I recently joined an AI company as a Product Manager and I’m having a blast. What do Product Managers do at AI companies and how are our roles different?
The answer, as many PMs would immediately respond, is that it depends on the product.
Short Answer:
AI Product Managers design and build products around datasets and their inherent attributes and distributions. We design products to solve problems, by thinking about what data to feed our AI models.
What is an AI product ?
If a restaurant adds online delivery service via a mobile app, we would not consider it an internet company. Similarly, if the same restaurant then adds an AI chat-bot or integrates with Amazon Echo to allow customers to place orders by shouting at it, we would not jump to label them as an AI company.
Simply put, an AI company is one that develops AI technology as a core competency. We can extend this criteria to AI products.
A consumer travel app that leverages Google Translate, while enhanced with AI super powers, is probably fairly light on AI development and would not likely develop custom translation models.
A consumer travel app that looks at an exotic plate of food and tells you the name in both your native and its local language, may be developing custom AI models.
The AI discussions for many PMs today center around strategy for leveraging available solutions. Not every product needs custom AI models (and that’s fine); those that require its teams to invest and develop AI competency will be organized differently.
Roles and workflow
Teams that do build above mentioned custom AI models optimize for different goals, and naturally have different roles and workflows, such as:
• Data collection and management
• Machine learning and model development
Depending on scale of product, the Engineering team may also have additional roles or workflow for:
• Model testing & QA
• Production model deployment and monitoring
Typically, PMs at web and mobile product teams work with Design and Engineering members to roll out new products/features. In contrast, AI PMs may coordinate with 2 additional roles: Data team and Model/Algorithms team.
Working with data at the core of product is not unique to PMs at AI companies. In fact, this may sound very similar to Data Scientists (or PMs that work with them), with a very key distinction; data is used to build AI models, a core part of the product itself. Data science may be a best practice for data driven decision making, but its outcomes are not itself part of a product.
Designing AI products
Industry leaders today have world class AI labs in addition to deep learning research and engineering teams across various business groups. Deep learning is still a relatively new technology and the industry as a whole is still figuring out best practices. It is clear however, that deep learning AI models are best developed with optimal datasets and designed with laser focused goals.
Let’s think about the travel app in the earlier example. We want to build an deep learning model to recognize images of food, and tell us the name of the dish.
- Be specific; who are the users, and what data would they feed into the AI models?
Our users in this case, are regular tourists traveling to exotic places where they might encounter new and exciting looking food. We might assume that users would take pictures of their meals with their smart phones. Our users might come across interesting looking dishes on social media, and want to find out what it is. In this case, it would be important to build dataset with both professional and user generated content, from foreign restaurants such as Hispanic or Asian cuisines.
- Not all datasets are built the same; depending on how you build your dataset for training AI models, you may introduce certain bias to your models.
One assumption that we might make is that the photos are foodie shots, where the dish is sole subject of the image. What about selfie shots with users eating the food? Or images with multiple dishes on the plate? What about beverages? Datasets have inherent attributes and distributions that may make an AI model more or less effective.
—
PMs all think about users and how the product needs to solve a particular problem. AI products look to solve this with deep learning AI models. AI Product managers are responsible to think about the specific task the AI model needs to solve, and what relevant data to feed it.
