Data Science Success: Knowing When and How to Make Decisions Based on Data Science Results

Last updated on Sep 21, 2021

Intuition is an over-romanticized term in our generation. While there’s nothing wrong with going with your gut, doing so may be detrimental for your business. After all, no one wants to create a Highest Paid Person's Opinion (or "HiPPO") culture in their company.

To avoid a HiPPO culture in your office, it’s crucial that once you’ve analyzed your data, you and your team are empowered to make decisions based on your results. In this article, we’ll discuss when and how to make data-driven decisions to fuel your business's growth.

You can learn more about how to design a data science project as a leader in other Edlitera posts, but let’s first examine a summary of potential outcomes that can tell you that your data science project was successful:

  1. You learned something new.
  2. You used your results to create an impactful final communication product (for example, a report, presentation, or app).
  3. You learned that your data can’t answer your question.
  4. You made decisions and new policies based on your results.

Let’s think more on that last point. If taking action based on the results of your data analysis is your goal, you need to design your project for success from the beginning. For instance, model choice is an important decision at the start of a project. For many projects, active models are better than passive models because they help avoid human decision override. When choosing your model, consider the following qualities of a successful model: look for a model that is as simple as possible, explainable, and properly limited. And when designing your project, don’t only consider the benefits of your model but take a moment (and tap into your paranoid side) to consider how your model may also cause harm if used incorrectly. 

Why is an Explainable Algorithm So Important?

Artificial intelligence (AI) tools like neural networks and deep learning have working mechanisms that humans may not fully understand. Even so, we must understand what an algorithm is being optimized for in order to successfully use it. It’s also essential to know why an algorithm made a decision. This is crucial for industries like consumer finance, health care, education, military, and government.

For leaders, the “what” question is fundamental–you need to ask this of the teams that design and build your automated solutions. To better understand the question, let’s consider the paperclip maximizer thought experiment proposed by Nick Bostrom in 2003. The story goes that an AI system is given the goal to manufacture paperclips as efficiently as possible. However, in doing so it transforms the whole earth and space into paperclip manufacturing facilities. In this example, the outlandish results are not technically the AI’s fault, but the fault of the humans who didn’t provide the right goals and constraints to the system. A similar–but real-life–example is the AI system set to create school and bus schedules in Boston. While the idea seemed like a great way to improve efficiency, the designers received plenty of complaints after implementation from parents whose schedules were not considered because the AI system was too focused on saving money. 

Let’s also discuss why it’s important to know why an algorithm makes a decision. While an AI scientist may understand the technicalities of a model, these same technicalities are still hard to explain to non-experts because machine learning makes decisions based on patterns that defeat human logic and intuition. Because AI’s sophistication makes it sometimes difficult to understand, the public is raising plenty of concerns about the potential negative impact of AI on our lives and our jobs. While I don’t think AI will take your job, I do believe it will allow you to take on higher-value tasks at the job you have. For further assurance that your work is safe, data and algorithm regulation in the future will very likely increase AI accountability, and at the same time create incentives to make AI more explainable. And as we’ve discussed, explainable AI enables us to challenge AI-based decisions. 

There is no doubt that data is a precious tool for businesses. This article has hopefully shown how understanding the “what” and “why” of your algorithm can enable you to make data-driven decisions once you have your results. So the next time you’re faced with a decision to make, you can make it the better way–based on data, not intuition.

About the author

Claudia is a data scientist, consultant and trainer. She is the CEO of Edlitera, a data science and machine learning training and consulting company helping teams and businesses futureproof themselves and turn their data into profits.

Before Edlitera, Claudia taught Computer Science at Harvard, and worked in biotech (Qiagen), marketing tech (ZoomInfo), and ecommerce (Wayfair). Claudia earned her degree in Economics from Yale, with a focus on Statistics and Computer Science.