Several studies and surveys show that businesses are ramping up investment in data science. Yet it’s been estimated that most data science projects fail. The consequences of a failed data science project aren’t pleasant, with money, team morale, and competitive edge on the line. It’s clear that nobody sets out on a pricey project expecting it to crash and burn, yet it’s not an uncommon experience.
In this blog, we explore some of the key reasons why data science projects fail, and how to avoid them. Before we dive into it, it’s crucial to note that many of these reasons are interlinked and sometimes a result of a domino effect from neglecting an area or consideration earlier on in the project. With that in mind, we’ve tried to lay them out in the rough order of occurrence.
Reason 1: The project doesn’t line up with the business strategy
The first big reason why data science projects fail can be found in the disconnect to business strategy. With a lot of buzz around data science and AI, some businesses get caught up in the desire to capitalise without considering whether the solution will truly line up with the business’s strategic and commercial goals.
Here it’s important to outline how the project contributes to the bottom line. What KPIs will it drive? If it’s a new product feature, is there demand for it? You may need customer research to validate your idea.
Data science can be used to solve business problems, and successful projects start with the problem, not with the vehicle to overcome it.
Reason 2: Lack of buy-in and direction
This point links back heavily with the previous reason. When the advocates for the project are unable to identify the value it will bring, they struggle to get senior buy-in. This means that the leaders of the business won’t consider the project a priority, dampening momentum and causing issues with resource allocation. Embarking on a data science project requires time, commitment, and effort, and without these, even a good idea won’t materialise, let alone produce the desired results.
Reason 3: Lacking quantity or quality of data
As the name suggests, data science projects require data. Even if you have a brilliant idea that’s well aligned with the business strategy with full senior leadership backing, the project will fail if the data is poor or insufficient. Let’s take prediction models as an example; if there’s not enough data or it’s inconsistent, the model will struggle to produce meaningful insights.
Luckily, these issues can often be solved. If there are gaps in the data, you may want to reassess your processes to fill these gaps. If you have insufficient data, you may be able to buy more. Sometimes, you can even apply more machine learning capabilities to actually enhance the data that you have!
It’s critical to assess and address the quantity and quality of data before embarking on a project. Committing to a model before fixing data issues can become a costly uphill battle and can ultimately cause the project to fail.
Reason 4: Not having the right skills in the team
Team constitution plays a huge part in the success of a data science project. If you don’t have the right skills in the team to achieve the goal, failure becomes more likely. This spans further than having skilled data scientists, other team roles can be equally crucial.
While a project’s success requires the hard skills that data scientists bring, it also requires a person who owns and steers the project and can offer the sometimes-needed external perspective.
If communication and project management are weak, projects can lose momentum, and cracks start to show. This, in turn, is substantially related to the risks that come with the lack of buy-in and resource allocation discussed earlier.
Data science projects can be expensive to execute, and getting skilled talent comes with a price tag, but a failed project can cost the business much more. Investing in the project, all the way from your team constitution to the technologies used are vital components of a well-functioning data science project.
Reason 5: No infrastructure to maintain the solution
Data science solutions are often built on human behaviour, which is not a constant, and we cannot always foresee how it may change. It’s crucial for the long-term success of a solution that the business embraces an adaptive approach, and this should be understood from the start of the project.
Viewing a data science solution as a one-time exercise is a mindset that can set the project up for failure. Data science solutions require maintenance and upkeep to ensure the longevity of desired outcomes. This means that the appropriate infrastructure should be in place, even after the project is complete. Even though a solution has been put in place now, the wants and needs of the product can change over time. That’s why revisiting and reassessing is vital to keep it up to date. In a nutshell, the best results require a commitment to provide the right infrastructure for the long term.
Conclusion
It’s clear no two data science projects look the same, but we can identify some common reasons why they might fail. For a project to succeed, it must be aligned with the business’s overall strategy, and have senior management buy-in. Only then will the project warrant the skills and resources it needs to succeed. Data quality and quantity should also be considered from the outset, along with creating the right infrastructure to maintain the solution.
Thanks to Lead Data Scientist Tina Urba for her vital contributions to this piece.
