The hype around artificial intelligence (AI) and machine learning has led to lots of jargon, so that this very powerful technique has become more difficult to understand. The tips below have all helped me, so I hope this article will help product managers to cut through the noise and better understand how AI can fit into their daily work.
A Broad Definition Enables Better Problem-solving
Let’s forget the buzzwords for a moment: what is AI anyway?
In my experience from working on data-driven products, I’ve found that a broad definition of AI helps me to focus on defining the problem I’m trying to solve, rather than fixate on specific techniques to use in the solution.
As product managers, we must thoroughly understand our problem space so that we can properly define requirements and allow our team to solve the right problems. This often leads to the topic of “how” coming up prematurely; before we know it, we have accidentally biased ourselves towards specific solutions before we’ve properly defined this problem.
To help avoid this, I prefer to define AI as “automated decision-making”.
Most products we work on require decision-making based on data, though the method for making decisions can vary. For example, decisions can be made by machines or humans, and the data can be static or dynamic. A focus on decision-making abstracts away the intricacies of specific methodologies or the noise of industry jargon. This broader definition enables product managers to be more attentive to the problem space – it removes distractions that can lead us to think about solutions too early in our process.
AI is a Tactic That Helps to Solve Problems
There are three key concepts that serve as the foundation of everything we do as product managers:
Vision: The end goal we aim to achieve
Strategy: Doing the right things to realize our vision
Tactic: Doing the things right to properly execute our strategy
Depending on the product lifecycle, product managers need to operate on all three of these levels at any given time. We must align our teams to the vision we are trying to achieve, while ensuring everyone understands the strategy and how our daily, operational tactics fit into the overall plan. It’s important to remember that AI is a tactic that can be used to solve specific problems rather than a strategy or vision. Deploying AI without an end goal usually brings no value to end users.
To give a concrete example, here is how the application of AI might fit into a team at Netflix, one of the first companies to effectively productize AI at scale:
Netflix Vision: Become the best global entertainment distribution service
Netflix Strategy: Drive member retention via engaging, personalized UX
Netflix Tactics: Ratings system, recommendations, personalized hero shots, usage tracking, etc
As you can see, there are a number of tactics within the personalization strategy that could be used to achieve the goal of improving member retention. The degree to which data and AI are used varies from tactic to tactic, while the vision and strategy statements abstain from dictating which technologies or algorithms must be used.
AI can Empower Humans, Rather Than Replace Them
The current discussion on topics like automation has raised some interesting ethical questions about the future of work, and subsequently, the narrative around how AI can empower humans has become a bit lost. A common example that illustrates this point is self-driving cars. Within the industry, autonomous capabilities for vehicles are classified into five different categories, with much of the conversation fixated around what would happen in a world where cars were fully self-driving (level 5 autonomy).
It is important for product managers to recognize that AI capabilities are typically developed in stages over time, rather than turned on instantaneously. Machines are good at different types of tasks from humans, so certain decisions are easier to automate than others. High-performing AI capabilities require a sizable training dataset to get started, and training datasets need to be well-structured high volume, and machine-readable. Ideally, the dataset should also have well-defined notions of success and failure, where past outcomes are predictive of future results. Here is a framework that I often use when considering how to apply automated decision-making:
Along the y axis, routine scenarios happen in high frequency and have low variability in how they unfold, while nuanced scenarios seldom occur and could contain hard-to-replicate subtleties. On the x axis, informational decisions provide additional context to the end user, while action-oriented decisions perform an action on behalf of the end user. Routine scenarios tend to generate more reliable training datasets and are therefore easier for machines to learn; informational decisions tend to be lower risk than action-oriented decisions. Combining these two dimensions yields four categories of automated decisions:
- Routine information: Easy to predict and low risk if wrong
- Example: Car estimating remaining driving distance based on remaining fuel and driving behavior
- Nuanced information: Hard to predict and low risk if wrong
- Example: Car warns when the driver is falling asleep based on image recognition and driving behavior
- Routine action: Easy to predict and high risk if wrong
- Example: Car autonomously driving on highways under normal conditions
- Nuanced action: Hard to predict and high risk if wrong
- Example: Car autonomously driving through busy construction zones
AI can Drive Impact in Three Different Ways
From working on data-driven products over the last decade, I’ve identified three primary buckets of use cases for how product managers can drive impact with data:
AI can optimize and/or automate operational processes. Products with proper behavioral tracking can generate a dataset that empowers teams to make more informed decisions about how to run the business. For example, customer touchpoints and communications can be optimized based on data in order to increase conversion or reduce churn. Support requests can be triaged or routed more effectively based on the predicted topic or outcome. In this sense, AI serves as an advanced business intelligence tool that drives productivity and effectiveness for teams.
AI can dramatically improve the user experience of products. Here are some examples of how some companies have been able to use AI to create delightful experiences for their customers:
|Brand||Product||AI-enhanced UX Example|
In each of these examples, the fundamental product delivered to end users remains the same (for example, mobility via Uber), but the experience around the product is better with the application of data (being matched to a nearby Uber driver). Using this pattern, product teams can often create unique user experiences that become long-term competitive advantages for their company.
AI can fundamentally change products themselves. Perhaps the most famous example of this is the story behind Netflix’s “House of Cards” series, where the usage of data redefined how entertainment is created. This series not only won many awards, but it is also loved by Netflix subscribers. It also marked the beginning of a new era of significant growth for the company. It shows that AI has the potential to create new categories of products and new trends for an entire industry.
To sum up, here are four ground rules for how product managers might think about integrating AI into their daily work:
- Defining AI broadly frees us to focus on problem-solving rather than the end solution.
- AI is a tactic that helps solve problems, not a strategy or an end-goal.
- AI can empower humans, rather than replace them.
- AI can drive impact in three different ways: optimize operations, improve product experiences, and create new product categories.