Search
  • Usama Fayyad

Data Science Career Talks - Usama Fayyad

Updated: Nov 14, 2019

We continue to reach notable personages of the Data Science world and talk about their careers, analytics & data science teams they manage and their advice for young professionals in this area. Hamit Hamutcu reached Nuria Oliver, PhD for an interview, who is the Director of Research in Data Science at Vodafone on the first episode.


The second guest is Dr. Usama Fayyad, who is also the Co-founder of IADSS, Co-founder and CTO of OODA Health, and Founder of Open Insights.


Please note that you can read the interview transcript here, or watch it directly at IADSS YouTube channel. Please follow our channel to stay up-to-date with latest interviews and news & insights from the IADSS Research.


Kate Strachnyi: Usama, please tell us a little bit more about yourself.


Usama Fayyad: Okay. I guess I've been dealing with data mining and data science for a lot of years. Started out with a PhD in AI machine-learning. Then seven years at NASA Jet Propulsion Lab, then worked for Microsoft for five years. Then I did my first two startups in the Seattle area, the second of which got acquired by Yahoo in 2004, where I became the world's first Chief Data Officer. Then I did a few more startups, did a stint running an accelerator and technology investment fund in The Middle East, out of Jordan, moved to Barclays in London for three years, came back to Open Insights in 2016, and have been also very involved in OODA Health, which is a healthcare technology startup.


Kate Strachnyi: How has popularization of the data scientist role changed the expectations of the ability and the skill-set of a data scientist?


Usama Fayyad: If you have a role which is highly paid, it's only natural that many people looking for jobs will try to reclassify themselves as, "Hey, I could do data science." In most cases, by the way, I don't blame them because it's so diffuse in the definition. They'll say, "Well, I'm a data scientist because I understand something about this domain, and have done some analyses, and can do some charting." In other area, the data science is much more specific and needs a lot of more technical knowledge. In some cases, it needs a lot of data engineering, etc.


What I think happened is we have now developed both an awareness and sensitivity to the importance of the role. Organizations are now basically coming up with better ways to evaluate how a data scientist is doing. By the way, one of the surprising results in our IADSS research survey, asking around this question is, "How do you evaluate data scientists and how are they evaluated?" There's many ways to evaluate through management, whether it's how long you've been in your role, and your seniority, versus how much value you're delivering to the business, etc. It's highly skewed. In data science, you're being measured by how much value you're actually delivering to the business. The two biggest factors are manager evaluation, not a very surprising thing, and the second one, which is even bigger is value to the business. I think that's very healthy because that is typically objective. It gives you the right incentive, and the value could be money. The value could be are you getting a better Net Promoter Score from your customers, are you servicing them better. It could be are people feeling better about the brand. There's many ways to do it.


Kate Strachnyi: Or seeding money to the company.


Usama Fayyad: Right. Any of these ways to do it, cost savings, more revenue, etc. So that, to me, is very healthy. All of these have driven the behaviors and tools that allow you to measure more systematically whether somebody is qualified or not. I think there is a danger there because if we don't define the proper standards, you end up defining it too narrowly. So it now becomes data scientists are only deep machine-learning, algorithm people. That's not true. That's why it's important to define the standards correctly so we define the right testing tools and the right approaches to evaluating.


Kate Strachnyi: A popular question that usually comes up is, what an aspiring data scientist should do first? Do they go back to school? Do they take the online courses? Do they just start going an internship and try to learn on the job? What are some advice to them?


Usama Fayyad: Yeah. My personal advice, there are several things you need to learn if you want to be a data scientist. Some of them are slightly technical, but not impossible. Nowadays, knowing Python or some programming language that helps you achieve tasks is very important. I would insist that you know some of the statistics and understand what these algorithms are actually doing. Personally, when I interview data scientists, one of the areas most candidates fail under is I'll talk to somebody, and they'll pull up an algorithm's name, and they'll say, "I applied it here and there." If I don't feel like they understand how this algorithm works and where it fails, then your knowledge is not that useful, because most of the game is about failure and how to avoid failure.


Actually, anybody can read up about an algorithm and run it from a program. The real problem is are you aware enough to understand where this thing can fall apart. When does the data become junk? As far as its concerned, what are its particular weaknesses? What kind of noise sensitivity does it have? So a fundamental understanding of the science of data, of what does it mean to build a model, which is really a summary of a data-set, and what are you trying to achieve with it, and how do you measure the health of the model that you generated. Those are all important aspects that you need to have a fundamental understanding. So little bit of statistics, some programming, and of course, familiarity with machine-learning. In the fields I work in, big data is very useful. Big data, again, doesn't mean volume. It actually means variety more than anything, meaning if I take any data-set today, and I add to it a professional graph, a social graph, and some documents, including videos, and images, and so forth, what do you do with it today? Well, for most databases, there's nothing they can do. That's why a lot of these data-sets today are becoming a lot less structured, a lot more unstructured, and therefore, having some knowledge of big data as a platform that allows you to manipulate these kinds of data-sets easily is also becoming necessary.



Help Set the Standards for Analytics & Data Science Professionals


IADSS is conducting a global scale research study. This study aims to gain insight into the analytics profession in the industry and help support the development of standards regarding analytics role definitions, required skills and career advancement paths. This will help set some industry standards which in turn could support the healthy growth of the analytics market.


If you want to be a part of this initiative and help collectively define industry standards, you can participate in the research. Just leave a note to us, which takes about 15 seconds, and you will receive an invitation later.


© 2020 by IADSS