Why Organizations Need to Industrialize Data Science

Datamation content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.

This is the fifth article in a series from Accenture Applied Intelligence on Data Science Transformation. It focuses on how to ensure that data science can deliver the best value for an organization.The prior article in this series is Unlock New Intelligence from Data.

‘Fail fast to succeed sooner’ is the key to disruptive ideas. Why? Because the ability to imagine, experiment and learn, is critical to driving innovation. To do so requires a high-velocity environment. But all too often, data scientists are not equipped to move fast enough, either because they are saddled with aging technology, or because there are outdated or insufficient tools and data available to them. This hampers their ability to answer questions from the business in time or even reach the right people with insights.

Data science can have a transformative impact on businesses. But to achieve that impact, organizations must harness the power of new (and old) technologies to provide data scientists with an ever-evolving, fit-for-purpose data science workbench. Or, as we’ll be referring to it, a “data science playground.”

This playground is essentially a workbench of the right tools and systems for data preparation that allows data scientists to concentrate their time and effort on the math behind the business issue and, as a result, drive tangible value.

A good example is KDDI Corporation, the second-largest telecommunications provider in Japan.

In a saturated mobile marketplace, KDDI wanted to become a ‘life design company’ that provides truly personalized experiences for their customer. Collaborating with Accenture, the company transformed its technology landscape in order to enable improved customer experiences and successfully provide value added solutions to partners in allied industries.

The playground they built features an artificial intelligence-based, real-time, cross-channel recommendation engine fed by centralized customer data from across KDDI affiliates. The data scientist team can access comprehensive real-time data, for example, sensor data from connected cars. As a result, KDDI’s data scientists have what they need to iterate at speed (including data prep, feature detection and algorithm development), allowing them to focus on business value realization and innovative customer experience design.

Unfortunately, this is not a reality in most organizations.

The Data Science Playground – a Reality Check

Instead, in most organizations, data scientists need to make do with limited and siloed desktop tools, inferior data that is unavailable at the speed and granularity required to drive business impact, and archaic batch deployment models. As a result, the business often sees data science as a cash drain, and simultaneously, data scientists become disillusioned, disengaged and, eventually, leave the business.

There are typically a number of issues to blame:

· Limited assortment of tools and systems: With a lot of time spent on data preparation, the little time data scientists are afforded on true value creation is handicapped by desktop statistical and data mining tools. Often the time crunch and limited technology ability to experiment with advanced techniques lead to most data scientists working on BI & reporting tools.

· The speed of technology dictates the pace: Data science workbenches at organizations are not evolving at the same speed as technology advances in the industry, and therefore they’re always playing catch up. This compromises the data science program and leads to limited use cases the business can put forward.

· Lack of industrialized intelligence: Most data science programs fail to reach their full potential because of the complexities involved in creating enterprise adoption and scale. For instance, edge devices involving thousands of real-time deployments require a great deal of work to keep data models updated and fresh. Without robust model management and a real-time deployment environment, business and data science programs often become disjointed and irrelevant for business.

Where Organizations Need to Play. Seriously.

· Establish and maintain the data science playground – Maximizing data scientists’ effectiveness requires an array of ever-evolving, fit-for-purpose technologies, including AI, analytics programming and integrated development environments (IDEs), machine learning, and content analytics. Organizations should consider creating a role that will act as a conduit between business requirements and the data science team’s evolving technology needs.

· Industrialization and automation – Industrializing data science and model management is key to getting new intelligence into the business stakeholders’ hands. For example, where thousands of edge devices are in operation, the organization should deploy the data model in real time and manage it through an automated ecosystem.

· Use continuous model management — Keep models fresh, and develop design-led applications to make them as relevant and accessible to the broader business as possible. Applying advanced deep reinforcement learning algorithms that reward optimized behavior will help data science systems remain relevant longer. Automated continuous improvement will free up data scientists’ time to focus on key business issues.

Investing in knowledge returns the highest interest. Yes, establishing and maintaining a data science playground – let alone, industrializing it – can appear to be very complex and expensive. Nevertheless, organizations can achieve many benefits without massive outlay. Cloud-based data science and analytics solutions offer flexibility at relatively low financial commitment. Deploying open-source technologies can avoid having to pay substantial sums for every new component. And by focusing on ‘human’ interfaces and business processes, rather than a plethora of dashboards, data science outcomes will gain faster traction and broader acceptance across the business.

About the authors:

Robert Berkey is a managing director at Accenture Applied Intelligence, where he leads the Strategy & Transformation offering globally.

Dr. Amy Gershkoff is a data consultant; she was previously Chief Data Officer for companies including WPP, Data Alliance, Zynga, and Ancestry.com.

Takuya Kudo is a managing director at Accenture Applied Intelligence.

Monark Vyas is a senior manager at Accenture Applied Intelligence.