Choosing a data scientist is hard. They come in lots of different flavors and they get hired up quickly. If you are a hiring manager, deciding who to interview and hire can be confusing. And if you aren’t a data scientist , it can feel like shooting in the dark.

The first trick is to stop looking for unicorns. They don’t exist. The master-of-all-trades, complete package data scientist is a myth. Searching for them is a distraction to you and your recruiters.

Instead, look for mutants. They totally exist. Each one has their own superpowers, like retractable Adamantium claws or the ability to walk through walls, but none of them can do everything. Here are some of the superpowers you can be on the lookout for. (For a similar list, check out Mango Solutions' Data Science Radar.)

Creating beautiful and informative pictures from your data.

Creating a system that takes what you already do and makes it run ten times faster or on a thousand times more data.

Creating a system that can take in your data in its raw form, process it, store it, use it and get the results to the program or end user that needs them.

Experiment design
Creating a plan to collect enough data of the right type to give you meaningful results.

Creating clean and useful data from dirty real world collections.

Creating a mathematical cartoon of your data.

Algorithm development
Creating new mathematical methods for using your data to answer your questions.

It’s helpful to think hard about exactly what you need on your data science team. Hire for the specific skills you need. You’re probably looking for two or three more than the others. If that’s the case, focus on your short list in your recruiting and interviewing. If you’re looking for someone to scale your current process from Terabytes to Petabytes (engineering), don’t require that she also know the convergence rate of a neural network (algorithm development).

If you’re looking for three or more of the items on the list, make more than one hire. Everyone knows that mutants do their best work in teams. Trying to find a one-size-fits-all data scientist may leave you disappointed and your new hire over-extended. Even if one person has the skills of three, he still only has 40 hours in his week (or 60, depending on your culture). A diverse team can divide and conquer, and they’re less likely to be taken out of commission by a single cold.

You may know that your business needs some data science muscle, but have no idea where to even start. Don’t despair. Your course is clear. You need to hire your Charles Xavier—a technically savvy leader. This woman or man will not be a unicorn. Others will code faster, have more publications and win more Kaggle competitions. Your candidate may even have a middling GPA, a low h-index and an unfamiliar alma mater. But your Professor X will have several superpowers of their own, and you’ll be able to evaluate these without having a technical background yourself.

Ask them to explain their research, teach you something or solve a simple problem that you know the answer to. If you have to work to follow their train of thought, they don’t have it. If their ideas effortlessly take shape in your head, simple and clear, they are communicators. When they’re done talking, if you feel like they’re smart, that’s not good enough. If you feel like you just got smarter, then they are communicators.

Here you are looking for examples of when they got themselves unstuck. Ask yourself “How long of a project could I leave this candidate to execute on their own?” and measure autonomy in hours, days and months. Judge it based on what they have built and work they have done, rather than “what if” questions. Ask them about projects they have worked on and teams they have led. Autonomy includes a fearlessness about learning new tools and skills and a willingness to try unusual approaches.

Whoever you interview, look for blind spots. No one knows what they don’t know. Ask them to show, rather than tell, you about their skills. Asking the candidate “Are you a good communicator?” is guaranteed to be less informative that asking the candidate to tell you about a project they worked on that failed.

I hope this gives you a jump start in hiring your next data scientist. Happy hunting.