5 Data Science Best Practices CDOs shared at DataX New York

By Melissa Todisco, November 25, 2019

A few weeks ago the Concentric team was at DataX New York. DataX continues to be one of our favorite events to attend due to the quality content and connections we always make there. Here is an overview of the highlights we heard:   

How to prioritize business needs for building an analytics process.

Hafton Eckholdt, Chief Data Officer & Chief Science Officer at Understood.org shared criteria for deciding which business questions to build an analytics process around. He suggested looking for areas where information can improve decision-making.

“Only certain business questions are worth prioritizing an analytics process around,” says Hafton Eckholdt Chief Data Officer & Chief Science Officer at Understood.org.

The obvious choice would be to focus on iterating and improving a model to help stakeholders who are already using data to make decisions – your happy internal customers. Eckhardt suggested that there is greater value to be gained by looking for people who aren’t happy with their analytics.  Eckhardt said, “The goal is a model for every customer. Find the unhappy, the difficult, and the unfamiliar. That is the right thing.”

Making the mindset shift from data projects to data products.

A panel discussion with Laura Hamilton, Vice President of Product at Rally Health introduced the idea of data as a product versus a support function that generates insights. She used Waze as a good example of a data product. It has found a way to deliver insights on how to help people get to their destination faster by avoiding traffic.  Thinking about how to use data to make someone’s experience better helps to package insights into something tangible that delivers business value. Taking this approach requires a business case with a long-term commitment as well as a shift from project management to product development. How can you rebrand your executive dashboards to become a data product? How can you create a better experience with a data product?

The importance of education in building a data science community.

Michael Berger, VP/Chief Data & Analytics Officer from Mount Sinai Hospital spoke about how he has built a Data Science Center of Excellence at Mount Sinai Hospital. One of his core areas of focus has been on education – not only educating the data science team, but also business executives. Part of his role is training executives how to ask questions. Executives often only think to ask questions that they have seen the answers to previously. By educating executives on the potential of what can be solved through analytics, it opens up the potential of where analytics innovation can take place.  Berger also shared his approach to staff development, which included doing an assessment of data skills and then negotiating bulk training services for key skills, such as R, Python, Tableau, statistics and storytelling. 

How to break through the data quality paradigm.

Hafton Eckholdt, Chief Data Officer & Chief Science Officer Understood.org, shared his perspective on the age old data quality paradigm. Eckholdt said, “The build it and they will come mentality will fail.” He suggested organizations need to focus on building out solutions and tackle the data quality issues in the context of the business problem. “It’s easy to abstract a business problem into a mathematical problem,” said Eckholdt. It’s only when you start touching the data in toy models and meeting the people consuming or generating data in some way that you can truly evaluate data quality and develop a plan to solve for it.

Choose explainability over accuracy.

Patrick Surry, Chief Data Scientist at Hopper and Eckhardt at Understood.org both spoke about the important of explainability.  When you deliver complicated ideas to people who don’t understand math, you are losing credibility and diminishing the value.  Surry said, “If we can’t make a connection to value, why are we doing it?”  Eckholdt suggested erring on the side of explainability even if that means using a slightly less accurate model.  Collaboration is needed in order to deliver value.  Slightly more accurate models that are too complex and can’t be explained limit those opportunities for collaboration. 

 

Our team enjoyed our time at DataX meeting and networking with some of the brightest minds in the industry all working to deliver on the promise of data science and machine learning:  value creation.

RELATED CONTENT