Five reasons why your data science project can fail, and what you can do about it

A Gartner study of 2013 says over half of all data analytics projects will fail—either because they are not completed within budget or they miss to deliver the desired benefits.

Afrozy Ara Oct 24th 2017 A-A+

The rise of data science is largely driven by the rise in volume and variety of data and our ability to access it—data being the key component driving digital transformation. However, developing data science as a capability within enterprises has its own unique challenges.

According to a study by Gartner in 2013, the worldwide market for analytics will remain the top focus for CIOs through 2017; but over half of all data analytics projects will fail—either because they are not completed within budget or they miss to deliver the desired benefits.

The ability to recognize some of the common mistakes, or ‘predictors’ of failure can put CIOs in a better position to lead these projects.  Here, we take a look at five common reasons why data science projects fail, and what you can do to avoid it.

1. Losing sight of the ‘big’ picture

As data scientists, our job is to extract signal from the ‘noise’ hidden in data, which will impact our business. However, an obsession with accuracy and depth—at the cost of breadth—can cause data science projects to fail as it makes one unable to zoom in and out of the problem being solved. For instance, if you are building a model to forecast product sales, you may get 85% accuracy in a month but moving from 85% to 90% could take you 6 months.  Does the marginal benefit of higher accuracy justify the time and effort spent? To find the right balance of depth and breadth, get constant feedback from your stakeholders. Engage them early to ensure that your recommendations generate value for them.

2. Lack of engagement with key stakeholders

The right people, processes and culture is a bedrock for building viable frameworks and infrastructure for data science. If you hear quotes like – “We have some data science folks but no one really knows what they do”, then it signals a lack of executive sponsorship and engagement with stakeholders.  Hiring data scientists is not enough. This talent needs to be integrated into the existing organization and new structures that enable value creation, as required. One approach could be for data science folks to get involved with business units so that they share responsibility for BU performance. Whatever approach is taken, success is driven by executive sponsorship and buy-in from senior management.

3. Putting the ‘how’ before the ‘why’

When you solve a problem, address the ‘what’ and ‘why’ first. In fact, problem formulation meetings that begin with ‘how’ become inherently short sighted. This can happen when you have a team of bright and technically inclined folks who want to try out the latest tools and technologies at the word go.  To avoid this, question your objective definition to sharpen the ‘why’ before moving on to the ‘how’. For example, instead of articulating your problem as, “We want to use Spark MLlib to build an ecommerce recommender system”, consider defining it as, “We want to build an ecommerce recommender system to improve the customer experience on our website”.  Focusing on the ‘why’ will ensure you are aligned to the desired outcomes.

4. Not solving the right problem

My favorite quote from the statistician John Tukey captures the essence of this issue perfectly: “An approximate answer to the right problem is worth a good deal more than an exact answer to an approximate problem.” This challenge manifests itself in many forms: you don’t know what problem to solve, or how it fits in the overall scheme of things. Sometimes, you’re mandated to fix a specific near-term problem. However, not understanding all the dimensions can result in band-aid approaches, where you miss the opportunity to address strategic underlying issues. Avoid this situation by asking for a use case, getting diverse opinions and questioning the objective. You need to identify the right problem, solving which will result in the desired outcomes.

5. Looking for data scientist ‘unicorns’

Data science is a field which requires an interdisciplinary skillset—you need to be good at math and statistics as it provides a foundation of methods to analyze and interpret data; domain knowledge is required to understand data and business processes; and coding is a prerequisite to convert theory to action. It is hard to find one person who has all the capabilities, especially with data scientists in such high demand. Instead of struggling to hire them, you can build diverse teams. Eventually, data science is a team sport and constructing a team with a strong combination of these skills will help you lead with analytics on the path to digital transformation. 

In some cases, mistakes are inevitable. As Murphy melancholically puts it, "anything that can go wrong will go wrong". Having said this, enterprises cannot afford to disregard bad practices. It is vital to track every data science project, learn from mistakes as early as possible, and avoid them in the future.

The author is Head of Data Science Practice at Incedo.

Disclaimer: This article is published as part of the IDG Contributor Network. The views expressed in this article are solely those of the contributing authors and not of IDG Media and its editor(s).