We recently sat down with Mansha Mahtani, a data scientist at Instagram, to discuss her transition into the field, what a typical day looks like, what it’s like to be a woman in data science, and what advice she has for aspiring data professionals.
The full video Q&A is below, but here are some of the highlights:
What is data science?
The way I think about it is, it’s a way of making business decisions [by] aggregating data from different sources and really making sense of that data. So, what’s nice about data science is you actually have proof for the decisions that you plan to make as a business.
How did you get started in data science?
When I was in college, I wasn’t sure specifically what I wanted to do. I was studying CS, I was studying engineering, but I knew I didn’t want to be a software engineer. I wanted to solve business problems, which is why I moved to consulting and specifically tech consulting. When I was in consulting, I realized that I was too far removed from the actual business decisions. So, we’d be giving recommendations and then the company would execute it and you kind of leave and you’re not really part of that process, which is why I decided I wanted to move to a smaller company, so I could see how all the nuts and bolts were working.
I specifically was looking for a consumer company because I was really excited to understand how people behaved and how people made purchasing decisions. I reached out to my network and ended up being hired for an analyst role at Blue Apron and it was a very not well-defined role. I was just basically doing anything and everything—and that’s essentially what startup is—but eventually got enough of an understanding of statistics and data analysis and data science. Once I was at Blue Apron, I decided that I wanted to understand how bigger companies worked and really understand how a social network ended up making product decisions, and that’s why I decided to move out here to the Bay Area and join Instagram.
What is a typical work day like?
I would say I would split my time into three equal parts. One would be doing the analysis—coming up with insights. The second part, which is pretty important, is just the storytelling, the communication—so how do I actually get people excited about this, what is that narrative I’m building? And then the third would be really communicating with other people—evangelizing my thoughts. The third part is probably one of the most important parts of a data scientist’s journey, just because the value of a data scientist is really only seen if the business decision is actually implemented. And the third part is crucial in actually getting your idea from just an idea to an actual implication and an actual product.
What’s your favorite and least favorite thing to do as a data scientist?
The least favorite thing would probably be just doing all the data wrangling and data cleaning that comes up with data science. I think it’s part of the job you have to data wrangle a bunch just to make sure that the data is right, but it gets tedious and you never really know if the data is reliable, so that can be kind of hard.
Most favorite part of the profession? I would say that it’s just incredibly fascinating to me that you can actually predict the future with numbers and I find that just mind-boggling—just the fact that you can look at patterns, really understand what people are doing and then decide: this is what I’m gonna launch and actually have strong confidence that it’s gonna work.
What’s a cool project that you’ve worked on?
One interesting problem I solved when I was in my last role was: we were trying to figure out if a certain promotion in a physical store was going to cannibalize our product. In a software product, this would be pretty easy to test. You would run an A/B test and test: is the test better than the control or is the control better than the test in terms of sales? However, in the real world you cannot do that—you cannot run an A/B test. So what ended up happening was I had to actually come up with a simulated control group to compare against the sales of the promotion. And what was really hard about that was not just the fact that technically it was really hard to do this—I used R and a bunch of matching algorithms to actually match test vs. control—but the fact that I had to come up with really interesting features to match people, and I think there was more of an art to that than a science.
Eventually what happened was we realized that the sales did not cannibalize the product… we launched this specific promotion across the country and this pilot essentially became one really huge revenue driver for us.
What’s one essential tool that you can’t live without?
Stack Overflow is probably my most essential tool/website that I’ve ever used. I feel like it just houses every single answer that you need, especially if you’re in a small company where you don’t have much guidance. So, in my last company, for example, I was the first analyst they hired. I didn’t have too many people to really speak to and as a result I had to rely on external resources like Stack Overflow. So, it’s a really great tool when you’re learning something new and I recommend actually looking at Stack Overflow first before even asking people around you because you get used to learning on your own.
Related: 24 Free Data Science Tools
What kinds of people typically become data scientists and what advice do you have for them?
Data science is both an art and a science, and I do think that people tend to fall in a spectrum, so some people tend to be really good at the communication, really good at the storytelling and the product intuition part of things, and then there’s some people that really prefer working on optimization problems, really prefer building those statistical models and getting into the math of it.
My recommendation would be: figure out where you land and then make sure that you develop some sort of way to communicate that when you’re looking for a role or a job. When you do that, you’ll really stand out amongst your peers because every employer knows that there is a spectrum and it would just be helpful if you were able to communicate your strengths and play those up.
What types of things do employers look for when hiring?
Employers, I think, are looking for maybe two or three main things. One, they’re trying to figure out: do you have the right product intuition or the right understanding of what the problem the end-user is trying to solve is? That’s the first thing. The second thing that they’re looking at is: can you communicate those insights clearly? And then third: do you have the technical ability to analyze data? And I think with the technical ability, a lot of it is just potential. Just because you don’t know a certain skill now doesn’t mean you aren’t able to learn it very fast. In fact, when I was a data scientist at a startup, I actually didn’t know SQL at all. I learned it on the job and that was because I was able to pick it up relatively fast. Of course, I did have a computer science background, which definitely did help, but there was no requirement for me to know a specific language. And I think if you can show that potential, you’ll go really far.
Regarding questions, an example question would be: should we launch feature X, for example, and if so, how do you know feature X is successful? That’s generally a very typical question you’ll hear. And people will typically answer, “Once you launch it, measure KPI 1, KPI 2, and these other KPIs, and then launch an A/B test and see how well the feature is doing from those metrics.” And although that’s a good answer, it really doesn’t show the fact that you’re a product thinker. To take it the next step, what you should really be doing is understanding why is that feature being launched in the first place, what is the job that that feature is going to solve for the end user and then come up with actual metrics that would solve that job—and then talk about the metrics. So, going back to the why is really important with answering questions like that.
What should a candidate bring to a job interview?
When going to a job interview, some companies expect you to do a take-home exam, so you’d be showcasing a lot of your work there. You can also send in samples of some of your work. My key advice for those sort of projects is: make sure that the problem that you’re solving is important. A lot of times, people solve problems that are interesting, which is great, but unless you can actually showcase that what you’re solving is important, and there’s a true opportunity, it won’t seem as important for employers because at the end of the day your time will be limited in the job and an employer wants to understand: are you able to prioritize the most important items?
What’s it like to be a woman in data science?
One thing I found with data science is that there’s actually more representation of women than in other engineering professions and I truly believe, if I can be so bold to say this, women tend to have a lot of empathy and empathy is important for the role—not just for the end users, but also you need to collaborate with your cross-functional partners and really come to a decision with them. I think that’s where women tend to excel in the profession.
However, another part of data science is evangelizing your findings and really advocating for your work and women tend to be a little bit worse at that as compared to their male counterparts. I think that’s something even I’m trying to work on: really making my work more visible. Because a big part of the job is actually coming to a decision and until you evangelize you’ll actually never get there.
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.