Insights and tips on how to prepare for a successful transition
As artificial intelligence becomes more and more popular, more companies and teams want to start or increase their use of it. Due to this, many positions are coming into the market or gaining importance. A good example is what a machine learning/artificial intelligence product manager looks like.
In my case, I transitioned from a data scientist role to a machine learning product manager role two years ago. During this time, I have seen a steady increase in job offers related to this position, blog posts and talks discussing it, and many people considering or interested in moving into this position. I was also able to see my passion for this role and how much I enjoy the day-to-day work, responsibilities and value I can bring to the team and company.
The AI/ML PM role is still quite vague and is evolving almost as quickly as cutting-edge AI. Although many product teams are becoming relatively autonomous with AI thanks to plugin solutions and the GenAI API, we will focus on the role of AI/ML PMs working on core ML teams. These teams typically consist of data scientists, machine learning engineers, and research scientists, along with other roles involved in solutions (traditional ML use cases, need for LLM fine-tuning, specific disciplines) where GenAI via API may not be sufficient. . House use cases, ML as a service product… ). Check out one of my previous posts for an example of such a team. “I work in a multidisciplinary machine learning team to deliver value to our users.”
In this blog post, I will cover the key skills and knowledge needed for this position, how to get there, and the lessons learned and tips that helped me in this transition.
There are many skills and knowledge required to succeed as an ML/AI PM, but the most important ones can be divided into four groups: product strategy, product delivery, influence, and technical fluency. Let’s take a closer look at each group to better understand what each skill set means and how to obtain them.
product strategy
Product strategy is about understanding your users and their pain points, identifying the right problems and opportunities, and prioritizing them based on quantitative and qualitative evidence.
For me, as a former data scientist, this meant immersing myself in the problem and the user pain to be solved rather than a specific solution, and thinking not about where to apply this cool new feature, but whether it can provide more value to the user. AI model. I find it important to have a clear understanding of Objective Key Results (OKRs) and to care about the bottom line impact of the initiative (delivering results instead of outcomes).
Because product managers must prioritize tasks and initiatives, we’ve learned that it’s important to balance effort and reward for each initiative and ensure this influences decisions about what and how to build solutions (e.g., the project management triangle (consider scope, quality, time) ). An initiative is successful if it can address four key product risks: value, usability, feasibility, and business viability.
The most important resources I have used to learn product strategy are:
- Ben Horowitz’s Good and Bad Product Managers.
- Everyone recommended it to me, and the reference book I now recommend to aspiring PMs is “Inspiration: How to create tech products customers love“, by Marty Kagan.
- Other books and authors that have helped me get closer to user space and user problems include:Continuous discovery habits: discovering products that create customer and business value“, Teresa Torres.
Product delivery
Product delivery is about being able to manage your team’s initiatives to efficiently deliver value to your users.
I started by understanding the product feature stages (discover, plan, design, implement, test, release, and iterate) and what each stage meant to me as a data scientist. We then discuss how you can bring value. “efficiently”: Start small (through minimum viable products and prototypes) and deliver value quickly through small steps and iterations. To ensure the initiative is moving in the right direction, we’ve found it’s also important to continuously measure impact (e.g. through dashboards) and learn from quantitative and qualitative data to adapt to next steps through insights and new learnings.
If you would like to learn about shipping your product, we recommend:
- Some previously shared resources (such as the Inspired book) also cover the importance of MVP, prototyping, and Agile as applied to product management. I also wrote a blog post about how to think about MVPs and prototypes in the context of ML initiatives. When ML meets product – less is often more.
- Learn about Agile and Project Management (e.g. through this crash course) and which Jira or project management tool you currently use at your company (through videos like this crash course).
affect
Influencing is the ability to gain trust, collaborate with stakeholders, and guide teams.
Compared to the role of a data scientist, the day-to-day work of a PM is completely different. It’s no longer about coding, but about communication, coordination, and (lots of!) meetings. Excellent communication and storytelling are key to this role, especially the ability to explain complex ML topics to non-technical audiences. It also keeps stakeholders informed, provides visibility into the team’s efforts, and ensures alignment and buy-in on the team’s future direction – demonstrating how it will help address the biggest challenges and opportunities and earn their trust. ) has also become important. Lastly, it is also important to learn how to challenge, how to say no, how to act as an umbrella for the team, and sometimes how to deliver bad results or bad news.
Here are some resources I would recommend on this topic:
- A complete stakeholder mapping guide, Miro
- Here are some must-read books for data scientists and ML product managers:Storytelling with Data — A Guide to Data Visualization for Business Professionals“, by Cole Nussbaumer Knaflic.
- To learn more about how you can influence and empower your team as a product manager, click hereEmpowerment: Ordinary People, Extraordinary Products”, Marty Cagan and Chris Jones.
technical fluency
Technical fluency for an ML/AI PM means knowledge and sensitivity in machine learning, responsible AI, general data, MLOP, and backend engineering.
your Data Science/Machine Learning / A.I The background is probably your strongest asset. Take advantage of this! This knowledge allows you to speak the same language as data scientists, have a deep understanding and challenge of the project, and have a sense of what’s possible, what’s easy, what’s not, potential risks, dependencies, edge cases, and limitations.
We’ll be driving products that impact users, including: Responsible AI Awareness becomes paramount. The risks associated with not considering this include ethical dilemmas, company reputation and legal issues (e.g. specific EU laws such as GDPR or AI law). In my case, I started with Fast.ai’s Practical Data Ethics course.
Normally data fluency You’ll also need (which you’ve probably covered too): analytical thinking, curiosity about data, understanding where it’s stored, how to access it, the importance of historical data… Above all, it’s important to know how to measure impact. , business metrics and their relationship with OKRs, experiments (a/b testing).
Since your ML model may need to be deployed to have a final impact on users, you may want to work with a machine learning engineer on your team (or an experienced DS with model deployment knowledge). You will need to develop a sense of. MLOP: What does it mean to put a model into production, monitor and maintain it? At deeplearning.ai you can find a great course on MLOP (Machine Learning Engineering for Production Specialization).
Finally, your team may also encounter situations where: backend As an engineer (typically responsible for integrating deployed models with the rest of the platform), in my case it was a technical field and outside of my expertise, so I had to invest time learning about BE and getting a feel for it. At many companies, PM’s technical interviews include BE-related questions. Get an overview of several engineering topics such as CICD, staging and production environments, monolith and MicroServices architecture (and PRO and CONT in each setup), pull requests, APIs, event-driven architecture, and more.
We’ve covered the four most important knowledge areas for ML/AI PMs (product strategy, product delivery, influence, and technical fluency), why they’re important, and some ideas about resources that can help you achieve them.
As with any career advancement, I found it important to define my plan and share my short- and medium-term desires and expectations with my managers and colleagues. This allowed me to transition from a data scientist position at the company to a PM role. This made the transition much easier. I already knew the business, the products, the technology, the way we work, and the colleagues. I also found mentors and colleagues within the company where I could ask PMs questions and learn and even practice certain topics. interview.
While preparing for the interview, I focused on changing the mindset of thinking about whether to develop or not to release. I find BUS (Business, Users, Solutions) to be a great way to structure your responses and apply new ways of thinking during interviews.
What I’ve shared in this blog post may seem like a lot, but it’s much easier than learning Python or understanding how backpropagation works. If you’re still not sure whether this role is right for you, know that you can always try it out, experiment, and then return to your previous role. Or maybe you, like me, will love being an ML/AI PM!