We recently interviewed Levi’s Director of Data Science, Pallav Agrawal, to discuss how he hires DS candidates and the 5 mindsets aspiring Data Scientists need to master to get a job in the field. Here are the key takeaways.
We’ve heard that the data science team at Levi’s is big on puns. What are some of your favorites?
We have quite a pun-tastic group of data scientists that Levi’s. So one of the first puns we came up with, our thought was to name our tech blog Ma-Jean Learning. Another one that is actually in production today is this project where we look at the different attributes that define the look of jeans.
- What kind of wash does it have?
- What kind of damage does it have, what’s the leg flare or what’s the rise of the jean for different types of jeans, and how those attributes can then be combined to create a final look?
And we call that Project Jeanome — just like the way the genome often defines the structure of an entity.
You believe there are 5 principles – or mindsets – that Data Scientists should master. What are they?
- Develop pithy, actionable insights that improve decision quality
- Have a good BS Filter – Practice healthy skepticism towards most remarkable claims and cautious optimism towards the applicability of positive results
- Think like a lawyer – Define the business problem in unequivocal and quantifiable terms
- Aim to fulfill user needs in a financially-viable manner
- Learn to communicate and collaborate within interdisciplinary teams – it’s as essential to the success of a data scientist as the knowledge of programming
Principle 1: Develop pithy, actionable insights that improve decision quality
The core value proposition for data scientists is enabling better decisions. Whether they are made through an algorithm or through something that’s delivered to a human who then makes a decision, the core value proposition is that through our work, we will help our company make better decisions across the board.
It’s really important for data scientists to create good decision aids for decision-makers whenever possible by leveraging data in a very scientific and rigorous way. It’s a lot like uncovering the laws of nature as to how do people behave, how do certain quantities fluctuate when other quantities fluctuate and bringing them all together so that the final decision maker — whether that’s an algorithm or a human being — has the right tools to make the right call at the right time.
Principle 2: Have a good BS Filter
I always check whether the candidate has a good BS filter — BS here could stand for bad science or biased statistics or more conventional definitions of the term.
It’s important to practice healthy skepticism towards the most remarkable claims and cautious optimism towards the applicability of positive results.
Data science is going through an interesting time right now where we have this perfect storm brewing in many ways. On one hand you have companies who are claiming at unprecedented rates that they’re using AI and machine learning to make their businesses more efficient. On the other hand, you have startups who just by adding .ai in their domain name stand to raise 3.5x more money than companies that don’t. And so it’s this perfect storm in some ways. It can often be tricky to navigate the entire spectrum of companies out there.
And if that weren’t enough, Cassie Kozyrkov said in a blog once that data analysis is the ultimate Rorschach test – that humans are so prone to apophenia, where we will find patterns and randomness that we can often find meaning in data that doesn’t actually have any meaning to it. And it’s important to not fool ourselves and understand that often times, results that look too good to be true are not true.
So it’s important to keep ourselves honest and to have a good BS filter. Richard Feynman once famously said that, “The first principle is to not fool yourself because you’re the easiest person to fool.” And I think as data scientists, we are so excited to bring our work to the world that we sometimes don’t pass it through some of the sanity filters that we should.
Principle 3: Think like a lawyer – define the business problem in unequivocal and quantifiable terms
Having the ability to think like a lawyer is crucial for data science. You should be able to take apart a statement and remove all the subjective components and really keep the objective core so that it is tangible and measurable. And the way it applies to the work that we do as data scientists are in the fact that a lot of times business stakeholders will come to us with a fairly simple ask, which to a human being makes a lot of sense.
But translating that to a machine can be challenging.
How do you explain things like we want to make our consumers more loyal or we want to detect fraud to a machine? What lawyers are great at doing is using very precise language to cut out the boundary of the playing field because otherwise the problem statements can become too challenging and too broad for us to solve in a meaningful way. The disconnect occurs when the business wants to solve one problem while data scientists are solving a different interpretation of that problem. It’s important for both parties to be super aligned around what is the exact definition of the problem that they’re trying to solve. There’s this famous quote which is, “A problem well defined is a problem half solved.” So I think spending more time upfront in defining the problem before jumping directly on the solution is important is an important mindset.
Principle 4: Keep a keen eye on consumer experience
As data scientists, we often tend to be so enamored by technology that we often forget that there are many times there’s a human who is being affected by the outcome of our algorithms. There are stories around algorithms being trained for some objective which is not aligned with what the users necessarily want. And so I think it’s very important to have a keen eye on what kind of consumer experience is being enabled to the application of this algorithm.
Principle 5: Data Science is a team sport
Data scientists can never be truly effective if they work in a silo. A lot of data scientists I meet come from academia and research labs, or they used to be Kaggle Grandmasters where they were often a team of one and they operate as such. When they come to work in companies, they often realize that they’re working on just one small part of the overall problem.
And so the analogy I like to use there is that they might be working on the car’s engine, but for the car to actually go to market, there are a lot of other components to it — the transmission, the seatbelts, the audiovisual system, things of that nature. It’s important for data scientists to be able to speak the language of their interdisciplinary partners from operations to finance to dev-ops to product management and project management. Data Scientists need to understand how their work can fit in to solve a larger problem — so having a team mindset is very important in this role.
Watch Pallav’s Real Talk Here:
Ready to start or grow your data science career? Check out our Data Science Career Track —you’ll learn the skills and get the personalized guidance you need to land the job you want.