Data analytics life cycle at the Global Innovation Network and Analysis (GINA)

Team Overview:

• EMC’s Global Innovation Network and Analytics (GINA) team comprises of senior technologists located in centers of excellence (COEs) worldwide.
• The team’s primary focus is to drive innovation, research, and university partnerships by engaging employees across global COEs.

Initiative in 2012:

• In 2012, a newly hired director within the GINA team aimed to enhance innovation-related activities.
• The director sought to establish mechanisms for tracking and analyzing information related to innovation, research, and university partnerships.

Enhancing Knowledge Capture:

• The GINA team aimed to improve the capture of both formal and informal information.
• Special emphasis was placed on capturing insights from informal conversations with thought leaders within EMC, academia, and other organizations.

Global Knowledge Sharing:

• The team envisioned a mechanism to facilitate global knowledge sharing among GINA members, even when geographically separated.

Data Repository Objectives:

Planned creation of a data repository to achieve three main goals:

• Store both structured and unstructured data.
• Track research activities conducted by global technologists.
• Mine the accumulated data for patterns and insights to enhance the team’s operations and strategy.

Strategic Impact:

• The anticipated impact of this approach was to foster a global platform for sharing ideas, increasing collaboration, and improving knowledge sharing within the GINA team.

Discovery Phase:

- Began identifying data sources
- Consulted experts (Tom Davenport, Peter Gloor) to help decide to crowdsource work by seeking EMC volunteers

Roles filled:
— Business User/Sponsor/Manager: VP from Office of CTO
— Business Intelligence Analyst: IT Representatives
— Data Engineer/DBA: IT Representatives
— Data Scientist: Distinguished Engineer

Data fell into two categories:
— 5 years of EMC’s Innovation Roadmap idea submissions (mix of structured and unstructured data)
— Innovation/research activity minutes and notes (mix of structured and unstructured data)

Key Initial Hypotheses
- 10 main initial hypotheses around mapping innovation, evaluating ideas, measuring knowledge transfer, identifying research boundary spanners, etc.

Initial Hypothesis

Grouping of Hypotheses
- Descriptive analytics: analyze current activities to spark creativity, collaboration, asset generation
- Predictive analytics: advise management on where to invest in the future

Data Preparation phase:

• Set up a new analytics sandbox to store and experiment with the data
• Data scientists and engineers noticed during exploration that some data needed conditioning and normalization
• Realized some critical missing datasets were needed to test analytic hypotheses
• Recognized that without sufficient data quality and accessibility, subsequent lifecycle steps wouldn’t be possible
• Had to determine what level of data quality was sufficient for the GINA project aims
• Discovered issues like misspelled researcher names and extra spaces around names in the datastore
• Needed to address these small data problems to enable better analysis and aggregation in later phases

Model Planning phase:

• Social network analysis techniques seemed feasible for much of the dataset to analyze innovator networks
• In some cases, lacked data to appropriately test hypotheses
• For IH9, decided to initiate a longitudinal study to start tracking data over time on people developing new intellectual property

This future data collection would allow testing of:

• IH8: Whether frequent knowledge sharing reduces time to generate a corporate asset from an idea
• IH9: Whether lineage maps show when knowledge sharing did or didn’t result in an asset

Needed to establish goals and parameters for the longitudinal study:

• Identify the right milestones to meet the end goal of an idea becoming a successful corporate asset
• Trace how people move ideas between milestones towards the goal
• Trace ideas that fail and those that succeed
• Compare the journeys of successful and unsuccessful ideas
• Compare the times and outcomes using statistical tests like t-tests or classification algorithms

Model Building phase:

Employed several analytical methods:

• Natural Language Processing (NLP) on textual idea descriptions
• Social network analysis using R and RStudio
• Developed social graphs and visualizations of innovation networks using ggplot2
• Figures show social graphs depicting relationships between idea submitters across countries
• Identified “hubs” — people with high connectivity and “betweenness” scores
• Cluster in one graph had geographic variety, proving hypothesis about geographic boundary spanners

One person stood out with an unusually high score — queried data to learn about his influence:

• Attended top conferences and visited teams globally to share insights
• Presented at widely attended virtual sessions with global attendees
• Introduced researchers to dozens of corporate innovators
• This suggests the hypothesis about identifying influencers spanning geographies/units is correct
• Used Tableau for visualization and exploration
• Used Pivotal Greenplum database for repository and analytics engine

Communicating Results:

• Identified most impactful and relevant findings
• Project successful in identifying boundary spanners and hidden innovators
• CTO office launched longitudinal studies to track innovation over time
• Promoted knowledge sharing about innovation and researchers spanning company and externally
• Enabled cultivating additional intellectual property and new research topics
• Forged university relationships for joint academic research
• Accomplished with limited budget using volunteer force of skilled engineers and data scientists

Key Finding:

• Disproportionately high density of innovators in Cork, Ireland office
• 15% of innovation contest finalists and winners were from Cork despite its small size
• Learned Cork received focused innovation training from consultant, increasing contributions
• Would have been hard to identify this innovator cluster through traditional methods
• Social network analysis revealed a highly contributing pocket of people

Communication:

• Shared findings internally through presentations and conferences
• Promoted externally through social media and blogs

Operationalizing Results :

• Running analytics against innovation activity data yielded great insights into innovation culture
• Key findings:
• Need more data in future, marketing initiative to convince people to share innovation/research activities
• Some sensitive data, need to consider security and privacy regarding who can run models and see results
• Need parallel initiative to improve Business Intelligence — dashboards, reporting, queries on global research
• Mechanism needed to continually reevaluate model after deployment
• Assessing benefits and defining retraining process are main goals
• Showed how analytics can drive new insights into traditionally hard to measure areas
• Informed investment decisions in university research and identified high-value innovators
• Developed recommender systems using topic modeling to help idea submitters refine proposals for new intellectual property