How to Build Disruptive Data Science Teams: 10 Best Practices
Building a data science team from the ground up isn't easy. This strategic roadmap will help hiring managers with tactical advice and how to properly support a data science team once established.
By Praful Krishna, founder of Coseer
In 2012, an HBR article named Data Scientist the sexiest job of the 21st century. The same article defines the role as a “hybrid of data hacker, analyst, communicator, and trusted adviser. The combination is extremely powerful—and rare.” These rare birds are in high demand. The August 2018 LinkedIn workforce report found that there were 151,000 data science openings open, and the US Department of Labor and Statistics forecasts double-digit job growth through 2026. That’s on fire compared to the average of 6% growth.
We’ve spent the last seven years working directly with data scientists at Fortune 500s and government agencies. We can attest to the HBR definition of data scientist - it’s a little bit of everything. Data scientists take a concept (big data) and turn it into something valuable and concrete. The role requires someone uniquely tech-savvy and creative.
If this explanation seems vague, you’re not alone. The value proposition of data scientists is clear; you have data, likely in an unfriendly format, and you need someone to help make use of it to solve business challenges. But with little or no experience hiring this new breed of knowledge worker, how do you get started putting together a data science team? We’ve identified a few key trends over the years.
1) Know who to hire.
To add to the general confusion, "Data scientist" has ballooned to encompass a huge variety of job descriptions. There are actually four distinct roles within data science, usually held by four very different people.
- Data Theorist: Depending on your application, the highest value-add comes from people with the technical skill to sift through raw data and the creativity and intuition to identify patterns and trends with the right context. There is an art as well as science at work here. Here at Coseer, these "jack-of-all-trades" types have written a completely new algorithm called Calibrated Quantum Mesh that is better for natural language applications than deep learning.
- Data Engineers: Next comes the people who take these algorithms and design models to find solutions. At companies like Google, Netflix, and others, such data engineers are using deep learning to solve some amazing challenges.
- Data Modelers: This is probably the group you're thinking of when you hear "data scientist." They take existing models and apply relevant data to make business decisions. For example, Convolutional Neural Networks work best for images. Even with such a clear path forward, they must problem-solve constantly, such as figuring out which set of images are best for which category of problems.
- Data Analysts: Finally, we come to the data scientists whose primary role is to hunt for and clean data so that more advanced analyses can be conducted. At first glance, this may sound like busywork, but it's far from it. Talented data processors can add a lot of value here. This is also the most significant segment of data science roles in terms of person hours.
2) Don’t rely on recruiters.
If you, a hiring manager, are having a hard time understanding data science, imagine what a recruiter is going through. You may be out of your element, but you have a firm grasp of the business challenge you’re hiring for, and you know your organization. You’ve done your research and read through the different roles data science encompasses in #1 above. You’re more capable of making data science decisions that affect your sphere than an HR pro.
3) Be specific, be flexible.
It’s good practice to be specific about your wants, but open to suggestions about your needs. Given the confusion about what makes a good data scientist, the more specific you are on what your team needs, the more transparent you’ll be.
At the same time, keep in mind that data scientists tend to be comfortable with uncertainty. They’re used to management not quite understanding them. If you signal that you’re open-minded, they’ll be happy to bring out their creative side with you. You wouldn’t be hiring a data scientist if you had all the answers. Casting a wide net and keeping an open mind can open the door to amazing solutions you would never have thought of.
4) Don’t isolate data scientists.
The same HBR article that named Data Science the sexiest job of the 21st century found that data scientists fundamentally want to build. They want to be “on the bridge” - at the center of the action, observing and responding to evolving situations, and affecting real change.
They can’t do this in isolation. They need experts in the right areas to add color to the data. Context is key in separating actionable insight from noise. Make sure to facilitate free-flowing communication between data scientists and product management, business development, engineering - especially customer-facing people. You’ll get better solutions and happier data scientists.
5) Be open about your tech stack.
Remember when we told you to be open-minded about your needs? Be just as open about your tech. Yes, AI is an amazing tool. It’s transforming everything about the enterprise. But it’s not the solution for every problem. Less glamorous techniques like linear regression will get you there faster much of the time, and many big data problems can be solved without advanced AI.
If you do decide to go the AI route, remember that there is more to AI than just deep learning. For example, if your data is unstructured (e.g., emails, scientific papers, images, or audio), or if you need results quickly without a lot of training time or expense, an NLS (Natural Language Search) solution might be a better fit for you.
Your data scientist will be trained to identify the best solutions, so understand that “cool” algorithms probably won’t be doing the heavy lifting.
6) Run world-class interviews.
There's no "checklist" for data science hiring, but that's not all bad. In fact, it allows for a lot of creativity. We've found that by asking open-ended questions based on real-world data problems, we're able to open up a conversation that allows us to see a data scientist's thought process in action. This is essential in sizing up how he/she will tackle your business challenges.
Here’s a list of questions we ask every data science candidate at Coseer.
7) Manage expectations.
This is data science, not magic. Even with the most advanced training and tools, your newly assembled team will need time to learn your systems, your data structure, and your strategy. Your process probably isn’t as robust as you think. Projects may get off to a slow start.
Hear your data scientists’ out when they ask for new software/hardware and give them time to clean and organize data before diving into any huge projects. Remember Rome wasn’t built in a day - the quality data you’ll get at the end of this adjustment period will pay for itself a million times over.
8) Strive for accuracy.
Familiarize yourself with the virtuous cycle of AI. A great product attracts users, who add data to the engine with every use, which in turn improves the product. No one will use a tool which doesn’t produce results. Ever wonder why your company’s intranet search sucks so much? Because it wasn’t designed well from the start, which discourages use, and because no one uses it, it never gets any better.
Bottom-line pressures can be intense. We know you’re excited to see results. Resist the urge to put tight deadlines on your new data science team. You don’t want them pushing a half-baked product that no one will use. This is a waste of everyone’s time.
9) Keep hiring trends in mind.
By now, you’ve assembled a rock star team, but the field is evolving fast. Data science encompasses so many skills that you’ll likely have a team with very different expertise.
Data scientists are in high demand, yet alarmingly unhappy in their roles. Commonly cited issues are misaligned expectations, “office politics,” and isolation. You can avoid this by facilitating clear communication, assigning cross-functional projects, and giving data scientists space to experiment and follow their creative hunches. You’ve chosen these people for their tech-savvy and creativity; don’t let it go to waste. You’ll end up with a happier team, and, go-getters that they are, they’ll keep up on all the amazing developments happening that may make all the difference for your business. It’s a win-win.
10) Supplement with the right partnerships.
You can’t do everything by yourself. You’ve hired a formidable new team to do the heavy lifting, but you’ll still run into situations where you need to bring in outside help.
Working strategically with vendors gives you a degree of flexibility that’s hard to get elsewhere. You’re going to run into unexpected issues, and vendors will provide you with bandwidth where you need it - until you need to move on to the next frontier. Don’t be averse to working with specialists who’ve been solving unique challenges for years.
Much of the advice you’ll run into when hiring data scientists make the process sound like trying to catch a few fish in a small pond - while competing with 1000 other fisherman. Understanding data science and the capable people in the field will give you more than a leg up.
In order to attract A+ people and keep them happy in your company, you need to have the right foundation - a culture with a basic understanding of their field and what they can bring to the table, a realistic and challenging set of problems for them to solve, and a support structure that can offer direction and context. The whole world is available to top talent, so retaining such driven and capable people is a matter of presenting your most interesting business challenges in a way that speaks to them, and offering them room to be inventive in solving problems.
As Thomas Davenport and D.J. Patil so eloquently sum up the growing need for data science in enterprise, “think of big data as an epic wave gathering now, starting to crest. If you want to catch it, you need people who can surf.”
Bio: Praful Krishna is an AI/Cognitive Computing expert who helps Fortune 500s automate tedious text-based tasks. He founded Coseer after a career with a hedge fund and a turnaround CEO role. Praful is an alumnus of McKinsey, Harvard, and IIT. He is also a hapless dad living in San Francisco. Connect with him on Quora or on Linkedin.
- How to Make a Success Story of your Data Science Team
- On Building Effective Data Science Teams
- A Winning Game Plan For Building Your Data Science Team
Top Stories Past 30 Days