P08: Accelerated Decision Making with Analytical Solutions: Case Studies
Poster Presenter
Sujith Kurup
Director, Software Development
IQVIA India
Objectives
The objective of this presentation is to share practical, real-world visual analytics use cases for faster data review and use of ML models to create leading indicators for faster and proactive decision making.
Method
Multiple visual analytics and ML models were compared and the suitability and acceptability was evaluated based on efficiency and quality achieved. The solutions were also evaluated for replicability and sustainability to compete with current landscape of the industry.
Results
Visualizations should present a quick and easy way to get the answers the user needs to make the right decisions. More data doesn’t mean more insights. We have more data than we know what to do with, which means that opportunities to create charts are endless. While developing a visualization, it is tempting to include a lot of data points, make it very fancy, and ignore the requirements of the end users. Visualizations should be developed as tools to support decision-making, not just as a laundry list of KPIs. HCD, an easy to follow visual communication plan and efficient way to establish visual hierarchy helps data managers to develop simple yet effective visualizations with minimal efforts that helps the entire team (from individual contributor to leadership) to make quick and independent decisions required at their level and be ever ready for difficult milestones such as database locks.
1) Efficiency: No inspection findings (3 inspections). No additional man hours.
2) Highly replicable: Adopted by 6 operations (study build, SAS, analytics, etc.)
3) High visual preference & efficiency: F pattern1, Arrowhead Model2 and RAG based hybrid – 67% (N=88) preferred RAG2 and 78% preferred interactivity2.
4) Granular actionable insights: pending issues, quantity, and stakeholders. Proactive identification, and resolution of inspection findings.
5) High end-user satisfaction: 100%, (N=7)
6) Additional benefits (Gemba3): 15% time saving, manual error-free, uniformity, better documentation.
7) Predictive modeling4-9: Optimum output: Multi-Class Logistic Regression Model (TF-IDF4 vectorized inputs) - 76% vs. average 60%
Conclusion
One size “generally” does not fit all. This is an era of data, and there are a lot of analytical solutions available to manage them. When the race of data currency is on, in most use cases, analytical solutions are custom made and developed differently based on individual requirements. Visual burden is real. End-users, most of the time, tend to avoid “new solutions” and prefer to go back to the usual practice unless the provided analytical solutions are easy to adopt, exciting, and with minimal or no learning curve. Solutions capable of providing leading indicators, easy to understand and allow end-users to gain granular insights without shuffling between applications and tools are always appreciated better. Moreover, the definition of insight varies based on roles and responsibilities. Developing a one-size-fits-all solution capable of supporting all such requirements is difficult and requires a multidisciplinary approach and robust research on user acceptance. Use cases pertaining to a visual analytics model entwined with ML and NLP, replicated into two different successful projects of inspection readiness and faster database lock. The roadmap, concepts, efficiency of the visualization models, scalability, replicability, user acceptance, efficiency of the monitoring systems are key aspects that could make the solutions successful.
