Table of Contents
What was your first job in the industry and what was the learning curve?
My first job was in BI consulting at OpenBI (now Inquidia). It can be a big leap. When I applied for that job, I had never heard of “business intelligence”. I had some coding experience, but I studied pure math (representation theory). I didn’t use a computer for a single result in my thesis. Steve Miller, the president of OpenBI, has a statistics degree from UW-Madison, where I was a PhD student. He liked my math background and took a chance that I could learn on the job. “Data warehouse” was a new concept to me at the time.
In that first job, I often worked remotely. I don’t recommend that for someone new to the workforce. Being in the office and having casual work conversations can be very motivating and reassuring. It’s hard to get the same feedback sitting on your couch. I struggled with time management at first. It helped me to set my own daily milestones and check in frequently with my manager. Since my time, Inquidia has set up a downtown office, and I hear the new hires really like it. They have a great team and attract top-notch talent.
What was your most memorable mistake?
I accidentally dropped the control table for our ETL application. Fortunately there was a tape backup, and the system was back online within 24 hours. I learned how important it is make sure that you’re covered from silly human error. That means having a backup and well-defined process in place to limit the opportunity for a developer to alter production in catastrophic ways. The safeguards are well-known, but it’s hard to overstate their importance: restrict access to production to admin roles, and use your own admin account only when you absolutely have to. Establish discipline around releases. And remember to test functionality at the scale you need! I made the mistake of running our standard QA on a deployment in our testing environment only to discover that it broke under the pressure of our production scale.
Data Science is composed of so many different disciplines, which has surprised you most as being particularly helpful?
It’s surprisingly useful to have a background in ETL. So much of what data scientists do is extract and clean data in order to feed our models. The models themselves are often well-known. Getting the data can be the hard part. The “plumbing” can be time-consuming and complex.
If I could redo my education, I would take more stats courses, especially related to Bayesian analysis. There are so many powerful Bayesian techniques for creating models and understanding model output. When you make your prior beliefs about the model explicit, you get not just a confidence interval for your prediction, but an expected distribution as well. Even when you are not sure what you believe, running the analysis with different prior distributions can help measure uncertainty, so you “know what you don’t know”.
What aspect of your work with data do colleagues misunderstand most?
There’s a tendency to mistrust or downplay what can be gained from predictive models based on big data. While I understand that skepticism, if you think about how particular decisions are being made in the absence of data, the bar is pretty low to improve those decisions. To win the trust of skeptical colleagues, it can be helpful to unpack your assumptions and techniques. If you’re using logistic regression, explain why. Explain what the output of the model means and what it doesn’t. Try to say in English what the equations are doing. Your business stakeholders may surprise you with their sophistication. At the same time, it’s important to put a disclaimer on predictive models. They can be incredibly powerful, but they’re not always going to be right, and they need to be supplemented by human intelligence.The bar is pretty low to improve decisions. #datascience Click To Tweet
What have you learned from observing data science being done wrong?
As data scientists, we always approach a problem with assumptions about how the underlying system should work. We need to make our a priori assumptions explicit and be careful about relying on them too much.
At Conversant, my work is focused on bidding optimization. My goal is to understand how much should we bid for a particular advertising opportunity in order to maximize the benefit to our clients and company. It would be easy to make assumptions about how the advertising market works. For example, I might assume that if I am willing to pay more, I should get better quality advertising and that it will be more effective. It could be dangerous to build an assumption like that into my model without rigorously testing whether the premise is sufficiently valid to meet the business objectives.
Who has been your most important mentor?
My first boss at Dotomi, Oded Benyo. At the time, Oded was the VP of Operations. Now he runs Conversant Europe. During my first week at Dotomi, he defined three problems, big issues that were important to the company. Then he gave me total ownership to figure out the solutions. When I needed support from management and other teams, he backed me up. His approach had a huge impact on my motivation and personal growth. He had a really high level of technical skill, but so much of his effort went to empowering the people who worked for him. With Oded, I saw for the first time that becoming a manager can be a way for someone who is a highly skilled contributor to have an even bigger impact. Now that I’m managing people, I want them to feel that same sense of ownership and excitement that I felt working for Oded. I never want my team members to feel I’m just coming to them with a list of tasks.
What advice would you share with a younger version of yourself?
A person’s ability is not a static thing, and you don’t just come with a set of pre-programmed abilities. Approach life with a growth mindset and identify areas of opportunity to improve. Over time, that’s incredibly powerful. When I started my career at OpenBI, the idea that I’d one day be managing a data science team would have been pretty farfetched. Learn, challenge yourself, and don’t be complacent. The people I admire most all share that quality.
Also, do not let yourself or the skills you cultivate be defined by the role you’re in today. In grad school, there’s a pretty narrowly defined set of things that make you a successful grad student. But I didn’t spend my entire career as a grad student, and the skills that ultimately became most important weren’t part of that job description.
Question for readers?
I don’t have a good answer myself, but I think a good question to ask is: if what you’re working on today didn’t exist and wasn’t paying your bills, would you find it necessary to create that thing?
This interview is part of Schuyler’s Data Disruptors series over on the StrongDM blog.
(image credit: Duncan Hull, CC2.0)