Why So Many Data Science Projects Fail to Deliver

by James Taylor, CEO, Decision Management Solutions

If you are working in data science, you may be frustrated with the progress and ROI of your data science and artificial intelligence projects. The struggle is real, and you’re certainly not alone. Recent 
research conducted by the Alliance Manchester Business School at the University of Manchester and Ivey Business School at Western University identified five common mistakes that prevent companies like yours from getting the highest level of value from data science. The research focused on the banking sector, but it can apply to any industry. We believe this is an excellent list that represents what we see as some of the most common issues. The first step in achieving success with data science is recognizing these problem areas in your approach.

Mistake 1: The Hammer in Search of a Nail

A failure to achieve business value sometimes results from an infatuation with data science solutions. The authors suggest training and process-based fixes, but our approach goes beyond that. We have found that decision modeling is the best way to focus (or refocus) projects. Before all else, it’s critical to model the decision the data science is intended to improve. For a deeper explanation on the importance of leading with the decision, take a look at our blog, “Decision-Driven Data Analytics.”

Mistake 2: Unrecognized Sources of Bias

Models are often built using data sets that contain biases—both illegal bias, such as race or gender, and bias in terms of the customers you had targeted in the past or only customers in a certain geography. It’s critical to discern the sources of bias at the outset and make sure that the bias doesn’t find its way into models. How do you minimize bias? Bring together data scientists with business owners to discuss the central decision-making problem. And ensure these groups spend time talking to data owners too before using datasets.

Mistake 3: Right Solution, Wrong Time
It’s common to build a solution that won’t be used due to budget constraints or a change in strategy—and that is a waste of precious time and resources.
To prevent this from happening, data science needs to be in synch with the business strategies and systems of the organization. As the authors of the research point out, “Generally speaking, data scientists ought to be concentrating their efforts on the problems deemed most important by business leaders.” And to our way of thinking, this means the decisions that impact your top metrics. The best way to find these decisions is through decision modeling.

Mistake 4: Right Tool, Wrong User

A data science solution may not work because it hasn’t been integrated into user experience and processes. The researchers suggest shadowing, noting that “the use of shadowing in data science projects contributes to a better understanding of the processes that generate data, and of solution users and delivery channels”. We would add that, once again, decision modeling comes to the rescue. Decision modeling helps you capture what you learn when you shadow decision-makers. And because decision modeling is part of a focus on continuous improvement, you can keep shadowing and fine-tune and adjust your model along the way—you don’t necessarily have to solve the whole problem in the early stages.

Mistake 5: The Rocky Last Mile

In the home stretch of data sciences projects, there’s often a disconnect between data scientists and the expectations of the business team. This can occur for any number of reasons and can result in perpetual pilots, failed projects, long project cycles, and minimal business value. The solution is to ensure that the data scientists really grasp the business value of the project they are working on and think in terms of building solutions to the whole decision-making problem. It’s up to leadership to broaden the role of data scientists and encourage them to coordinate with other employees who are responsible for problem diagnostics, process administration, and solution implementation. These groups can collaborate around a shared decision model and deliver value using a mix of technologies—not just data science technologies—that will enable faster implementation and time to value.

 

We hope you avoid these five mistakes in your data science projects. If you’d like to learn more, attend our upcoming webinar, “Three Things to Maximize Machine Learning ROI” on Wednesday, May 12, at 10:00 am PST, 1:00 pm EST. Register today!