On November 3, Venky Ravichandran, Zilliant director of customer success science sat on a panel with other big data rock stars to discuss best practices on the road to creating business models and processes that can unleash the power of predictive analytics. Moderated by Joshua Greenbaum of Enterprise Applications Consulting, the panel also featured Ben Coverston, DSE architect at DataStax and Dr. Kirk Borne of Booz Allen Hamilton. Below are some key takeaways from the discussion:
Ravichandran: Predictive models aren't a "black box," they're simply math! When leveraging these models, it's important to avoid jargon, but always have an explanatory model that provides context to the guidance.
Borne: A predictive model shouldn't be a Rube Goldberg machine. A 50-page equation is extremely brittle! Change one thing and it all breaks. Lesson here? Flexibility is key!
Greenbaum: What do we do about model failures? We depend on automated systems, what do we do when they fail us?
Coverston: That's an age-old problem, but we've given ourselves the tools to overcome it. We had the same issue with automobiles; they failed so we had to learn how to manually fix them. Now, we're working at a higher scale; robots build and diagnose cars.
Ravichandran: We used to analyze data using averages, however that was flawed because outliers skew averages. You can do beautiful things with data, yet if the model is incorrect the guidance will be wrong. The key is to have a flexible model that empowers repeat-ability with stakeholders. Then, you have the opportunity to recall and tweak the model as many times as is necessary.
Borne: Agreed! If you are working in averages, you are skewing your business! This is critical with artificial intelligence: Always think of the consequences. For example, NASA cites a successful mission as one where the astronauts are safe home with their families.