BY Meghan MalasSeptember 02, 2022, 2:15 PM
Chetan Sharma was among the first data scientists at Airbnb. He’s now the founder and CEO of Eppo.
As data-driven strategies became more and more focal to businesses in the past decade, Big Tech companies’ demand for data scientists has sky-rocketed. And while data science professionals can land six-figure jobs at some of the largest companies in the world, the world of startups is also full of opportunities for data scientists.
Silicon Valley, where Big Tech giants like Apple, Google, and Meta are headquartered, is also home to many startups. Across this spectrum of companies, there’s a need for people to make sense of their data—and more than 3,500 data scientists work in the San Jose, Sunnyvale, and Santa Clara metropolitan area, earning an average salary of more than $157,000. The San Francisco, Oakland, and Hayward metro area is home to more than 5,200 employed data scientists who make an average annual salary of about $153,180.
In 2012, Airbnb was one such data science-hungry startup in Silicon Valley, and Chetan Sharma was among the first people to help meet this demand. He was one of the first data scientists hired at Airbnb—the fourth member of the team, to be exact. Sharma since left to work at other startups and found his company Eppo, a product experimentation platform, where he is also the CEO.
By the time Sharma left Airbnb in 2017, the data team had grown tremendously—he estimates there were about 100 employees at the end of his stint. Before landing the job at Airbnb, he earned a bachelor’s degree in electrical engineering and a master’s degree in statistics from Stanford University. Sharma worked in health care policy before beginning his role at Airbnb, marking the beginning of his career in data science.
Fortune spoke with Sharma about what startups are seeking in their data science teams, and how data science graduates can set themselves apart.
The following interview has been edited for brevity and clarity.
Experimentation is key to becoming a successful data scientist
Fortune: How does a data science team operate at a startup?
Sharma: In contrast with companies like Facebook or Amazon, the founders of startups are not typically data people. Startup leaders tend to lean into intuition and celebration, idolizing a Steve Jobs or Walt Disney approach—without heavy consideration of data and metrics. So we had to win a culture war in addition to setting up infrastructure, and we did that through experimentation.
The defining traits of the first 10 data scientists at Airbnb, were they were very product-oriented, great communicators, and able to be autonomous. Data scientists have to introduce fresh new ways of understanding business. When we were hiring at Airbnb, there was this kind of arbitrage—which was that there were a whole bunch of really smart Ph.D.s who didn’t have clear options in the job market, especially tech companies, so we hired all these physics Ph.D.s and social science Ph.D.s. What these people had in common is that they were credible thinkers and able to self-operate.
Experimentation is a really core part of that toolkit. Machine learning gets so much attention when like 2% of data science teams actually do any machine learning but a very large share use experimentation.
As a data scientist at a startup, you are literally going to be teaching people a new way of thinking to guide consequential decisions for company strategy, careers, and promotions. There’s a lot of trust that goes into a metrics-oriented point of view, so we really need people who can communicate to instill that trust. These are people who can take complicated topics and make them understandable and compelling to the audience.
3 tracks for data science hires
Fortune: What are startups looking for in their data scientists?
Sharma: By and large, there are three types of data science tracks for those looking to get hired.
The first one is a data engineer, these people are defined by their ability to basically build reliable pipelines to model data. So if a company is early in its foundation—or at really any scale—then they usually prioritize building up reliable reporting. Data engineers don’t actually need great soft skills to do that part of the job, and this track is very employable because everyone needs reliable infrastructure.
The second track is analytics. So this is someone who would look for opportunities in this funnel. There’s a lot of taking metrics and slicing them by segmentations. building these reports, and presenting them to people. The analytics data science role is primarily a role of persuasion and you are driving decisions, so communication skills matter a lot. This track is the most competitive, and I would argue that some analysts drive the most impact of anyone on the data team. However, you can come into analytics from pretty much anything, and you do not need as much technical expertise.
The third track is what I call inference. This role tries to interpret and explain what’s going on in the data in a way that’s more rigorous than a bar chart, and experimentation falls into this world. These data scientists run statistical tests on top of data sets to see if it’s just random noise or if there are useful inferences. A lot of times people in this track use econometrics, so when you see an economist going into tech companies, it tends to be this inference track.
If I was the hiring manager of those teams, I would recommend against specialization in one of these three tracks early on. I think your first 10 to 15 data team members need to be what I call “full-stack data scientists.” These are data scientists who can perform the tasks associated with all three of these tracks: data engineering, analytics, and inference.
Without a doubt, the track that will get you a job the fastest is the inference track. It’s very specialized knowledge, so there are just not that many people with the skill set. So, the inference track is most employable if you have technical data skills, and you also have kind of an economist imprint skill set.
You need to be familiar with R and Python toolkits, and you definitely need to know SQL very well. You don’t really learn SQL much in college, but if you are even two or three years into your experience, you’re going to need to be an expert in SQL.
How to navigate data science careers at startups
Fortune: What types of startups should entry-level data scientists work for?
Sharma: Data scientists who are early in their career should try to work at a company that is highly leveraged by data. And while it’s obviously very hard to figure out what is and what is not leveraged by data from the outside, one thing to know is that many consumer businesses that are post-product market fit have enough data available to make meaningful conclusions and have an impact.
Some companies are focused on developing completely emergent technology. In those cases, it is going to be hard to contribute as a data person because for something that never existed before, you don’t have a history to pull data from.
You will also want to make sure you understand the company’s strategy. For instance, at Airbnb, during most of the years I was there, I was very focused on growth—I was not focused on revenue. If we had chased after a revenue goal, then we would have focused on only America and Western Europe, where people are wealthier. But since we focused on growth, we cared about the number of bookings, not how cheap or expensive they were. That was important to know because once you know the strategy and your key metrics, that is something you can use to experiment with the products and marketing, and it becomes a very data-driven process.
In a recession, every piece of a company is gonna get scrutinized for revenue ROI. I think for the past five to 10 years, data teams have not really had to answer what exactly is the marginal value of an additional data scientist. If you’re a person fresh out of college, this can be problematic. If you get put on a team that doesn’t have this clear ROI equation because the data leader never had to pass that scrutiny—that’s ultimately detrimental to your career.
A certain amount of data team members can be justified as essential infrastructure because the board needs numbers to understand how the business is doing—that usually takes three or four data scientists. So, larger data teams will have to very visibly and tangibly show how their work affects decision-making.
See how the schools you’re considering fared in Fortune’s rankings of the best master’s degree programs in data science (in-person and online), nursing, computer science, cybersecurity, psychology, public health, and business analytics, as well as the doctorate in education programs MBA programs (part-time, executive, full-time, and online).