Machine learning (ML) models in production often start with a single objective, such as maximizing conversion rate in the payment industry. However, real-world business contexts are often more nuanced where other aspects relevant to a transaction, such as transaction cost or fraud risk, come into play. These objectives can be inherently conflicting: while optimizing for authorization may drive more revenue, it could also lead to higher costs or increased risk exposure.

Addressing such trade-offs necessitates the consideration of multi-objective optimization (MOO), while key information in the payment context plays a role in determining which objective should get more weight when considering the trade-off. In this talk we will share how we (Optimize ML team at Adyen) use the contextualized scalarization approach to improve our Intelligent Payment Routing product with a focus on conversion rate and transaction cost optimization.