Will every Product Manager become an AI Product Manager in the future?

Photo by Andrea De Santis on Unsplash

Thanks to the AI hype around chatGPT, I recently heard more and more people state that in the future every Product Manager (PM) will be an AI Product Manager.

Is that true? And if yes, what does this mean for every PM out there now who does not have a Computer Science degree, are they soon out of jobs?

To answer this question, let’s take a closer look at the statement that ‘every PM will be an AI PM’. There are three immediate questions that pop up in my head that I want to clarify in more detail:

1. What is meant by AI in this context?
2. What is an AI Product?
3. What is an AI PM? And do you need to have a CS degree to become one?

What is AI in this context?

The interested reader will likely already know that AI is a super broad concept, that describes the capability of a computer system to mimic human cognitive functions such as learning and problem-solving and that AI has many sub-fields, including for example Machine Learning (ML), Computer Vision, Robotics, Expert Systems, Speech Processing, Natural Language Processing, …

Especially the terms AI and ML are very often used interchangeably in conversations — but in the context of this discussion the difference matters:

• Products that are powered by simpler rules like expert systems are not that difficult and not that risky to build and maintain.
  Some human defines a list of rules that are then used by the system to make decisions. These systems operate in a deterministic framework, the outcome of the system is certain given a specific input.
• Products that are powered by machine learning are much harder to build and maintain.
  The rules the system uses to make a decision are in this case not defined by a human, instead, the system learns those roles from data (during what is called a ‘training phase’ in the development process). And depending on the specific ML algorithm used, like with neural networks, the rules are a black box; we humans won’t be able to understand the rules the system learned from the data it was fed with. These systems operate in a probabilistic framework: the outcome is a prediction or decision based on the likelihood of a certain outcome.

So when people talk about AI Product Management, AI Products, or AI Product Managers, they most often refer to building products that use Machine Learning in some way. Building ML systems require different skills and knowledge, special risk management to launch and work well for users & business, compared to ‘traditional’ software development as we knew it: According to Gartner approx. 85% of AI/ML projects fail due to unclear objectives and obscure project management processes; 87% of AI/ML projects never get to the production phase; while 70% of clients indicated minimal or even no impact from AI/ML.

What is an AI Product?

I want to differentiate here between a few possible types of AI products:

1. Products that apply AI to enable (core / unique) new user experiences.
2. Products (or rather platforms) that are built for engineers to build AI technology — i.e. to help them train, launch, and operate ML models.
3. Products that offer AI-as-a-service (off-the-shelf)
4. This last category is not a very serious one(yet ;)) Products that have been built mainly by AI, not humans.

Most people refer to category 1 products when they speak about ‘AI Products’ and mention as typical examples:

• Netflix: The recommendation capabilities are a key value proposition to users.
• Social Media Applications like Instagram or LinkedIn make you spend hours scrolling through our your personalized feed.
• YouTube would not be able to handle the millions of videos uploaded without for example it’s classification capabilities (spam, bad content, content-specific features for Gaming, News,….)

What all those examples have in common, is that AI technology is not only enabling new user experiences, but these user experiences are adding core and unique value for the user and the business.
Products, in which AI is only powering one or more non-essential, non-business-critical feature(s), I’d argue are not AI Products. But sure, if the company is willing to invest in these features, they may still hire an AI PM to build these.

In the early days, building AI/ML products meant that companies would also build all the necessary AI/ML technology and infrastructure in-house. They would hire for example machine learning engineers, data scientists, and data engineers and spend months and years building and maintaining their ML models.
In the last few years, however, we see a massive rise in AI-as-a-Service products (category 3). With those, companies can now implement and scale AI techniques at a fraction of the cost of a full, in-house AI. Any team that wants to use AI to solve key user problems will have to have a ‘build or buy’ discussion (to which I may dedicate a separate blog post in the future). Whether or not products, that merely use AI via an API for truly core user experiences are AI Products or not, is up for debate. ChatGPT just told me it would not classify these as AI products — I would argue they do. One key reason for me is that no matter who built the AI technology (in-house or external), these products expose the same risks to users, society and our environment and hence should be equally regulated. If we don’t call them AI Products, we create a false impression of lower risk.

Quick side note related to the ‘build or buy’ question:
Not every company that claims to be an AI company (and hence builds an AI product) actually uses AI — it sometimes may just be an aspirational claim about what something they might be able to do at some undefined point in the future. The Verge for example reported at some point that 40% of AI startups in Europe don’t use AI.
From a moral standpoint, I don’t support adding AI-related buzzwords to pitch decks to get more venture funding if the product really does not need AI to work well. I do though highly favor experimenting quickly with manually coded rules or AI-as-a-service to validate ideas and build MVPs — but to succeed long-term as an AI Startup / with an AI Product, you will likely need to own some critical IP in-house.

What is an AI Product Manager?

As the role of a ‘Product Manager’ matured in the last years, specializations like ‘Platform Product Manager’ (i.e. https://www.svpg.com/platform-product-management/), ‘Growth Product Manager’ or ‘Technical Product Manager’ started to emerge. Similarly to those, we can define ‘AI Product Management’ as a PM specialization that has all the essential PM skills at its core but adds additional skills and capabilities on top that a PM needs to master to succeed in building AI Products.

Let’s look at the unique capabilities a PM needs to have to build AI Products in-house effectively due to the different nature of AI product development (vs. traditional software development).

• Understand key ML concepts and speak the same language as your engineers, data scientists, and data engineers.
• Build a realistic landscape of AI capabilities within your industry to assess opportunities and use cases. Is ML the right solution for your problem or not?
• Take a new approach to strategy and planning — AI Products require more upfront work and data explorations to identify and validate the use case / opportunity.
• Become exceptionally data-literate and data-driven. Don’t mind getting your hands dirty here — any ML product is only as good as the data it is being fed with.
• Apply a foundational understanding of data science practices and lifecycle to manage the uncertainty and non-deterministic nature of these products.
• Work closely and effectively with data scientists and researchers to build the core models that will power your product experiences.
• Cooperate with various roles at the right time in the development lifecycle. Many roles add value, not just engineers.
• Manage and design for AI-specific risks to make sure your product is trustworthy (inclusive, secure, robust, inclusive, …).
• Be a change agent to help the organization understand the challenges AI products introduce (hint: expectation gap), and help upskill and transform your organizations' culture.

You may have noticed that I deliberately wrote above ‘building AI Products in-house’ — unless you do that (or have never done that ever before) I would not necessarily consider you to be an AI Product Manager.
Working with your engineering team on adding AI-as-a-service for a smaller feature in an existing product is relatively easy and does not require the full list of capabilities and skills I described above. Building ML models from scratch for experiences at the core of your product is a much more challenging endeavor — and definitely requires the full spectrum of the skills and capabilities listed above to be successful.

I also deliberately wrote about ‘building AI Products’. Please notice that I did not say an AI PM is someone who uses AI Products to make their PM life easier. Just because you leverage chatGPT to write a compelling vision document, you are not an AI PM ;)

To summarize, AI Product Management is an emerging PM specialization. Yes, every PM will need to understand the fundamentals of AI and will need to leverage AI tooling— but not every PM will become an AI Product Manager.

Do I need a CS degree to get started with AI as a Product Manager?

To be frank, having a CS degree will give you a big advantage for two reasons: First, it will be much easier for you to learn and understand this new technology. There are a lot of resources available for engineers to get into AI that you can leverage then easily, too. Second, you have an easier time building credibility with your engineering teams.

That said, I strongly believe that anyone, who is a great PM, can learn the fundamentals to help build better ML products for the benefit of our society. AI, including Machine Learning, is complex, but it is not magic.

Here are a few free recommended courses that get you started:

• AI for Everyone on Coursera — from Andrew Ng, co-founder of Google Brain and DeepLearning.ai. designed to teach non-technical individuals the basics of AI and its applications.
• Making Friends with Machine Learning on YouTube— from Cassie Kozyrkov (Google Chief Decision Scientist). This course is for everyone looking to gain an intuitive understanding of how machine learning works, the main models or algorithms in use, and what it takes to take machine learning systems from planning to deployment. She blogs regularly on medium in a really entertaining and understandable way.
• Human Factors in AI on Coursera — from Duke University. This course is a great add-on for PMs on top of the tech-focused one of the two other courses I recommended above. It covers human-centered design and the unique elements of user experience design for AI products. It also touches on data privacy, designing ethical AI, and approaches to identify sources of bias and mitigate fairness issues.

To get deeper, I started to host courses around AI ProductManagement — live (on Maven.com) and soon also video-based (sign up to waitlist).

Like what you just read? Follow me on LinkedIn or Substack, and check out my website. I am available for contract work as a product consultant and leadership coach!

This story has been written solely by a nice human being without any input from chatGPT or similar tools.