Hiring a data scientist or machine learning engineer is difficult enough, what with the competitive job market right now. It’s even tougher to hire that first data scientist, because this person will set the pace for the company’s data efforts for years to come. There’s a lot of pressure to get it right.
Conventional wisdom sometimes leads companies to look for a freshly-minted PhD who lives and breathes neural networks or statistics. This might make for an excellent hire down the road, sure. But for this first data hire, conventional wisdom will lead you astray.
When creating those job postings and interviewing candidates, what you’re really looking for is an experienced practitioner with a broad skill set. Bonus points if they also have some software development experience on the side.
The first person on the data team will lay the groundwork for your company’s AI efforts. Your future data hires will work with the tools this person has built and code they’ve developed. Even if they prefer to remain in an individual contributor (IC) role as the team grows, that first hire is still a de facto leader.
Given that, you really want to hire someone who’s “been there.” Someone who is well-versed in industry best practices. They’ve seen the various ways an ML effort can go awry and they know how to course-correct it. They also have experience working closely with stakeholders and product teams, so they’ll be able to guide your company on what AI really is and what to expect during projects. Related to that, they’ll know what it means to connect AI to the business mission.
By hiring an experienced practitioner, you will save time and effort in getting your data projects off the ground, and improve the chances that they succeed over the long run.
An unfortunate trope in this field is the Data Scientist Who Only Builds ML Models. This person refuses to handle any other tasks – working with stakeholders, deploying models to production, whatever – so they kick it over the proverbial wall to someone else. (Anyone else, really.) And there are plenty of data scientists who embrace that trope.
Such a data scientist might be a good fit in a company that has an established data practice. Maybe. But your company is just starting out, so you don’t have the bandwidth to bring on someone who is so hyper-specialized.
You want someone who keeps up with the tools and techniques of the field, so they’re ready to tackle a grab-bag of data-related projects: train ML models, perform ad-hoc analyses, work with databases (to pull data), develop data pipelines, and deploy models to production.
Beyond their technical skills, your preferred candidate has a strong aptitude for collaboration. They’ll need to explain the finer points of data science and ML modeling to various people in the company. As the team grows, they will likely work side-by-side with specialists (perhaps one of your projects calls for deep experience in a certain area of AI) and must be able to do so in a way that welcomes this new, focused expertise. They’ll also work closely with your software developers and ops teams to get models into production and keep them running.
And that takes us to the bonus points of software skills:
Data scientists and machine learning engineers write plenty of code in order to train ML models, sure. But they also need to be able to write code for reproducible data pipelines and feature engineering work. This code needs to be performant (so it doesn’t just run the analysis, but does so efficiently) and the data scientists should manage it with a source code control system, such as git.
All of these tasks are second nature to a software developer. A data scientist who comes from a software background will have a much easier time doing this part of the job. They’ll be able to work more independently and their code will be maintainable by others.
And even if your data science team doesn’t manage model deployments themselves, they’ll still need to perform a “handoff” of that model and code to the operations or app dev team. An all-too-common friction occurs when a data scientist just hands the model to the developers and says “OK it’s your problem now.” By comparison, someone who speaks the developers’ language – because they’ve done that job before – will provide a much smoother handoff.
Every company is different, and the data field is very young. As such, there’s not a lot of detailed, actionable guidance that generalizes across all companies and industry verticals.
That said, when it comes to that first data scientist – frankly, your first few data scientists – you’ll want to hire someone who has done this kind of work before, who can work independently, and ideally who has been responsible for production-grade code. Doing so should improve your company’s data efforts in both the short- and long-term time horizons.