This post part of a series on undervalued practices in ML/AI:
- Part 1: Getting started with ML/AI
- Part 2: Hiring and Team Structure (this post)
- Part 3: Planning projects
- Part 4: Project Execution (this post)
… until you actually have demonstrated a need for a full-time role.
Experienced data scientists can tell when there’s no work for them to do and they’ll steer clear. They can usually determine this from the interview. Some may even spot it in your job posting. So if you wait until the proper time to look for a full-time hire, that will shorten your search to fill the position.
Holding off will also reduce the chances that you’ll need to reopen the search later on. It’s not at all uncommon for a data scientist to settle into a job, get bored because there’s not enough interesting work to do, and then leave.
Postponing that first full-time hire doesn’t mean that you have to put your data work on hold. You can partner with an outside team on a part-time or per-project basis in the meantime.
This arrangement also helps with the “cold start” problem in hiring: you’re much better off starting from a fractional data scientist than from zero. If you already have (part-time) data scientists on-board, job candidates can meet with them to discuss the deeper technical details of what you’re doing, and in the same language.
Your first full-time data scientist will likely be an experienced data generalist. This is someone who has a well-rounded skill set – some data engineering, a variety of ML/AI methods, business knowledge, and some software development knowledge on the side – so they can lay the groundwork for future hires. This person will not just “analyze data,” but build out your company’s data practice.
Resist the temptation to hire that data scientist who has a very deep, but narrow skill set in whatever is the latest AI technique. They might be a great fit for you later, but not now.
In the previous post, I mentioned that everyone who interacts with the data scientists should understand all of the high-level concepts and terminology. That also holds true for the other direction: your data scientists should understand your industry vertical in general, and your company’s challenges in particular.
It’s rare that your data scientists will have this deep knowledge when they join the company. Take the time to explain it all. Not just the public-facing mission statement, but the deep mechanics of how the company works. Connect them to people in various departments as needed.
Doing this early on, before you turn the data scientists loose to analyze data and build models, will pay off in the long run. They’ll have a better understanding of when their analyses have produced meaningful results. As a bonus, they’ll be able to come up with ideas on their own.
The last thing you should ever do is stick your data science team in a corner and hope that they produce magic. Besides being unproductive for the company, that is usually the first step to boredom for the data scientists … which leads them start their next job search.
Teaching your data scientists about the business is closely related to the next undervalued practice:
Straightforward data analyses can yield useful insights, especially if your company has never done any kind of data work before. But the true power of ML/AI happens when you determine how (or, “whether”) to use ML/AI to drive your products. That requires data scientists to have a seat at the product table, and to get a voice in product decisions very early on. If you’ve completely mapped out a product and then handed it to the data scientists, it’s too late. You’ll likely have to backtrack on some decisions or even scrap large swaths of the product idea in order to bring it in-line with the realities of ML/AI.
As a bonus, some data scientists may take on product management roles over time. This is a very powerful skillset and you would do well to nurture it.