
It is observed that more than 80% of the data science projects fail and never deliver the return on investment (ROI) for the business. What are the actual reasons behind this failure and how can we overcome it?
The high failure rates
When data scientists talk with each other related to the number of successful data science projects that they have done in that past, a recurring theme comes up. They ask themselves, which of the data science projects made it through the deployment and are used by the company till date, and which are the projects that failed?
I think most of the people in the data science industry knows that only a fraction of them ends up making a difference for the employer.
According to a recent Gartner report, only 15% to 20% data science projects are completed. Out of those projects, CEO says that around 8% of them generate value. If these figures are correct, then this would amount to an astounding 2 percent of success rate.

The image created above casually depicts the actual success rate of data science projects in making a difference to the problems or issues faced by the organizations.
What is the root cause of failure?
So, where the problem lies?
If you talk with the data analyst, you might hear,
I developed a wonderful data model, and it had accuracy as well, but surprisingly nobody use it?
The business stakeholders and executives were hard to contact and making them engaged.
On the other hand if you talk with the stakeholders, they will say,
the data scientist made a brilliant model, and we were impressed by their qualifications, but it doesn’t answer our question.
Possible Reasons of Failure
On the corporate side,
- It is normally seen that there is an expert of data science among the business employees, but that person had a hard time to get traction with the executives to bring in the changes that data scientists recommended.
- The person who initiated the project move on to some other part of the organization and their successor is not interested in the existing project because he will not get credit of it in any way.
- Communication has broken down as the business stakeholders were too busy in their corporate work. Once they are free and can take on the project, it’s hard to rescue it. This situation arises a bit more when data scientists are geographically apart from each other.
- Data science projects are a long term. In that time, either the business may have changed their direction already or has lost patience waiting for an ROI.
- Although the stakeholders were interested, the person whose sign off was needed was never interested in the project and the fruitfulness it may bring to the table. This often happens in large organizations with overly complex hierarchies.
On the data science side,
- The data scientist has lost focus on the project and spent too long experimenting with models as if they were during his tenure as a student.
- The data scientist couldn’t communicate his findings to the right people.
- The data scientist was following the wrong metric all along.
- The data professional didn’t have the right tools and methods to solve the problem.
How can we stop data science projects to fail?

It is imperitive for the business stakeholders and data science team to sit together and convert their project in a series of steps so that the communication doesn’t break in between them.
- Business Question:
First, all of the stakeholders should work on the business question instead on focusing on technologies and methodologies. The data scientists and business executives should spend time in formulating what is the question that they want to resolve. - Data Collection:
After formulating the question, data scientists should work together as a team to analyze as to how and which relevant data should be used to solve the business question. Additionally, if they are unable to to find the relevant data, then they can purchase 2nd and even 3rd party data sets available on the internet. - Back to relevant stakeholders:
Furthermore, data scientists need to present the initial insights to the stakeholders so that the project can be easily scoped, and we can establish what we want to achieve through the project. At this point the business stakeholders should be thoroughly involved in the project and it’s the duty of the data scientists to help making them understand that what will be the ROI if the project succeeds. However, if the decision makers are not properly engaged then it would be a waste of money to continue with the project. - Investigation Stage:
Now it’s time for the data science team to proceed with the project. It’s important and critical for the data team to have at least one weekly meeting with the main stakeholder, and slightly less regular catch ups with the high-ranking business executives whose support is compulsory for the progress and success of the project. At all points, both parties should ensure that the project is heading to the right direction and towards the ROI which was mutually decided in the earlier phase of the project. - Presentaion of Insights:
Finally, at the end of the project, data scientists should present the insights and recommendations to the relevant stakeholders. You can use as much materials as you can: produce of presentation, video recording, source code to be presented, a white paper or in any other forms to the stakeholders, so that all the technical data is readily available from the stakeholders to the CEO.
If the above steps are followed, the point of value related to the project should be clear to the high-ranking executives. The frequent two-way between the stakeholders and the data science team ensures that not only the project will be successful, but also keep the data science team on track to deliver value by the end of the project.