Hi all - I am a Product Manager in technology, and I have spent the last ten years of my career working primarily on Machine Learning-powered products such as Web search, personalization and recommendations in online travel and e-commerce, and on recovery of hacked accounts. Before becoming a product manager, I was a data analyst and then an analytics team lead in a small machine translation company (before Google Translate made it all the rage! :)

In my effort to make tech a slightly better place for everyone, I have been mentoring folks who want to join the industry or advance in their careers. I mostly focus on product management, but I also get questions about the specifics of other tech roles, which makes me wonder how transparent, or not, tech is for those on the outside and for newcomers. These questions inspired me to write about the role of a tech product manager in general and my specialization in particular. If you find this article helpful, I’d be happy to write about other tech roles, too.

Let me start with a very brief and high-level explanation of what the Product Manager (PM) role is about before I deep-dive into the specifics of Machine Learning product development. Since there are numerous articles on the topic available online I will only touch upon some key concepts. 

A product manager's main mission is to satisfy customer needs while making the most of the company's strengths to bring in extra revenue. In tech companies, the strengths in question are usually technology and talent.

PMs balance the needs of multiple internal (business, development teams, legal and compliance, etc) and external (customers, regulators, etc) stakeholders and find solutions that drive maximum cumulative value. 

Working with cross-functional product teams —software engineers, UX and content designers, user and market researchers, data analysts —they set the direction and define success, and help the team prioritize and execute. On the outside, PMs manage communications on behalf of the team and the product, as well as drive alignment on strategy, priorities, and resources with stakeholders across organizations and functions. 

Typical career progression for a PM looks as follows (source: Product Plan):

At the start of a PM career, the scope is generally smaller (e.g. you oversee only a specific product feature or a single product among many) and the focus is primarily on execution. With time, the scope and the team grow and the focus shifts from execution towards strategy, resource planning, and business profitability:

Personally, I find the role exciting because of its broad, generalist nature and the high level of agency it gives;  I also enjoy working closely with people from different roles, sharing our diverse perspectives on how to find and then deliver the right solution for the customers, the business and the team. As a PM, you are empowered to call the shots but you are also accountable for landing the message with your team and stakeholders and ultimately for getting the promised result. It is a leadership role with its pros and cons, and if you pursue it, you need to get comfortable with both the power and the responsibility.

Machine Learning (ML) is “a field devoted to understanding and building methods that let machines "learn" – that is, methods that leverage data to improve computer performance on some set of tasks.” (Wikipedia).

Nowadays ML applications are everywhere; to give you a few examples, they drive web search, speech and image recognition for authentication, fraud detection, product recommendations, online ad targeting, computer vision in robotics and self-driving cars, diagnostics in medicine, predictive maintenance in oil and gas, algorithmic trading and more. The most recent advancement is, of course, generative AI that utilizes deep learning to learn to produce texts, images, video, and audio content. 

In the context of B2C applications (e-commerce, social media, search, streaming, etc.), products that leverage well-optimized ML offer superior performance and increase product stickiness by making user experience frictionless, more engaging, and serendipitous. They automatically adapt to changes in user behavior and they scale well to more users without losing performance (assuming no constraint on computing capacity). Due to its nature, ML products lend themselves to systematic improvement, which means continuous revenue increase, especially at a high scale when a small delta in performance leads to major gains thanks to amplification by traffic. A well-tuned ML solution is hard to replicate, and that gives ML-investing businesses a sustainable competitive edge. 

That being said, high-performing ML comes at a high cost. It requires high-quality data and investment into compute capacity, data infrastructure, and specialized talent — ML engineers, data scientists, and data engineers. On top of that, for ML to thrive a company needs to have an established data culture and tolerance to longer development cycles and opacity stemming from the complexity and limited interpretability of sophisticated ML models. 

ML products have unique advantages and unique demands, and managing them comes with a distinct set of perks and challenges. As a PM in the ML/AI space, you get to work with cutting-edge technology and the industry’s top talent, building products that just recently were deemed impossible. Yet, you will need to do a little extra compared to your non-ML colleagues. 

Understand the tech. ML done well requires a deep understanding of the technology and a systematic, rigorous approach to model improvements. High-performing ML teams often include members with Ph.D. degrees and show scientific rigor in approaching the task at hand. I find it extremely gratifying to work with such intelligent people on problems that are so complex, but also rewarding.

To do so effectively, you need to have a good grasp of how the tech works and what the development process is like. Otherwise, you’ll struggle to create realistic product plans and get buy-in from the team and other stakeholders. Gaining this expertise and maintaining it is a serious time investment, but if you find the topic exciting, it is a time well spent.

Evangelize. Advanced ML is effectively a black box that is impenetrable for people who don’t work on it directly. This opacity, or a lack of explainability and interpretability, makes many people uncomfortable with ML/AI, especially if they witnessed poor implementations with underwhelming performance. A job of an ML PM often entails educating stakeholders on how ML works, what its unique strengths and limitations are and how it is best leveraged for applications within the company.

Deal with the dark side of ML. ML is an incredibly powerful technology that can offer unparalleled performance; but at the end of day, it is only as good — or as bad — as its training data and the algorithms under its hood. ML tends to amplify biases, i.e. to make certain predictions at a higher rate for some groups than expected based on training data statistics, and thus exhibit discriminatory behavior when applied in real life: e.g. Amazon’s recruiting AI showed preference towards male applicants since it was trained on historical hiring data with the majority of male applicants and hires, and had to be scrapped [Reuters].  

Another problem with really advanced ML is that the accuracy of its predictions may become uncanny or outright creepy. For example, the belief that Facebook listens in on your phone calls has persisted for so long because of the extremely well-targeted and well-timed in-app ads [if you are interested to learn more, check out this article]. Building explainability features and data controls into your user-facing products may be needed to mitigate user concerns — and sometimes to meet the quickly emerging regulatory requirements. 

Always think ahead. ML is an expensive investment, and as a PM, you should look for as many applications and additional revenue streams as possible to improve its ROI. The last thing you would want is to stand up a team and build the needed ML infra for a single-use model. Thus, whether you are setting up a new ML team or joining an existing one, build and continue refining a portfolio of multiple short- and long-term ML projects to make sure your team always works on impactful, business-relevant goals. 

This type of planning is not that necessary for regular product teams that can switch from one problem or business area to another with reasonable ease, given a ramp-up period. ML engineers rarely can, or would want to, work on non-ML projects and thus need a backlog of ML projects prepared ahead of time.

Product management is still a fairly new role, and it does not have an established college-to-career path. This is particularly true for PMs who want to specialize in Machine Learning. So how does one become an ML PM?

Firstly, how does one become a PM? The most typical route is a transition from another tech role: data analyst, engineer, UX designer, or Product Marketing Manager. Most companies, large and small, support such transitions and allow their employees to do on-the-job training. For companies that do ML, this also works, including the transition from a non-ML PM to an ML PM. So join a company that does ML, and when the opportunity presents itself, make the move. 

Another route, especially when you are fresh out of college or want to make a drastic career and industry change, is Associate / Rotational PM programs at large tech companies [check out this overview for more details]. During these programs, you will get a chance to work with different product teams, which may include ML. If everything goes well, upon completion you will be offered a job and you can likely choose an ML team at that point.

At non-ML companies, you may be able to build a business plan for investing in ML and get buy-in from the leadership to set up a team. This path is more feasible for senior PMs who carry weight within their organizations, or for smaller, bolder, more nimble companies open for experiments. 

And of course, if you are bold enough, there is always an option to start your own business and choose whatever field or technology you are passionate about and see potential in. If this is the path you go for, kudos! The world needs more high-tech businesses to keep the competition alive. 

Product management is an exciting career, and if you get a chance to work inML, it is even more so. I hope to see more and more people discovering and joining this space, and more badass new products getting launched. If you are interested, best of luck! We might bump into each other one day ;)