Why Most Data Science Projects Fail
Simply hiring a team of data scientists does not automatically result in a top-performing data team or produce great business outcomes. This is one of the top misconceptions that enterprises have when they set out to harness the power of data, notes Nicolas Paris, head of data at CloudCover on a call with CDOTrends.
Paris speaks from experience, having worked for a decade in various data science and business intelligence fields before becoming the CTO of a healthcare startup implementing data-centric SaaS projects. Today, he helms a team at CloudCover developing cloud-based data solutions designed for top startups and large enterprises.
Support needed
On a positive note, enterprises now understand that they can significantly improve the bottom line with data science and are hence treating their data initiatives seriously. Why then, do so many data science projects fail?
“To derive value from data science, enterprises must look beyond the data models and technology. Instead, they must think deeply about the business outcome of their data science projects. And that is actually very difficult,” explained Paris.
In a nutshell, data science is merely a tool that must be applied correctly to achieve the desired results. Moreover, Paris says data scientists need access to clean and organized data to even begin. In some cases, this might necessitate the hiring of data engineers or at least supporting them with access to repositories of curated data.
“Data scientists need time and support from business executives to validate and formulate use cases. You cannot expect an isolated data scientist to come up with relevant use cases all by himself or herself,” he added.
Hearing from Paris, success can only happen with the active participation of key stakeholders: “Business leaders and departmental heads must be willing to engage with their data scientists, and this engagement must happen on multiple levels of the organization. Crucially, the example and direction must come from the top. Only when this happens will leaders and heads be willing to collaborate with the data department.”
Winning with data
Apart from consciously engaging with data scientists and supporting them on the data front, how else can enterprises position themselves to maximize their benefits from data? It turns out that patience matters, too.
Establishing data competency is a process that takes time, observes Paris, “Working on data is hard work. Success is not going to happen overnight. When organizations kick off their data initiatives, they are really building up and enhancing their capacity to work with data. A ‘get-rich-quick’ mentality is never going to work, which unfortunately is the default mindset of many enterprises.”
An effort must also be made to establish a data culture for long-lasting success. “There needs to be KPIs or incentives to reward data collaboration and successful data-centric projects. Because if there are no benefits or encouragement to use data, people are going to stay stuck in their ways.”
And while data had historically been a cost center, it is on the cusp of becoming a revenue center. How can that happen?
“[Data can be a revenue center] by leveraging it to empower sales or product development. When companies talk about digital transformation, what they often mean is: ‘Let’s improve the customer experience’. And that is one of the key promises of data – using data to create products that are more compelling than the competition.”
Tools first, or people first?
Finally, which should enterprises start on first: Hiring data talents or acquiring data tools? They should ideally be done at the same time, says Paris: “People that you hire must be able to use the tools. If you buy the tools first, then your use of the tools will likely be constrained.”
“We had clients where the data tools were acquired at a different point in time from the hiring; they found themselves in a bind because their data professionals were unable or unwilling to work with the tools,” he cautioned.
“One such case that I know about happened recently, and the data science team was let go because they could not work with cloud-based data tools.”
Paul Mah is the editor of DSAITrends. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose. You can reach him at [email protected].
Image credit: iStockphoto/koldunova_Anna
