(Image: cropped from the “bike stick bicycle” meme from imgflip)
Hiring is always important, no matter what the role. But the first hire in a new department is critical. Doubly so when that department represents a new field, like AI, that isn’t understood well.
Today I’ll explore four antipatterns – that is, practices you want to avoid – when hiring your company’s first data scientist or ML engineer.
Some of these statements may sound familiar. You may have even heard them uttered, with a tone of self-confidence, by your fellow CEOs or CTOs. Just know that these approaches rarely work out well.
Why it sounds reasonable: This hire will be the first person in the company who understands data and AI. Why not defer to their expertise on what to do?
Why it falls apart: While your data scientists should know the most about data compared to the rest of the company, they shouldn’t be the only ones to understand it. Your leadership team would do well to develop data literacy – an understanding of what AI really is and what it can(not) do – such that you’ll all know what’s realistic for your company.
This is especially helpful when hiring. The term “data scientist” is like “doctor” or “attorney.” There are certain things that anyone in that profession should know; but when it all comes down to it, you’re looking for someone with experience in a specific area. Do you need a podiatrist or a cardiologist? Someone to handle your real estate, or your recording contract? The same holds with data: do you need someone with skills in time series analysis? Natural language processing? Image recognition? Any data scientist can learn these skills; but if you need a certain skill and a hire already has it, you’re in a much better position.
Additionally, if you don’t already know what you want to do with your data, hiring a data scientist is a roll of the dice. You might get someone who has a lot of hands-on data experience and also understands your business model. Maybe. But many people who take the title “data scientist” expect to focus on the technical implementation of ML models and data analysis; they are neither interested in nor prepared to take on the role of devising data strategy.
What to do instead: Take the time to develop a plan for how you’ll use data. Line up specific projects (say, “let’s try to predict customer churn” or “can we classify incoming customer support requests?”) and make sure you actually have the data to execute on them (“do we have the last few years’ customer records and support requests in electronic format?”). Then, after that, you’re ready to develop a job posting and go on the hunt for your company’s first data scientist.
Why it sounds reasonable: Some early data scientists held advanced degrees in economics or statistics. Understandably, people unfamiliar with the field often assumed that this degree was a firm job requirement.
Why it falls apart: There are plenty of amazing data scientists who hold PhDs. There are also plenty of amazing data scientists who don’t. While there are some aspects of data science (and machine learning, and AI, and so on) that would benefit from a PhD, it is certainly not required.
What to do instead: Focus on candidates’ capabilities and experience instead of their academic credentials (or lack thereof).
The other big reason to not focus on a person’s degree is because that’s often a euphemism for “someone who is fresh out of grad school.” And that leads to the next antipattern:
Why it sounds reasonable: Maybe your budget doesn’t stretch enough to hire someone with a ton of experience. Or you have the budget, but the senior-level candidates you interview keep turning you down. The only people who are willing to move forward are fresh out of school, and you need to get started on your data projects, so … Going with an entry-level candidate is fine, right?
Why it falls apart: If this is your company’s first data hire, you are filling a very special talent gap. You don’t just need muscle to execute on projects; you need someone who can show the company how data projects are actually done. Someone with hands-on experience training and deploying models, and analyzing data. They’ll understand how to work with stakeholders, product owners, and software developers in order to bring your projects over the finish line.
Compare that to an entry-level hire. They’re smart, they’re eager, and they’ll definitely learn over time … but … they’ve never done this before. They’ll wind up making decisions well beyond their knowledge. Through no fault of their own, there’s a very strong chance that they will cost you money down the road as you undo (and then redo) their work.
**What to do instead: **Hold off on entry-level hires until you have built an environment that is able to support and nurture them. Ideally, that means inviting them into a team where experienced data scientists and ML engineers can mentor them on industry best practices and show them the ropes on how they’ve built out your company’s data shop.
Let’s say that you’ve developed a realistic road map and you have identified a time-sensitive use case for ML/AI. Great! A few months pass by but your job post just isn’t getting any bites. It’s tempting to keep holding out for a full-time employee (FTE), isn’t it?
Why it sounds reasonable: You really, really want an FTE in this role. You want someone who can grow with the company and really become a part of the team long-term.
Why it falls apart: For your first data hire, you face two competing concerns: one is to find an FTE to fill the role; the other is to get started on your data projects.
Even with recent layoffs from the bigger-named tech companies, you’re still in competition with … every other company that wants a data scientist or data engineer or ML engineer. It could take ages to fill that seat. And in the meantime, the clock is ticking on your time-to-market. What’s the opportunity cost of you waiting for your perfect hiring scenario?
The solution: It’s time to engage outside help to get you started. Whether you retain the services of an individual, independent consultant or a larger firm, want someone who has tons of experience. They’ll be able to deliver on the projects and also bring your company up to speed on industry best practices along the way.
All of these antipatterns certainly sound logical, don’t they? I can see why executives would follow them. The trick is that, as the name would imply, an antipattern is really a recipe for a bad outcome. At best, they work because of luck. At worst, they cost you time, effort, and money as you have to course-correct (or even start over).
Building out a data science team starts with this critical first hire. This may sound daunting but you can still shift the odds in your favor: develop a plan, focus on candidates’ skills, seek out an experienced practitioner, and kick-start your efforts with outside help if needed. These steps will improve your chances of success in hiring, and position your data efforts for a win over the long term.