Using Large Language Models to Provide Commercial Insights to a Leading Pharmaceutical Company

Overview
This case study explores the development of a one-stop-shop hub for insights from interaction data meant to unlock actionable next steps for medical and commercial leadership. Combining strategic experience from the client with Reveal’s AWS experience  in data engineering and Machine Learning, lead to reduced time and effort for client analysts to arrive at actional next steps. 

Opportunity | Key Issues
Many years of collecting interaction data in disjointed, independent data capture systems left the company with ineffective means to view their data strategically across the organization. If the data could be aggregated, standardized, and brought together in a single AI application, the company could identify emerging trends, understand patient and healthcare provider sentiment, discover unmet needs, and track the effectiveness of products and campaigns. This wealth of information allows the company to tailor its strategies more effectively, optimize product development, improve patient outcomes, and ultimately gain a competitive edge in the market. The client engaged Reveal and our experience with AWS robust cloud services to address this opportunity with an eye to future scalability, enhancement, and performance. 

Approach | The AWS-driven Solution Blueprint

Reveal’s approach to the project began with close collaboration with the pharmaceutical company’s data lake teams to explore and understand the large volumes of unstructured data. We focused on pre-processing the data to make it ready for AI analysis, ensuring accurate insights could be drawn. To keep medical and commercial data separate, we developed two distinct data indices, which maintained the necessary boundaries between these domains. Additionally, we used AWS services to build a cloud-based architecture that could scale efficiently as the data grew.  

As part of our data engineering efforts, we created a  data health dashboard within the hub that provided a clear view of data currently in use for the filtered data set, including the completeness of key fields. We supported this dashboard with a data imputation model that predicted the most likely values for missing data, considering the information in other completed fields. For the first time, stakeholders could see and understand the status of the data their teams had been collecting for years, which lead to an internal project to shift data capture policies in support of machine learning development in the years to come. 

With the data in place, we customized several AI and machine learning models for specific needs, such as semantic search, sentiment analysis, and topic distribution. These features particularly benefited technical analytics users and led to the hub being expanded to include more teams and datasets. We also gave users the ability to create custom dashboards with their preferred visualizations and filters. Throughout this process, we maintained strict security and compliance standards, like HIPAA, ensuring that all healthcare data remained secure. Our commitment to continuous optimization ensured the platform stayed responsive and compliant as it evolved to meet user needs. 

Reveal HealthTech leveraged advanced AWS-based infrastructure to build a robust, scalable, cloud-based platform. Key components of the solution included:

• Data Integration and Management: Reveal leveraged AWS Glue for ETL processes and overall data management, used Amazon S3 for data storage, employed Amazon Athena to extract model insights for downstream users, and orchestrated workflows using Airflow DAGs. 

• Data Security and Storage: Reveal utilized Amazon S3 to securely store logs, backups, and event data. Identity management was handled through Amazon IAM. DynamoDB was used for log storage, while Amazon OpenSearch Service was used to maintain separate indices for medical and commercial use cases, preserving the necessary boundaries between these domains. 

• Model Development and Deployment: Reveal developed machine learning models using Amazon SageMaker and deployed these models to production environments through the same service. 

• User-Facing Interfaces: Reveal used Amazon CloudFront to deliver content efficiently and employed React to build data visualizations, displaying model outputs to the user. Extensive filtering capabilities were implemented, including interactive features that allow users to click on visualizations for filtering effects. 

Features | Comprehensive Platform Capabilities 

• Role-Based Data Access: Integrating with the client’s IAM infrastructure ensured that when users logged in, they were directed to a version of the hub appropriate to their role.  

• Custom Visualizations: Infographics built from the model outputs included time series, sentiment analysis, and topic distributions.  Users could build multiple customized dashboards on which they had selected the visualizations they wanted. 

Outcomes | Impact and Improvements
Through this collaboration with Reveal, the platform unlocked the value in years of data across the organization, leading to significant improvements: 

• Time to actionable insights has been greatly reduced. 

• The initial rollout saved the client $1 million per year by replacing a third-party, black box application with fewer customizations. 

• The data completeness dashboard identified process and data gaps, prompting the client to overhaul interaction recording and generate a wealth of new data. 

• The improved UX, including interactive chat and expanded ML insights, led to a tenfold increase in the user base. 

• Insights generated by the Reveal machine learning platform were shared back to the client’s data lake, enabling downstream use by additional teams. 

• Additional teams are slated to be added to this hub, along with their sourced data.

Conclusion
Reveal’s work with the pharmaceutical company significantly improved how they use their data, unlocking value across the organization. By replacing an expensive third-party system and enhancing data management, Reveal saved the client $1 million each year and sped up the time to get actionable insights. The improved user experience and strong machine learning tools led to a tenfold increase in user engagement. As more teams and data are added, the platform continues to drive greater impact and efficiency.