I honestly feel for recruiters who are tasked with filing data-science and machine-learning job openings. The list of requirements that employers draw up for those roles is pure bravado with a side of madness: “10 years of data science with at least five years in natural-language processing and either a Master’s or PhD” (never mind that I can count on one hand the number of data scientists who were building for production back in 2007). Others ask for experience with three different programming languages, 10 platforms, a niche algorithm set, leadership skills—and by this point I’m typically only halfway through reading the job qualifications.
Ask any tech recruiter and they’ll tell you about the stack of job openings like these that they’ve been unable to fill for the past six months to a year. Every couple of weeks, the client calls and berates them for not being able to send them quality candidates. After awhile everyone involved throws up their hands and calls it a “skills gap.” It isn’t.
Scrap Your Stupid Job Qualifications
Google doesn’t require a PhD to be a machine-learning engineer. A recent survey found that only one in four data scientists has a PhD. Yet I still see advanced-degree requirements on the vast majority of data-science and machine-learning job descriptions. Most companies just throw it in unthinkingly. But unless they’re investing heavily in advanced research, it’s pointless.
Requiring a set number of years of experience is equally stupid. Forget years and start thinking in terms of problem-solving abilities. I love formulas, so here’s mine for hiring a great data scientist:
Platforms + Business Problems = Required Skills
There are no years of experience in the equation. Has the candidate solved the business problem on the same or a similar platform before? Great! Data scientists are used to working with uncertainty. We’re used to turning business problems into technical solutions. Tell us your problems, show us your platforms, and take us to your data. We’ll outline a roadmap in the job interview that leads to a solution. If you like it, hire us. It really is that simple.
Both these mistakes, and several of the other common ones, come from the myth that every data-science team member needs to be some kind of multitalented wunderkind. Most businesses just need one person with the rare trio of strategy, engineering, and mathematical modeling chops. The rest of the team is built around this person and supports their work. So setting more sensible job requirements for those roles can actually help you build the right team structure. When every member isn’t expected to do everything, hiring gets a lot easier.
There Are Thousands Of You And One Of Me
When job candidates start sounding like Kanye, you know demand is high. But who can blame them? We’re all trying to stay humble and grounded amidst a massive hype cycle that we neither started nor have any long-term interest in perpetuating.
When I was hiring software developers a decade ago, we tossed candidates aside who seemed to feel they were hot commodities; we didn’t want to hire a bunch of arrogant, aloof divas. These days, the race for data-science talent has built up egos in a similar way that does a disservice to hiring managers and job seekers alike.
That’s not to say that businesses don’t need to attract rare talent that gives them a competitive edge. Data scientists and machine-learning practitioners are in high demand for good reason. But the businesses that succeed in finding candidates for these roles don’t compel them to be as boastful as possible; they follow much the same approach as they do for recruiting senior-executive talent. It’s a relationship-building process, more focused on the company selling the position than the candidates selling themselves.
What Does “Better” Look Like?
The bigger problem is that many companies view data science and machine learning as checkboxes on their operational to-do lists. Hiring a data scientist checks the box, and they’re done. Businesses that haven’t really teased out the connection between that type of role and the return on investment they’re expecting from it won’t ever have the right tools to hire appropriately. This can also help you avoid both overpaying and underpaying data scientists. How much they’re worth is a lot clearer once you know how much value that person will bring to your business.
The hype cycle bears a lot of blame for this problem. Companies are afraid of missing out on the benefits of data science and machine learning. Investors are starting to ask tough questions about how these emerging technologies will play into businesses’ larger strategies. And that’s all good and well. But very few are talking about concrete solutions. Hype gets all the likes.
Doing better means integrating these emerging technologies as strategic solutions to clearly defined business problems. It means more executive oversight for these roles and sticking to a clear schedule of deliverables. And yes, it also means less sexy, over-the-top job listings—which will be much easier to fill.
Based on my experience in the field, there are actually plenty of data scientists to go around. Clarity around how data science and machine learning solve business problems is a lot less plentiful. If you ask me, that’s the real skills gap.
Vin Vashishta is founder and chief data scientist at V-Squared Data Strategy Consulting. Follow him on Twitter at @v_vashishta.