AI and product management | Marily Nika (Meta, Google)
Marilee Nika, Product Lead at Meta and Maven instructor, discusses AI's impact on product management. She covers how PMs can leverage AI tools, build AI-centric products, and navigate the evolving landscape, emphasizing problem-solving over 'shiny object' pursuits.
Deep Dive Analysis
18 Topic Outline
Introduction to AI in Product Management
Staying Informed on AI Developments
Overhyped and Underhyped Aspects of AI
Practical Applications of ChatGPT for PMs
The Future: Every PM Will Be an AI PM
Integrating AI: When and When Not To
Data Needs and Build vs. Buy for AI Models
Understanding AI Models and Their Training Process
Real-time AI Translation with Google Glass
AI as an Enhancer, Not a Replacer, for PMs
Why PMs Should Learn to Code and Where
Skills and Challenges for AI Product Managers
Strategies for Gaining Leadership Buy-in for AI Investments
Bridging the Gap Between PMs and AI Research Scientists
AutoML: A No-Code Tool for AI Model Training
Building and Iterating an AI Product Management Course
Encouragement to Create Your Own Educational Content
Lightning Round
5 Key Concepts
AI Model
An AI model functions like a brain that can process an input, such as an image or text, to recognize patterns and make predictions. It outputs not only its identification but also a probability score indicating its certainty.
Training an AI Model
This process involves feeding an AI model vast amounts of labeled data, like thousands of images tagged as 'cat' or 'dog.' The model then processes this data, learning to identify patterns and relationships within it, enabling it to make predictions on new, unseen data.
AI Product Manager
Unlike a generalist PM who focuses on building and shipping the right product, an AI PM focuses on solving the right problem, often by leveraging data and smart solutions. This role involves identifying problems that can be effectively addressed by AI and working with research scientists.
The Shiny Object Trap
The tendency to adopt AI technology simply for its novelty rather than addressing a genuine problem or pain point. It emphasizes that AI implementation should always be driven by a clear, identified need that a smart solution can address.
AutoML
A tool, such as Google Cloud's AutoML, that enables users to train high-quality, custom machine learning models with minimal coding or mathematical expertise. It simplifies the process of developing AI solutions, making it accessible even to non-technical users.
9 Questions Answered
Subscribing to newsletters like 'The Download' by MIT Technology Review or TLDR is highly effective. Even general technology-focused newsletters are increasingly sprinkling in AI content, reflecting its growing pervasiveness.
ChatGPT can be used to rewrite mission statements to be universally understood and inspiring, generate diverse user segments with motivations and pain points, and provide AI-enhanced ideas for product features.
Marily believes this because modern products increasingly require personalized experiences, effective recommender systems (like Netflix), and automation to drive technological investment and societal improvement.
No, AI tools will not replace product managers; instead, they will enhance their work by automating tedious tasks, freeing up PMs to focus on more strategic aspects and unlock new areas of product management.
Learning to code, even the fundamentals, provides PMs with a deeper understanding of how AI tools are created, fostering confidence and a different mindset to approach and leverage technology effectively, similar to learning classical music fundamentals for piano.
Challenges include managing the inherent uncertainty of research and model training outcomes, dealing with the constant change and action required, the difficulty of acquiring good quality data, and clarifying career progression in research-heavy organizations where launches are less frequent.
PMs can gain buy-in by referencing successful adjacent AI-first products within the company as examples, proposing contingency plans to de-risk the investment, and building trust over time, especially in cultures that welcome experimentation and learning from failure.
Large tech companies offering core services like speech recognition or their own generative AI should build and diversify their own models to ensure unique quality, as relying on common datasets will result in similar quality across products.
AutoML, offered by Google Cloud, is a recommended no-code tool that allows users to train high-quality custom machine learning models with minimal math or coding, primarily by uploading labeled data.
40 Actionable Insights
1. Focus on Problem, Not AI
Before adopting AI, identify a clear problem or pain point that requires a smart solution, rather than implementing AI just for technology’s sake.
2. Define Problem Before Implementation
Once a problem and a high-level solution are identified, then seek expertise to determine the actual implementation strategy for AI.
3. Rethink Existing Data
Shift your mindset to analyze all available data, even if it’s “just lying around,” to identify opportunities for AI-driven product improvements.
4. Start Data Collection
If your team isn’t already, begin collecting data and setting up dashboards, as this is a crucial foundational step for any AI initiative.
5. Avoid AI for MVPs
Do not use AI for Minimum Viable Products (MVPs) to validate market fit, as it’s a resource-intensive investment better suited for later stages.
6. Fake AI for MVP Validation
For MVPs, simulate AI functionality using prototypes (e.g., Figma) to gather user feedback and validate ideas before committing to actual AI development.
7. Use AI with Existing Data
Implement AI when you have existing data, either from your product or an adjacent one, that can be leveraged to create meaningful recommendations or automation.
8. Collect Diverse, Proprietary Data
To achieve high-quality and differentiated AI products, actively collect your own diverse and proprietary data rather than relying solely on generic datasets.
9. Define AI Quality for Launch
As a Product Manager, take responsibility for setting the acceptable quality bar (e.g., accuracy percentage) for AI features, determining when they are “good enough” for users to launch.
10. Understand AI PM Differences
Learn how AI product development fundamentally differs from general product management, often requiring a focus on managing the problem itself rather than just the product.
11. Bridge Research to Product
Actively bridge the gap between academic research and product production by translating research ideas into meaningful, monetizable user cases.
12. Monetize AI Research
As a PM, proactively identify and develop strategies to monetize AI research and models, converting their capabilities into viable business opportunities and user value.
13. Collaborate with Research Scientists
Get comfortable partnering with research scientists and understanding their role in developing smart models for automation, personalization, and recommendations, as this collaboration is crucial.
14. Embrace Uncertainty in Research
Develop comfort with the inherent uncertainty of AI research and development, recognizing that projects may involve experimentation, iterative refinement, and potential pivots.
15. Clarify AI PM Career Progression
When joining a research-focused AI organization, proactively clarify with hiring managers how progress and performance will be assessed, as metrics differ from traditional product launches.
16. Be a Cheerleader for AI Teams
As an AI PM, actively encourage and motivate your team through the often uncertain and iterative process of AI development, acting as the captain guiding the project.
17. Manage Change and Action
Be prepared to be the central point for managing change and driving action within AI projects, understanding that this leadership role can be tricky and challenging.
18. Be Creative in Data Collection
Be creative and willing to explore unconventional methods for data collection, even if it means directly engaging with people or finding novel sources, as acquiring good data is challenging.
19. Learn Basic AI Coding
Overcome intimidation and learn the basics of coding or training small AI models, even if not for full-time work, to gain a deeper understanding and confidence in how these tools function.
20. Engage in Hands-On Learning
Take online courses, get hands-on experience, and pair with others to learn AI fundamentals, building a foundational understanding beyond just using pre-built tools.
21. Shadow AI Researchers
If working at a company with AI researchers, reach out to them, shadow their work, and spend an hour weekly discussing their activities to gain context and identify potential applications.
22. Practice Explaining Complex Concepts
Practice explaining technical concepts (e.g., “a database”) to a three-year-old to develop crucial storytelling and simplification skills for communicating with non-technical audiences.
23. Subscribe to Newsletters
Subscribe to newsletters like “The Download by MIT Technology Review” or “TLDR” to stay informed about technology and AI trends, as AI will increasingly be integrated into all tech.
24. Explore Beyond ChatGPT
Read newsletters and online blogs to understand the broader applications of AI beyond popular tools like ChatGPT, such as lie detection or other emerging research.
25. Follow Academia & Research Blogs
To stay updated on cutting-edge AI, regularly check academic papers and research blogs, such as the website Archive, where new research is frequently published.
26. Use ChatGPT for Mission Statements
Leverage ChatGPT to rewrite mission statements by providing your initial version and asking it to enhance clarity and inspirational quality for diverse audiences.
27. Leverage ChatGPT for User Segments
Prompt ChatGPT with product ideas (e.g., “who would be interested in a fitness band that doesn’t have a screen?”) to generate diverse user segments, motivations, and pain points.
28. Guide AI with Initial Ideas
When using AI tools like ChatGPT, always start with your own initial ideas or mission, then use the AI to enhance or expand upon your existing thoughts rather than having it do the entire job.
29. Integrate Smart Features
Look for opportunities to integrate “smarter” AI-driven features into any product, such as enhancing security, personalization, fraud detection, ethics, speed, accuracy, or recommendations.
30. Utilize AutoML Tools
Explore and utilize no-code/low-code AI tools like Google Cloud’s AutoML to train high-quality custom machine learning models with minimal technical expertise, especially for image-based tasks.
31. Use Analogies for Buy-in
When seeking buy-in for big AI bets, use successful “adjacent products” or past “crazy” but successful initiatives as analogies to de-risk and illustrate the potential of your proposal.
32. Present Contingency Plans
When proposing AI investments, include a clear contingency or rollback plan, outlining the maximum negative impact to demonstrate risk mitigation and gain leadership confidence.
33. Take Stanford’s AI Course
For self-paced learning, consider taking the “Introduction to AI by Stanford” course on Coursera to gain foundational knowledge in artificial intelligence.
34. Attend Coding Bootcamps
If you prefer structured learning with others, explore online coding schools like Career Foundry, General Assembly, or Coding Dojo to learn basic coding skills.
35. Read “Inspired” by Cagan
Read “Inspired” by Marty Cagan to learn fundamental principles for creating tech products that users truly love.
36. Read “You Look Like a Thing”
Read “You Look Like a Thing and I Love You” to understand how AI works and its impact on the world in an engaging and accessible manner.
37. Read “Adventures of Women”
For women in tech, read the “Adventures of Women in Tech Workbook” by Alana Karan to gain insights and guidance on navigating the technology industry.
38. Listen to Boz’s Podcast
Listen to Boz’s podcast (CEO of Facebook) for valuable insights and perspectives on technology and leadership.
39. Try the Lensa App
Experiment with the Lensa app to transform photos into artistic styles, and consider trying the male version or the pet feature for unique results.
40. Create Your Own Courses
Consider creating your own courses to share your expertise; teaching can be game-changing for both you and your students, as people are eager to learn what you might take for granted.
5 Key Quotes
I believe that all product managers will be AI product managers in the future.
Marily Nika
don't do AI for the sake of doing AI. Make sure there is a problem there. Make sure there is a pain point that needs to be solved in a smart way.
Marily Nika
The model is like a kid's brain. It has the ability to take an input, which means it has the ability to take an image and say, Oh, I recognize what this is.
Marily Nika
AI will not replace PMs. If anything, it's going to free out time for me to do other things that are less tedious.
Marily Nika
Don't underestimate this. Try creating your own courses as well. People really want to learn what you take for granted.
Marily Nika
4 Protocols
AI Product Development Mindset
Marily Nika- Identify a problem or pain point that needs to be solved in a smart way (don't do AI for the sake of AI).
- Define a very high-level solution for that problem.
- Reach out to figure out how to actually implement the solution.
Learning to Code for PMs
Marily Nika- Take an online course (e.g., Coursera, Stanford's Introduction to AI).
- Consider online coding schools if you prefer learning with others (e.g., Career Foundry, General Assembly, Coding Dojo).
- Get your hands dirty and practice.
Becoming a Strong AI PM
Marily Nika- Understand how AI product development differs from general product management (e.g., managing the problem vs. the product).
- If at a company with AI researchers, reach out to them, shadow them, and spend an hour a week talking to them to gain context.
Building an Educational Course
Marily Nika- Treat course creation like a product: form hypotheses about the audience and their needs.
- Reach out to potential students to understand what they want to learn and what specific questions they need answered.
- Iterate on the course content and structure based on feedback (e.g., duration, focus).
- Foster a personal relationship with students by messaging them, reviewing applications, and meeting to address concerns.