Overview

Amazon Marketing Cloud (AMC) introduced a new capability called Custom Models for Audiences that lets advertisers run their own machine‑learning (ML) algorithms inside AMC’s secure environment. The feature, launched in early 2024, merges a brand’s first‑party data with Amazon’s proprietary shopping signals, giving sellers finer control over audience creation than the platform’s existing look‑alike tools. Sellers with mature advertising operations can now target shoppers with unprecedented precision and allocate spend more efficiently.

Key Points

• Custom algorithm import — Advertisers upload proprietary ML models to AMC, keeping the code and data inside their AWS account.
• First‑party + Amazon signal blend — Models can ingest a brand’s CRM or purchase history together with Amazon browsing, purchase, and category‑engagement data.
• Price‑sensitivity detection — The system can flag shoppers who are likely to chase discounts, helping sellers avoid unnecessary markdowns.
• Lifetime‑value forecasting — Predictive models estimate a shopper’s future spend, allowing advertisers to prioritize high‑value prospects.
• New‑to‑brand prediction — Algorithms identify shoppers most likely to try a brand for the first time, sharpening acquisition campaigns.
• AWS Clean‑Room security — All processing occurs within a protected AWS clean‑room, ensuring neither Amazon nor third parties see the proprietary model code or raw customer data.

How Custom Models for Audiences Works

1. Model preparation in the advertiser’s AWS environment — Data scientists train an ML model using a mix of internal datasets (e.g., loyalty‑program IDs, email‑open rates) and Amazon‑provided signal schemas. For example, a beauty brand might combine its subscriber list with Amazon’s “frequent fragrance buyer” signal to predict future purchase propensity.
2. Secure upload to AMC via API — The trained model package is transferred through AMC’s API endpoint. Because the transfer uses encrypted channels and lands in the advertiser’s dedicated AWS clean‑room, the model never leaves the brand’s infrastructure. A hypothetical sports‑nutrition company would see its churn‑prediction model appear in AMC without exposing the proprietary algorithm to Amazon staff.