AI-First thoughts? Well then, AI changes your role.
Let’s talk about why you need to think AI-first as a product manager. Every technological revolution has dramatically changed the roles of product managers, from the personal computers in the ’70s to the Internet in the ’90s and then the mobile in the first decade of the 21st century. Every tech revolution inherently changed the way we build products. And as a result, it has impacted the strategy and success of companies across the world.
Those tech revolutions force managers to learn and adapt quickly to compete in the marketplace. And those managers usually did it by dramatically changing the way they worked for years. When the mobile revolution happened, in just a few years, companies moved from desktop-only site to primarily a mobile experience, and it required everyone across the organization, from engineers to executives, to relearn how to rebuild products and rethink their strategy. AI is no different. I would even claim it’s the biggest tech revolution we have ever experienced.
Satya Nadella, Microsoft’s CEO, calls AI the “defining technology of our time,” and Sundar Pichai, Google’s CEO, was famously quoted to say that AI is “more important than fire or electricity.” To understand the role of AI in technology today, imagine a river rafting boat. The guide in the back that holds the two paddles pretty much navigates the boat. Those big paddles play the role of AI. They can dictate success and failure in your product today. That guide, then, better be you. If you lead products in your company, you have to have the knowledge and the skillset to know how to use those paddles and navigate your team and product to success.
A closer look at the role of AI in Medium.com
When you open your Medium app, you have seen that by now almost everything you see is powered by AI. AI decides which blogs to show you and how to present it to you. AI decides which people you might know and want to follow. When you share, AI figures out the right tags to help you reach the right audience. When you’re about to reply to a message, AI will show you suggested replies that are personalized to you. AI decides what topics and blogs are most relevant to you so you never miss out on a key moment. You get the drill. Almost everything you see at Medium is powered by AI. A lot of what you don’t see is also powered by AI. For example at Medium, AI powers plagiarism detection to make sure the content is original. Next time you interact with a product, any product, digital or non-digital, think of how AI might be powering your experience as well.
Let’s start with AI basics.
So now it’s time for us to learn some AI. I’ll talk about some of the fundamentals of artificial intelligence, which are the most relevant for product development and product thinking. Before jumping into the three components of AI, let’s talk about why it’s important to understand how AI works.
First, you can’t manage what you don’t understand. Let me repeat that. As a product manager, you can’t lead what you don’t fully grasp. Hopefully, that’s self-explanatory. If AI is powering your product, and potentially your company’s success, not understanding how it works means you’re not set up to be successful in your job.
And second, AI is not a magic box. For non-AI engineers, there’s a tendency to think of AI as this black box driven by some spells and some charms, and nothing can be further from the truth. For background, the term AI, artificial intelligence, was coined in the ’50s and has been researched heavily ever since. The biggest recent change in AI has not been the algorithms but rather the computing power and better proliferation enabling those algorithms to scale and reach incredible results.
So now, let’s talk about how AI works. And to simplify it, we’ll only focus on the three main components of AI that I think you should know.
Number one, the objective. What is the task you want your algorithm to learn? Remember that ultimately, AI learns to achieve a certain objective. What is your objective? It can be anything from predicting the stock market performance to knowing when to wake you up in the morning. It’s all about what you set up your algorithm to achieve by learning.
Number two, the algorithm. There are different types of AI algorithms that are used for different use cases. How will your algorithm learn? For example, the type of algorithm that is used to suggest films on Netflix is different than the one that is used for self-driving cars.
And number three, the data. AI learns from examples, both past examples, and real-time examples, and those data samples are what fuels the algorithm to achieve its objective. For example, if you’re trying to predict the stock market performance, you would need a lot of diverse data points from various sources like interest-rate increases or even the weather forecasts.
So now, only once you have these three components set up correctly, the objective, the algorithm, and the data, will you be able to build our learning program to achieve product success. Let’s dive deeper into each one of those components and learn how they work.
Set your AI objective
As a product manager, choosing the right objective for your algorithm is your most important job. It’s also the hardest job to get right. Knowing what you want your algorithm to accomplished is the most important step. Without setting your objective right, it doesn’t matter what you’re going to do next. An oversimplified view of what the objective should be can easily lead your algorithm on the wrong path. Also, small nuances in how you define your objective can dramatically change your product outcome. This is one of the hardest things to get right, and it requires a lot of thoughtfulness and a lot of experimentation.
Let’s take, for example, the ads algorithm. The algorithm decides what ads consumers will find most interesting. What do you think is the objective of the ads algorithm? Some of you might have already guessed it right. It’s to be relevant. For an ad to be successful, it needs to be relevant for the user who is seeing it. And that sounds pretty simple, right? And it is but it actually might be overly simple. Without being careful, that objective can lead to negative effects as well.
According to a recent Pew Research, consumers sometimes feel ads are becoming too creepy. Have you ever felt that way about an ad you saw? Feeling it was too intimate, almost as if you were being listened to? The reason might be rooted in the AI algorithm objective and its ability to connect the dots.
So for example, let’s say my wife and I talk about going, over dinner, to vacation in the Maldives. And later that day, out of curiosity, my wife searches for vacations in the Maldives. Batch figures a retargeting pixel from that vacation site to be associated with my wife’s Facebook profile. And as a result, Facebook now learns that my wife might be interested in vacations in the Maldives. So when I logged into Facebook the next day, the ad algorithm already made the association that I might also be interested in the Maldives vacations. Because Facebook knows we’re married and couples usually take vacations together. So as a result, I now see an ad about a vacation in the Maldives. And remember, I never told Facebook anything about my interest in taking a vacation, let alone talking about going to the Maldives. So imagine how I feel right now. While the ad algorithm was being very relevant, even intimately relevant, it also made me feel uncomfortable at the same time.
That’s why defining your AI objective right is critical. Ultimately the outcome of your objective can have significant, unexpected side effects. And later in this article, you’ll learn how to set up your AI objective right.
Deciding the right AI algorithm?
After setting the objective of the algorithm, let’s talk about the algorithm itself. How do you want your algorithm to learn? There are several types of algorithms and choosing the right algorithm depends on your use case. I won’t spend too much time here since in most cases your AI engineers are better equipped to know which type to use and how to use it. However, you need to understand the main three types of machine learning algorithms today. First, we have the supervised learning algorithm. It’s best to use this algorithm when you can classify your observations, your data, into specific labels. Meaning you know ahead how to bucket your results. For example, imagine classifying sorting emails into a spam or not spam. Or classifying different photos of birds based on their species. Second, we have an unsupervised learning algorithm. In this case, you don’t have a predefined classification and the goal is to classify your observations by identifying certain patterns.
For example, in the case of birds, you can cluster the same bird species by their shape and feather but you don’t know what type they are. This type of algorithm is very popular today. When you think of online recommendations from Amazon to Netflix, they all use unsupervised learning algorithms to discover your taste and suggest recommendations to you. The third type is what we call the reinforcement learning algorithm. This algorithm learns from trial and error. They take action and iteratively learn from the results. These algorithms are mostly used in robotics and gaming or navigation, for example in the case of self-driving cars.
One of the famous recent examples was AlphaGo, a computer program which has been given the rules of chess and then it taught itself how to play. It only took AlphaGo four hours of training using reinforcement learning, to beat world champions. Four hours of training, that’s just incredible. So now that you have some basic knowledge of the types of algorithms, let’s talk about the two aspects where you as the product manager can learn and influence and improve the algorithm itself:
The first aspect, it’ll be called features. And those are not the same as regular product features that describe the appearances or capabilities of a product. The algorithm features are the variables that help your algorithm learn and improve. Think of them as the personalization knobs that help you build a better algorithm. For example, some of the features in the Amazon search algorithm can be the price, the brand, or the shipping cost. Those can all influence the purchasing decisions of Amazon users when they come to do shopping on the site.
The second aspect you should know is the algorithm constraints. This is where you as the product manager can make strategic decisions on rules that limit the flexibility of your algorithm. For example, if you’re building an algorithm for designing meals, you should probably build a rule to make sure to avoid salt for customers with high blood pressure. As you can see there’s a lot that you can influence when it comes to the construct of your algorithm.
For example, let’s look at the following update from the Medium blog feed. What type of features and constraints would you build into your relevance algorithm for it? Take a look. In this example, for features, I might consider the likelihood of a user finding the person sharing relevant. I might also look at the topic itself, the type of content, or even the publication source. So now that you understand how to influence your algorithm, think of your products. What features and constraints are already in place? What else can be very relevant for you?
Let’s talk about the data that powers your AI.
A few years ago, The Economist published a report with this big, audacious headline, the world’s most valuable resource is no longer oil, it’s data. And that’s primarily because of AI. To intuitively understand it, think of data as the fuel that powers your AI algorithm. Once you decided on your AI objective, and then you coded your algorithm features, you’re successfully finished on the first setup. Data is what brings your AI to life. And generally speaking, the more data you have, the better your algorithm will be. Even more importantly, your AI will only be as good as the data it gets, like food for humans and soil for plants, without data, your AI will not thrive. With the right food, the right soil, your AI will learn faster and better. AI learns from examples to seek patterns.
Those could be past examples and real-time examples. So for example, when Netflix’s AI algorithm decides which movies to suggest to you, it leverages all the data available to it from past examples. This can be based on movies you saw, it can learn from how you ranked them, did you even finish those movies? It can also learn from movies that others who are similar to you in age, gender, and they select as well. The more data the algorithm can learn from, the better the suggestions you’ll see. This is especially true with new AI innovations like deep learning, which learns faster and better as you fill it with more high-quality data. AI can also learn very quickly. That’s why when you first open Netflix for the first time, it asks you to rank a few movies so they can get the AI algorithm going. Think of it as the first technician of your personalized AI engine. Every interaction a user takes is being fed back to the algorithm as data.
That’s why data can start a virtual success cycle. The more data the algorithm can learn from, the better the algorithm will be, which in turn, helps companies build better products, which in turn, drives more users to those products, which in turn, generates more data. And so it goes. That’s why data is the new oil.