A/B testing is a tried and true method for deciding which business alternative is better. In our data driven world, they proliferate, and much attention has been given to most efficiently when to stop the worse alternative (which costs money). A new generation of algorithms have been developed and are widely deployed by leading data-driven companies. These “bandit algorithms” go beyond A/B testing and should be a key competency for data driven enterprises.
- The traditional settings for A/B testing in the online (marketing etc) and offline (clinical trials etc) world
- The technical pitfalls in A/B testing
- The motivation and technical solutions provided by “bandit algorithms”
- Real business examples where “bandit algorithms” are better
- A/B Testing ML models in production using Amazon SageMaker