Supervised vs. Unsupervised Learning: Why It Matters for Executives

This series empowers you, the executive, to make informed decisions that drive AI initiatives, not by becoming technical experts, but by leveraging your critical thinking and business acumen. After all, these qualities are crucial for AI success in the business world.

In previous chapters, we discussed the importance of setting SMART goals and securing relevant, unbiased data . Now, let's delve into two key machine learning (ML) approaches: supervised learning and unsupervised learning. Understanding their differences and applications empowers you to make wise choices for your AI projects.

Supervised learning acts like a diligent student, learning from labeled data where each point has a predefined "correct" output. Imagine training a model to classify emails as spam or inbox. You feed it labeled examples – emails already categorized as spam or inbox – acting as the ground truth for the model to learn from. By analyzing these labeled examples, the model identifies patterns that differentiate spam from legitimate emails. Once trained, it can classify new emails with a certain degree of accuracy.

Think of unsupervised learning like a curious explorer searching for hidden patterns in unlabeled data, where the desired outcome is unknown. Here, you're asking AI to uncover valuable insights that might be impossible to find manually due to data size and complexity.

Let's illustrate this with a real-world example: IT incident categorization. Traditionally, organizations use manual, error-prone methods to classify incidents into pre-defined categories and subcategories. This is where supervised learning shines. By providing the model with enough labeled incident examples (training data), it can learn to automatically classify unseen incidents. The beauty? You can measure its success easily, like how accurately it classified 95% of incidents in a test set.

However, this high-level classification lacks the granularity needed for automated resolution, problem identification, root cause analysis and skills-based ticket routing. This is where unsupervised learning takes center stage. Imagine AI analyzing hundreds of thousands of laptop, network, or e-commerce website issues, automatically grouping similar incidents together (a.k.a 'clusters'). This unveils valuable insights, but unlike supervised learning, there isn't a single pre-defined 'correct answer.'

As demonstrated below the list of laptop incidents can be grouped into similar clusters in more than one way. The table shows two options and if you needed to analyze these incidents you would probably come up with a third or fourth option.

While unsupervised learning doesn't have pre-defined "correct" answers (otherwise it would be supervised), your subject matter expertise becomes crucial in evaluating quality and success. Remember, the ultimate goal is to achieve your business objectives. Ask yourself: "Are these unsupervised results good enough for the task at hand?"

Interestingly, the chosen approach (supervised or unsupervised) can depend on how you phrase your business needs. With the right understanding, you can define your goals to lead you towards either a straightforward supervised approach or a more open-ended unsupervised process.

Remember, unsupervised learning unlocks possibilities, not definitive answers. Your business judgment and understanding bridge the gap between the model's findings and real-world value. By understanding these key ML approaches and their strengths, you can be more accurate in how you define the goals and success criteria, actively shaping the AI journey of your business.

In the next chapter we'll explore Machine Learning vs. Generative AI : Understanding the difference and choosing the right tool for your goals.

Feel free to comment below and share your experiences! Your contribution to the discussion and knowledge sharing is valuable.