Company leaders will tell you they need more product managers with strong machine learning knowledge. But here’s the catch, they just aren’t ready.
When I use the term ML PM, I am thinking of a product owner or manger that understands the problems that can be solved using machine learning and which cannot. The job involves identifying the right problem, framing it to data scientists, and facilitating its successful evolution to a repeatable, scalable product offering. The role also includes being able to show and explain the feature to prospects and customers. To a product company looking to integrate data science work products, this individual can be very valuable.
That said, I can’t tell you how many times that I have been at conferences where business leaders bemoan the lack of individuals who deeply understand ML and product. So much so, that people with these two skills are referred to as unicorns. There is profound desire for this person to exist but a complete lack of belief that this person could actually be.
At first, I thought this analogy was funny. I am someone who went from developing models to incorporating them into a product as the product manager. They thought I was a mythical creature. And then I thought “Wait, this might be good”. I was excited at the thought of being a forerunner in a new and exciting profession.
But, the longer that I remain in this role, the more I realize that there are really good reasons why there aren’t more of us. And, in my opinion, it has nothing to do with the lack of people that could understand both domains.
In the product management roles that I have had, there is no doubt that I have benefited from understanding machine learning. However, most of my exposure to machine learning in products involved either leveraging ML services, using pre-trained solutions, or developing relatively simple classical machine learning solutions. The implementation of these solutions did not require many years of grad school and a deep understanding of ML.
In fact, to be honest, a considerable portion of my work has been involved in the data infrastructure, monitoring, and serving the results of ML. I would argue that my engineering knowledge (not ML) has been more than sufficient to fulfill a PM role within these particular areas of responsibility. So the problem is, companies want to hire a PM with deep ML knowledge and then assign them the tasks on the ML periphery. Bad fit for the PM (yawn) and bad for the company (not all ML experts understand design principles).
The crux of the situation is that many companies often go out looking to fill this role with individuals having an extensive level of ML experience and expertise. This can make the role difficult to fill because it is written for a PhD level of knowledge when in fact most PhDs working in data science are not ready to become PMs, since the field is relatively new. For PMs with an engineering, math or other scientific background, the job description is written in a way that can seem out of reach for everyone except those prone to lie. This leaves companies feeling like they can’t fill the PM role. I believe that what the hiring manager needs and what they have gone searching for, are at odds.
Additionally, speaking from my own personal experience on developing models, there can be a fundamental tension between the engineering goals and data science goals. If the industry want to grow this expertise, it takes patient mentoring to convert an “optimize the math” mindset to a “deliver the incremental value” point of view. And then there is the general lack of desire to grow people in industry — but that’s for an entirely different article.
Next time you write a PM job description, ask yourself — what do I truly need. Bad fit leads to repeated vacancy.
