Everything starts with inference about a business situation . It then moves to observation and experimentation. An finally if an case for impact can be built into ml production either through automation or detection