Summary: What is an AI Product Manager and how do you know when you need one.
The role of Product Manager (PM) can mean many things dependent on the specifics of the company, its markets, its channels, and the variety of its products. It’s almost impossible to put a single label on the responsibilities of a PM except that he or she is responsible for all the features and capabilities of the products in their care.
In a smaller or mid-size company the PM role can be quite broad extending to supply chain, contract manufacture, distribution, channel selection and maintenance, and pricing to name just a few. In a larger company it may be quite narrowly defined as determining product features and building and maintaining the product alongside many others.
Whatever the case the role of PM has always been a plumb assignment allowing the job holder a key role in company success and with a broad view of how their product or service enables that.
Recently there’s been a good deal of conversation about a new type of PM titled “AI Product Manager” (AI PM). There seems to be a good deal of confusion around this role.
- Does this mean that if you introduce AI/ML features into a product you need a separate AI PM?
- Is this in addition to or does it replace your traditional PM?
- What does this mean about the roles and responsibilities of your data science team and how they interact with LOB managers?
The list of questions goes on and a little clarification is in order.
When a PM is a Good Organizational Choice
If you are adding AI/ML to an internal function, perhaps to increase capability or reduce cost, these are not typically projects where you might assign a PM.
For example Walmart is using computer vision, weighted shelf sensors, and other IoT capabilities to assist employees in determining when specific product inventory is low or produce is going bad.
Exxon is using AI/ML for its Drilling Advisory System to optimize undersea drilling methods for safety, performance, and cost.
It also might be something as simple as an RPA application in accounts payable to reduce errors or speed processing.
These are clearly internally focused good uses of AI/ML. Your senior data science project leaders along with their LOB manager teammates will continue to handle these internal projects.
In general we would think about organizing around a PM only when there is a clearly defined external customer and a corresponding external customer need to be addressed.
So When Does a Regular PM become an AI PM?
We’re specifically going to leave out cases where the PM is working for an AI-first company. If you’re at Apple, Microsoft, or any number of AI startups building AI-first products like improving Siri results, making self-driving cars, or making photos searchable for ecommerce then you are by definition an AI PM.
It’s more likely though that you are enabling a major customer facing product or service using AI/ML like these:
- United Health implementing extensive NLP in authorizing doctor-recommended medical treatments.
- CVS Health building a system to provide AI-powered healthcare recommendations automating patient intake and asking a series of questions about symptoms and previous health history.
- AT&T using ML algos to better match advertisers with their ideal customers.
- Bank of America rolling out its in-app customer servicer bot.
Some of these like the CVS example represent wholly new products or services. The other examples however are major innovations applied to existing services.
In the emerging digital landscape all managers and particularly PMs need very high levels of data literacy. That extends to the capabilities and limitations of each type of AI/ML. The degree to which AI/ML will change an existing offering (or create a new one) will determine the set of skills your PM must possess.
What are the AI Skills an AI PM Must Have?
The question of whether your PM must have these skills personally or be assisted by others that have them depend mostly on the size of the project and the size of the company. Obviously, if investment and potential reward are very high then resources must match.
In smaller mid-size companies with more modest but equally as important and complex applications of AI/ML then you may need a PM with extensive hands-on data science skills. Whichever is the case that list of capabilities most likely includes these:
A fundamental understanding of what each type of AI/ML is capable of delivering, and equally as important, what its limitations are.
From here, the product development stage passes through much the same process as for a non-AI/ML product or service:
A clear statement of the customer need and a framing of the problem that sets scope and bounds.
Innovation, design, and UX development using the same mockups, wireframes, and user surveys as traditional development.
Selection of an agile and rapid development process.
AI/ML comes to the fore in feature development which frequently is dependent on both the selection of an AI/ML technique or approach and an assessment of data. Is the data available? Can it be made available at an acceptable cost and time? Can it be managed through the life of the product or service?
Creating the models or AI capabilities (text, speech, and computer vision) is an iterative procedure that may take dozens or even hundreds of experiments to get right. These are heavy duty AI/ML skills where your AI PM will need extensive support from internal or external data science resources.
Maintaining the models once selected is a separate task. If they are custom coded using Python or R then resource requirements may remain heavy. In product design consideration should also be given to automated or semi-automated platforms like DataRobot or Tazi that can make continuous monitoring, updating, and serving the production code a more efficient task.
The more extensive the AI/ML component of your project, the more the definition of an AI PM becomes relevant. A fully competent AI PM would have the general business acumen of a traditional PM with a full set of AI/ML skills. At a minimum though you AI PM will also possess experience in:
- software engineering
- specialized UX design unique to AI/ML
- knowledge of the full pipeline for development and ongoing support
- data lifecycle and pipeline management
- AI/ML experimentation and measurement
- and the unique aspects of AI/ML development like the CRISP-DM development process and continuous deployment.
That’s a very difficult set of skills to find in one person. Chances are you will need a team of internal and perhaps external resources to make your AI/ML project a success.
About the author: Bill is Contributing Editor for Data Science Central. Bill is also President & Chief Data Scientist at Data-Magnum and has practiced as a data scientist since 2001. His articles have been read more than 2.1 million times.