machine learning basics

Identify Your Machine Learning Opportunity: The Basics of Getting Started

machine learning basics
Share on facebook
Share on twitter
Share on linkedin
Share on email

The promise that machine learning can transform a company is real, but understanding the basics of getting started and succeeding requires a different approach than your traditional software development projects. The reason is simple, really: Make sure the math checks out before you commit to spending your money.  Here are two key things to consider when first trying to grasp the basics and identify your machine learning opportunity:

Identify the Best Business Use Case

The most important step by far is to identify the right use case from the business perspective.  It’s best that you start small, but I guarantee that if you get this part wrong, you will have failed before your project has even begun.

When identifying the use case, keep the discussion about the business opportunity, not the technology.  Understand the processes and meaning behind the statements.  You should be asking “why” repeatedly until you grok the perspective of the business.

In the beginning, I like to network with an executive who can act as a liaison to connect you with the right individuals in a company.  It’s these roles that are critical to the identification of the use case because they are the ones who face it every day.  They know the details and, to find success, you’ll need them in your corner.

Expect Bias and Keep it Human

I like to recognize two types of ML opportunity: revenue-driving and cost savings, or risk reduction.  It’s important to be aware of the different types because of the internal challenges you will face—after all, it doesn’t take much for progress to become derailed. Understanding how to navigate the politics and pushback plays an integral role in why projects either fail or don’t gain momentum.

Revenue-driving opportunities are typically an easier sell for two main reasons: they require less operational change, and they have a more straight-forward value proposition. Predicting customer acquisition, using machine learning to provide sales forecasts, and even predicting the right healthcare service line for patients are just a few examples of such revenue-driving opportunities.

By contrast, cost savings or risk prediction is usually the more difficult path because it requires automating a task that will alter someone’s job. Automating back-office processes and predicting disease based on image data are two off-the-cuff examples that will alter the jobs of finance or HR departments and physicians respectively. There are a few takeaways here: expect a natural push back and a strong dose of bias. It’s natural for us to resist change, therefore you must be intentional and consistent in your communication that automation is about a task, not a job.  If you are successful, you will have elevated roles and have gained allies for your next adventure.

Once you have identified a few candidate opportunities, you’re now at a place where you can move beyond the basics and start asking more technically oriented questions.  You can begin to understand what data is available, where it comes from, its quality and other factors.  All of these are inputs and contribute to framing the right problem to solve.  This can unlock exploration and preliminarily modeling which in turn informs the right decision to make.  It’s this due diligence that yields the evidence for moving forward with a larger investment.

Want to learn how to take the next step with your data? Check out Predictive Analytics Discovery and learn how we can help.

Stay Up-To-Date