TL;DR (Too Long; Didn’t Read)

The paper addresses the following topics:

  • Addressed the problem of silo’ing of data.
  • Addressed how we could update / upgrade Machine Learning models on the fly to accomodate data drift.
  • Discussed how to develop a new kind of data scientist who would possess both technical and business skills.
  • Discussed how this new kind of data scientist would fit into and be aided by Agile pods.
  • This approach would be quicker than waiting for a Data Scientist to develop business skills or vice-versa.

Background

Thanks to our team winning the 2018 Deloitte ML Guild Hackathon, I was invited to be a part of a panel of authors with the goal of writing a Whitepaper. The authors on this whitepaper other than myself include: Tami Frankenfeld, Kyle Harbacek, Arunima Gupta, Abhishek Dugar, Sharad Kumar and Devin Cavagnaro.

The paper was published on Deloitte’s internal knowledge sharing platform, called KX. The paper was the most widely read whitepaper on the platform in 2018.

What is the Dataspace in ML?

We identified 3 challenges extant in the ML / DS space, which we call the “Dataspace”. The challenges are:

  • Data in an organization is not always centralized. It is often siloed in different departments, and even within departments, it is siloed in different teams.
  • When ML models are deployed, they are often not monitored for performance. Models might not be making predictions on the same data that they were trained on, due to “data drift” and the model itself influencing future data perhaps.
  • The people, process and technology changes in an organization that are required to implement ML are not always well understood.

How do we address this Dataspace?

While I cannot go into much detail due to confidentiality agreements, if you are a firm that wishes to contract Deloitte to help you with your ML journey, you can contact the authors of this paper who are still working at Deloitte.

My contribution was a section on the people aspect of the Dataspace. We came up with a career roadmap for people with business skills who are interested in ML, and vice-versa. This methodology was combined with Agile pods. With this combination, we were able to chart a course for people at any career level to develop both the required technical acumen and business skills to be successful in their chosen business domain.