Best Practices To Succeed With Data Science

As we noted last month, the presence of data scientists in a room does not automatically translate into great business outcomes. Insufficient cross-organizational support, a poor data culture, and unsuitable tools are the top reasons why data science projects fail.

What are some best practices or steps that enterprises can adopt to ensure that they succeed with data science? Having highlighted the advice of Rama Ramakrishnan, a professor at MIT Sloan on leading successful data science teams previously, we now pose this question to practicing data science leaders in Asia.

Define the outcomes

According to Garrett Teoh, the senior director of data and analytics at Capgemini Invent, success starts with defining objectives with clarity. This should include business goals, challenges, opportunities, and areas of improvement, and cover how a particular product or service offering can be improved or enhanced.

Success metrics to address the objectives or business challenges must then be defined. This should be explicitly outlined, too, such as the percentage improvement in customer conversion, productivity increase, or time reduction to deliver an outcome. As you can imagine, this necessitates the availability of baseline metrics, and organizations that do not already have them might need to first take the time to obtain them.

Businesses must also make the effort to evaluate how the proposed solution integrates into respective business units, says Teoh. “Is there a way to automatically scale these insights across the organization and how does the solution evolve over time?”

Get the right data talents

Sourabh Chitrachar, the regional director and vice president of IT Strategy and Operations at Liberty Mutual Insurance, thinks that internal alignment and a clear understanding of the business strategy should form the foundation for one’s data strategy.

He noted that the biggest challenge is finding data scientists with domain knowledge. To address this, Chitrachar suggested that organizations first identify existing talents, preferably within business teams, with strong analytical skills who are already engaged in generating management information reports and train them in data skills.

“This is a long pathway, but it achieves two objectives: It addresses the knowledge gap in terms of domain knowledge and provides a career opportunity for your internal talent while making them more loyal to your organization,” he said.

Alternatively, businesses can reach out externally for data scientists and train them rigorously in their business. They can then be paired with selected IT employees or employees with deep knowledge of the business to help create a wider pool of talent, also known as citizen data scientists, says Chitrachar.

Invest for the long term

Chitrachar suggests using commercial off-the-shelf (COTS) data science software to get started quickly, though he cautioned that the organization would need to align their training programs with the selected tools. “Make use of COTS which does that heavy lifting in terms of data analysis, though this would require training on those tools [to] create your citizen data scientist pool.”

Unfortunately, there is no shortcut to the data science journey. Organizations should be prepared to make investments with “little to no ROI” for the first couple of years, says Chitrachar. Moreover, the existing data architecture will likely need to be maintained in parallel while the organization invests in modern solutions and builds a proper data lake.

Because the data science journey is ultimately a change management exercise to enlighten the mindsets of employees within the business and IT teams, the data science team can expect to work on various proof-of-concepts (POCs) with sample data sets to prove business value and garner stakeholder support.

Making it work

Finally, effective communication with stakeholders within the organization should not be overlooked. On this front, Teoh suggests that businesses should identify employees with the right blend of data science knowledge and business know-how.

“It is important to identify a data science evangelist who can act as the ‘bridge’ between business functions to the data science team. This data science ‘translator’ is someone who understands the business context and can identify the right data science solution,” he explained.

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/Nattakorn Maneerat