Separator

Why 60 Percent of Machine Learning Projects are Never Implemented

Separator
Nir Kaldero, Head - Data Science, Galvanize Nil is a tireless advocate for transforming education & reshaping the field of data science. His vision & mission is to make an impacton a wide variety of communities through education, science, and technology

We’re all familiar with the concept of the Industrial Revolution, but change has never stopped. We’re now well into the Fourth Industrial Revolution and it’s transforming the way businesses function and grow. Indeed, companies that fail to transform into data and model-driven organizations are doomed to failure. This is especially true of incumbents that risk being left in the dust by disruptive upstarts, companies comfortable with a data and model-driven approach from their inception.

As a brief primer, the First Industrial Revolution was about steam and railroads, the Second about electricity, and the Third brought about by the Internet. The Fourth Industrial Revolution is based on artificial intelligence (AI). The transformation it brings will be bigger than that any previous revolution has brought about. There are three fundamental pillars of AI adoption data, technology, and people/culture/process. In my experience working with many of the largest organizations in the world, I see a consistent pattern: they are willing to invest in cutting-edge projects with the power to revolutionize their organization, but they struggle to operationalize these projects. Approximately 40 percent reach implementation, while 60 percent stall or flame-out.

Why is this? It’s because business leaders do not pay attention to crucial change-management processes that need to be considered from the very beginning. If you don’t think about how to operationalize machine intelligence at the start of a project, you won’t be able to transform your business or realize ROI. As an executive and a leader, a big part of your role is managing the change that must, by necessity, take place if your organization is to embrace data science applications. Change management is the vital step in operationalizing data science projects and transforming your organization into a data and model-driven enterprise.

Four Ways to Solve Change Management Problems
Change management can be daunting. Fortunately, the issues I see are consistent, as are the solutions. In this section, I’ll give you a brief summary of the most common steps organizations need to take to solve their change management problems:

1.Business & Technical Teams Must Work Together: These areas of a company are usually home to very different types of people, with
different talents, priorities, and background. For successfully implementing and operationalizing data science, leaders need to build and nurture an environment where these two distinct types of people can work together seamlessly.

2. Lead from the Top Down: As an executive, you must understand that change comes from the top. Without active leadership on your part it’s highly unlikely that projects will move in the most productive direction. You need to be responsible for aligning the entire organization with your vision and guiding projects toward operationalization and the realization of ROI. It’s all about your people, the culture you cultivate, and the processes that will help your organization to best respond quickly to this game-changing technology.

At this stage in the progression of the Fourth Industrial Revolution, standing back and waiting is no longer a good option. It’s important to start taking advantage of data science and become a data and a model-driven organization


3. Allow Enough Time to Adopt Changes: Adopting new initiatives takes time. Many organizations fail to allow enough time to undertake such major changes successfully. To implement a data science project effectively, the typical enterprise must start working with their IT team and software vendors eight months prior to implementation. This is why you should think about the change management and processes required for implementation at the beginning of the project.

4. Get Strategic about Operationalization: To ensure that projects reach implementation, it is vital to stay nimble and create strategies that allow for unexpected scenarios. Too many organizations start out unprepared for the system enhancements and other operational factors that can derail a project. When these obstacles occur, they are too willing to drop initiatives entirely, rather than adapting to the new circumstances. Always think about these anecdotes and create short term strategies for implementation.

Implementation Isn’t a Data or Technology Problem, It is a People Problem
At this stage in the progression of the Fourth Industrial Revolution, standing back and waiting is no longer a good option. It’s important to start taking advantage of data science and become a data and a model-driven organization. If you engage in a new data science initiative however you want to do so successfully. You don’t want your project to be one of the 60 percent that never makes it to implementation.

You already have access to a tremendous amount of data. It may need to be cleaned and labeled, but it exists, ready to be put to use. Chances are that funding for large technology initiatives is readily available if ROI can be proven. Indeed, technology updates can be highly cost-effective because they move organizations from large expensive providers to open source providers.

The real issue preventing organizations from embracing the technology and becoming data and model-driven is management. In the 21st century, change management involves more than business leaders and executives. Technical people now have a prominent and integral role in the process. Fourth Industrial Revolution change-management processes involve both business and technical teams.

Most non-technical business leaders are not familiar with technical personas, their vocabulary, or their way of working.Realizing the ROI from implementation of machine intelligence models requires a paradigm shift and an understanding that the work of the technical team is integral to the implementation of machine learning initiatives. When you as a leader understand this, you will be in a position to take the other steps described above. Instead of allowing projects to drift, you will lead. Instead of rushing adoption, you will allow enough time. Instead of allowing operationalization issues to catch you unawares, you will expect them. And you will put your organization in a position to thrive, not wilt, in the Fourth Industrial Revolution.