You're never going to find a data scientist with that ad

Good data scientists are hard to find. Here are some common mistakes in recruiting them.


You're not having any luck hiring data scientists, are you? Unfortunately, it may be all your fault.

Part of the problem may be the type of data scientist you're seeking, something I've written about before. You may think you want someone that "speaks machine," when really you need someone that "speaks human."

But, as Gartner analyst Christi Eubanks suggests, you might actually have more fundamental problems than data science taxonomy. You might actually be recruiting the wrong person... for the wrong job... in the wrong place.

Other than that, you're golden!

Human vs. machine

When hiring data scientists, it helps to understand the audience for their output.

As former Google and Foursquare data scientist Michael Li writes, data scientists crunch data to yield analytics for machines or humans, but generally not both. "Unfortunately," Li says, "most hiring managers conflate the types of talent and temperament necessary for these roles."

A data scientist focused on machines will tend to need a much stronger "mathematical, statistical, and computational fluency" than her more people-minded peer. The latter needs to be able to tell stories with the data.

Sourcing the unicorn

But even if you're dialed into the right kind of data scientist, it's almost certain that you're aiming for the improbable, if not impossible, as Eubanks outlines.

"Admit it," she writes, "You and/or the recruiter took a bunch of LinkedIn posts and genetically engineered your perfect, beautiful quant baby with skills you didn't even know you needed."

And so the job spec asks for an MS or PhD (with an MBA, preferred!) that knows every business intelligence program ever invented, hacks NoSQL databases in her spare time, and also loves PowerPoint.

This leads Eubanks to conclude: "Data scientists are related to, but not the same as statisticians, who are not all versed in web analytics, which is not SEM, though all of the above can probably handle some SQL, and none would pick .ppt as their favorite medium. There are three or four different experts rolled into this one position, but none of them are going to apply for this job."

Far better, she argues, following her colleague Svetlana Sicular, to stop searching for unicorns and instead let someone with potential grow into the job. According to Sicular, the best data scientists will be those that know the right questions to ask. Learning the tools (like Hadoop) that help answer the questions is a secondary problem.

Rising to the challenge

A good data scientist wants to dig deep into a company's data, not muck around in entry-level, "data janitorial" work. As such, Eubanks calls out problems with job postings that advertise grunt work:

"'Not afraid to get your hands dirty' is always a red flag. 'Roll up your sleeves' is its ugly cousin. Experienced analytics people assume some level of janitorial data work, but they want to spend the majority of their time on tasks that leverage their interests and expertise."

The best place to discuss this with them--to really capture their interest--is not in an online job posting, of course. These people can pick and choose their jobs. If you're waiting for a data scientist to fall into your lap, someone else is already at their door, ensuring the best data scientists never see your Help Wanted ad.

"To land the most elusive analysts, you might have to meet them in their natural habitats... analytics conferences or industry associations are an obvious place to recruit."

So are online forums like Quora or online competitions run by Kaggle, she says. One additional benefit of finding data scientists in these places is that it's often possible to get a hint of the quality of their work.

Yes, all of these suggestions implies more work for the recruiter, but given the potential rewards, that extra effort is worth it.

Also see