Leaving academia to start a data science company: Looking back at our first year

Just over a year ago, a couple friends from graduate school and I decided to leave academia post-PhD and launch Strong Analytics, a data science consulting and machine learning development firm.

Starting a business wasn’t new to me. My early interest in programming from the age of 11 had allowed me to start three software companies between high school and graduate school, leading them to acquisitions at 17-, 22-, and 27-years-old, respectively. I had licensed software to tens of thousands of clients worldwide, and worked as a software developer and consultant with hundreds of firms.

But this wasn’t like my other businesses. I had never founded a consulting-first firm, and I had never attempted to wed my academic and entrepreneurial pursuits into one cohesive plan. To that point, the two had proceeded on separate paths, with academia energizing me intellectually and my software businesses satisfying my incessant urge to always build things.

It’s been a challenging year, and one in which my partners and I have learned a lot. In part for a bit of self-reflection, and in part to share our experience with the growing number of academics entering the field we call ‘data science’, I thought it would be fun to recap some of the year’s biggest lessons.

Learning what ‘data science’ really means

My partners and I decided to become data scientists because, to put it simply, we loved statistics, programming, and building things that lots of people could find valuable. Our skill-sets fit well with one of the most common definitions of the term ‘data scientist’ — a programmer who know more statistics than average, and a statistician who knows more programming than average.

While cute, this definition doesn’t capture the true breadth of data science. For one, data science is expected to have an ROI — which means that data scientists need to have a strong business sense, solid financial fundamentals, familiarity with marketing speak, and a general ambition to turn ideas that are interesting/fun into profitable ventures.

Programming and statistics also fails to capture the full toolkit of the modern data scientist. I have had numerous graduate students tell me that, if they can’t get a great post-doc or faculty position after grad school, they will just become data scientists because, after all, they are pretty good with R and enjoy statistics. While there are good entry-level positions somewhere for these folks, in order to be a self-sufficient (‘full stack?’) data scientist, you need to know more than just how to analyze data using R or Python using statistical models taught in graduate school. Here’s a few additional things that we use regularly:

  • SQL (and not just SELECT…WHERE…)
  • relational databases (e.g., MySQL, Postgres)
  • non-relational databases (e.g., MongoDB, Cassandra)
  • machine learning models (e.g., random forests, gradient-boosted trees, neural networks, survival models, Bayesian inference, k-means clustering)
  • optimization techniques (e.g., gradient descent, genetic algorithms).
  • distributed computation (e.g., Spark, Presto, Hadoop)
  • cloud services (e.g., AWS S3, EC2, Lambda, API Gateway, Redshift)
  • interacting with APIs (e.g., OAuth, REST)
  • probabilistic programming languages (e.g., Stan, JAGS)
  • web programming languages (e.g., Python, Node.js, PHP, Ruby, Go)
  • UX design for app development and data visualization
  • scaleable software application design

Oh, and if you decide to run your own data science company, like us, you can add sales, business development, marketing, accounting, and recruiting to the mix!

I think that this sprawling toolkit is exactly what makes it difficult to pin down what data science really is. In the end, I’ve come to describe a data scientist as someone who both recognizes the opportunities in data and has the skills to capitalize on these opportunities end-to-end. As data scientists, this means we can be doing entirely different things one day to the next, and that’s one of the things I love most about this field.

Learning about the hype around data science

One of the other reasons we decided to found a data science company was that it seemed everybody was talking about data science for our last couple years in grad school. Every startup wanted to hire a data scientist. Every programmer wanted to pivot to data science in just 3 easy steps. Every academic fantasized about just joining Facebook or Google as a data scientist and leaving the frustrations of academia behind. And every business writer was saying that big data and data science were the next big thing.

From the outside, it was difficult to assess how much of the data science hype reflected baseless excitement and how much reflected true opportunity for young academics like ourselves. At the time, this made me a bit nervous about the decision to jump in with both feet.

Now, from the inside (or at least somewhere on the fringe), the hype around data science is more entertaining than nerve-wracking. To be sure, we felt the hype from day one. We received more job applications in our first 3 months of business than inquiries from new clients. And we still find ourselves competing for some projects with web developers overeagerly selling themselves as machine learning experts or big data consultants.

But we have also seen first-hand how companies of all sizes and in all industries are recognizing real opportunities to use data science tools to increase the value of their core offerings or make their operations more efficient. Although we initially started out working with tech startups and e-commerce companies, some of our favorite projects to date have been in industries further outside the tech bubble, such as health logistics and electrical engineering. More and more, we are recognizing that the promise of data science to these businesses is very real, and we are excited to be their partners in their first forays into this new world.

Learning to sell data science consulting

With my previous software businesses, I worked hard to build a no-touch sales process. People found my software, checked out its features, and decided to purchase it all their own. My time was spend building software and supporting customers, not trying to convert new leads.

Consulting is different. Not only are sales processes very slow and high-touch, typically spanning several email chains, phone calls, and video conference calls. One reason for this is simply that wrangling everyone for these meetings takes time, and our clients are often talking to several other firms about their projects while in talks with us. But more importantly, selling data science consulting services cannot be done by listing off features with a clear price and expected ROI.

Part of this ambiguity in selling data science consulting is because it’s new. Sure, some clients are well-educated in the space and come with clear demands (e.g., “We would like to use deep learning for image classification.”). These clients know what they need and have usually already calculated the financial return on this investment.

Other clients, however, only have a sense that they could be doing something better with their data than they do currently and want our help to discover what that is. In these cases, even after a discussion or two, understanding the final deliverable or ROI of a project can be really tough. The ‘science’ part of data science means that we can’t always promise awesome, impactful research findings or accurate machine learning models. There is no way around the fact that, in the end, the data are what the data are.

The problem is that, while scientists might approve of this rigorous approach to data science, it can be off-putting to clients. They want to know exactly what they are paying for and what they will get in return. So, what we do now is offer most new clients a ‘discovery’ phase where we dive into their data, advise on data platform upgrades, and assess the feasibility of the various opportunities they are considering (and some they are not) before they need to decide where to make their investments.

This gives our clients a couple of quick wins in terms of data strategy: they (1) learn what is possible with their data, and (2) get a fresh set of eyes on their current data pipelines, warehouses, and analytics strategy. This also allows us to pitch them clearer projects with well-understood milestones, deliverables, costs, and expected ROIs.

Learning to stay motivated and excited

As I mentioned at the outset, co-founding Strong was the first time in my life that I had wed my academia and entrepreneurial pursuits together. In high school, I was studying physics and English during the day, then building content management software and a web hosting company in the evening. In undergrad, I was studying psychology, sociology, and French, then going home and building a universal payment gateway API and an e-commerce web framework. And in grad school, I was studying early language and conceptual development, then working on a new digital marketing SaaS product.

Data science had provided an exciting opportunity for my academic interests in statistics and research to come together with my software and business interests. Initially, it was really energizing to narrow my focus and not feel the pull of two independent interests all of the time.

Nevertheless, midway through our first year, I began to feel burnt out. I couldn’t understand why. I wasn’t working any more than I had before (which is probably too much, but still there was no observable increase). Moreover, things were working out as well as I’d hoped for the new venture. So what was wrong?

What I ultimately realized was that uniting my academic and business interests under one venture had had an unintended consequence: it had left me with very little time entirely dedicated to learning something new for my own enjoyment. Pursuing a PhD had always meant that I carved out lots of time to sit down and read academic papers or to understand new statistical approaches, sometimes because it seemed applicable to my research but sometimes (and most enjoyably) just for fun.

In data science, I was learning new things every day, but rarely for simple, intellectual pleasure. Everything I learned was a means to an end, and everything I learned was expected to be profitable.

If I had a boss at the time, I might have complained about the lack of time for self-development — a shortsighted move by any organization that wants to train and retain great data scientists! But given that I was my own boss, I had the more challenging task of understanding how I could fix this for myself. The answer for me was creating time when I turned off notifications, relaxed with a coffee, and simply began to read about things that I had wanted to learn but couldn’t justify because they weren’t related to a client project. It was tougher than it seems because, there are always 10+ things to do when you run your own business. But reading about new tools for fun soon re-ignited my motivation to build new things — just for fun — and got me through the burnout.

Looking forward

I’m happy to say that, with only one year under our belts, Strong Analytics is growing month-over-month and engaged in data strategy, machine learning, data visualization, and analytics projects with clients in a diverse range of industries.

We still have lots to learn about the tools and business of data science. But looking back at what we have learned this past year, this is more exciting than anything else!

If you’re an ex-academic or soon-to-be ex-academic looking to begin your career in data science, drop me a note at brock@strong.io and let’s talk. We’re hiring :)