Uplifting conversion across the acquisition funnel with personalization using contextual bandits on AWS
Amazon Payments achieved single-digit conversion lift using contextual multi-armed bandits on SageMaker AI
“The problem turned out to be the content, not the model.”
Amazon Payments ran a seven-week A/B test using a multi-objective contextual multi-armed bandit on Amazon SageMaker AI to select personalized content for an acquisition funnel, achieving high single-digit percentage conversion lift for one customer segment. A second segment saw no improvement, revealing the bottleneck was content quality rather than the selection model. The post argues that as generative AI accelerates content variation proliferation, bandit-based selection will increasingly complement traditional A/B testing.