We've all seen impressive examples of what machine learning and statistics can do when applied by experts. But how can we democratize the techniques used by researchers and turn them into better knowledge and decisions for everyone?

In this talk I outline principles for data science magic: how we can scale our techniques up from usage by a few experts by building services that everyone in an organization can apply to generate dramatically more value. I'll walk through examples of magical data science technologies from inside Facebook and from the statistics and machine learning communities.