions.
Additionally, Ads ML experimentation is more complex because it must integrate with bidding and budgeting systems, requiring advanced capabilities to separate feedback loops.
Big tech companies like Google, Facebook, and Amazon separate Organic and Ads ML models to allow teams to optimize independently.
This separation prevents changes beneficial for organic experience from harming ads performance and vice versa.
While using Organic ML as a foundation for ads is a good starting point, recognizing the need for separation early helps in resource planning and platform selection.