Machine Learning methods traditionally assume well-behaved data with well-defined boundaries between predictors and labels being predicted. Unfortunately, in the real world, we have years of minimal, inconsistent data. Using such data as a basis for a machine learning model is not an ideal start to AI. This bias with data, however, can be addressed and overcome, resulting in a robust ML model. In this session, you'll learn from Salesforce Product Management about data leakage, or label leakage, challenges while building Salesforce Einstein and how to detect and fix them automatically.