Five Roles for Your Data Science Team
To get the greatest value out of your organization’s data, your data science team needs to play five distinct roles: innovators, explorers, prototypers, optimizers, and responders.
- By Troy Hiltbrand
- January 9, 2019
Everywhere we turn today, businesses are reaping the benefit of effectively harnessing and maximizing the use of their data. To accomplish this target, many businesses are implementing data science teams to focus on delivering business impact through the effective use of their data. These data science teams play five distinct roles within the organization: innovators, explorers, prototypers, optimizers, and responders.
When a data science team takes on the role of innovator, its target is to find ways to disrupt the status quo by effectively using data. This disruption could be in the form of entirely new business models or dramatically different ways to frame an existing problem that gives it a whole new perspective.
A data science team in innovation mode is constantly asking what-if questions of themselves and the business, trying to identify if the application of data science to their data can have evolutionary impacts on how they do business. Teams focusing on innovation also need to look outside of their business to identify examples of how others are gaining value from the application of data science. Often teams look at other companies within their same industry, but effective data science teams include monitoring and learning from businesses from adjacent industries as well.
At times a data science team doesn’t have a clearly defined objective but instead a charge to delve into the data, trying to extract knowledge from it. In this capacity, team members could be termed data dumpster divers, tasked with taking all of the discarded data in the organization’s data lake and finding treasures of insight within it. Data science teams use tools such as anomaly detection and outlier identification, link analysis, and Monte Carlo analysis to uncover previously hidden patterns in the data that can be exploited for future organizational value.
As explorers, much of the time is spent searching without finding substantive results, but when they do find insights, those can be utilized to drive business value.
The most effective teams focus on specific outcomes, either good or bad events occurring in the business, and then work backwards to discover any key drivers in the data that have a bearing on these outcomes. The goal is to identify levers, backed by data analysis, that -- when pulled -- change the outcome in a beneficial direction. The benefit could be to increase the likelihood of a good event or to eliminate the likelihood of a bad event.
With these insights in hand, the next role that a data science team plays is that of prototyper. The team needs to find a way to take the patterns discovered during the exploration phase and develop them into a positive business result.
This process of prototyping could include the development of a model that can be operationalized to automate decision making and optimize business operations. It could also be testing a hypothesis derived from patterns in the data to see if changing elements of the business process have the desired results. It is in this process where the effort of researching starts to yield a definitive benefit for the business.
Once a data science team does the initial work in creating and deploying a model, their job is far from done. Team members must continuously monitor, improve, and optimize these models. This process could include evaluating the flow of data into the model, finding ways to accelerate the process or improving the quality of the data inputs. Data quality efforts include ensuring the quality of the attributes used in the model as well as ensuring they are utilizing the right attributes in the model and that no additional attributes would yield more significant results.
Also, as new data sources come online, it is essential for data science teams to evaluate the potential impact of utilizing this new data to improve the outcomes of models already in production. This evaluation could include adding new data attributes for the model to use or replacing existing data with more accurate or representative data.
Examples of this include new and changing human behaviors associated with a business’s risk model or additional customer touchpoints in the customer journey, such as a new mobile app or a set of marketing campaigns. The addition of this new data could dramatically affect the sensitivity and impact of models in production in either a good or bad way.
Optimization also includes improving technology and data science methodologies. With the increased emphasis on data science, both industry and academia are pushing to create new and optimized methods of extracting insight. Many of these techniques find their way into the public domain and can be consumed by data science teams to improve results.
Finally, a data science team has to be ready to respond to operational issues. As team members monitor their models, they need to be sensitive to performance degradation. They need to focus on both speed and accuracy issues within the model or the degradation of the business outcomes that the data science team is working to enhance.
This role is similar to that of explorer, but the difference is that when they identify performance issues, they must be ready to improve the models and deploy a stop-gap solution to ensure continued operations.
Five Roles, One Goal: Business Success
As a data science team strives to have a business impact, they are called upon to play many roles. Each of these roles is distinct, but each is important in the overall success of the company by taking full advantage of their data to achieve business results. When a data science team can master all of these roles, their potential impact to the business is significant, and the results have the potential of determining a company’s success.
Troy Hiltbrand is the chief information officer at Amare Global where he is responsible for its enterprise systems, data architecture, and IT operations. You can reach the author via email.