Innovaccer Delivers Population Health Insights in Under 60 Seconds with Generative AI

Innovaccer helps healthcare providers, payers, and public healthcare entities around the world to deliver better care through more effective use of data. The Innovaccer cloud platform simplifies access to patient records by integrating information from multiple healthcare systems, while its advanced analytics tools offer prescriptive and predictive insights for more equitable patient care.

Increasingly, Innovaccer is integrating AI into its healthcare solutions to optimize care, enhance patient outcomes, and lower costs. One of its focus areas is population health management (PHM), which involves analyzing data from various sources to understand the health trends, risks, and needs of a population. The insights drawn from PHM systems help organizations allocate resources effectively, addressing care gaps and informing targeted interventions.

However, extracting actionable insights from multiple data sources—such as electronic health records, claims, and patient-reported outcomes—can be challenging. Typically, users frame their questions in everyday language, but the underlying data resides in structured databases that require SQL commands to retrieve the necessary information. As a result, data analysts manually translate these questions into complex SQL queries spanning multiple interconnected tables, which can take hours or even days to write and validate. This time-consuming process delays critical healthcare decisions, especially when quick responses are needed.

Innovaccer saw an opportunity to bridge this gap by empowering healthcare teams to directly extract insights without needing SQL expertise. The goal was to use generative AI to automatically convert natural-language queries into accurate SQL commands. However, as the team began developing this solution using a large language model (LLM) through an API, they encountered a significant challenge: the accuracy rate of these automated conversions remained below 59 percent. This low accuracy posed multiple risks, including misleading insights, wasted resources, and reduced trust among users. Nilav Baran Ghosh, senior director of data science at Innovaccer, explains, “We needed a way to reliably convert healthcare queries into SQL without sacrificing accuracy or speed. Our initial attempts didn’t meet our standards, so we knew we had to take a different approach.”

Knowing it needed a more reliable way to convert natural-language queries to SQL without compromising accuracy, Innovaccer turned Anthropic Claude—an LLM known for its advanced natural language processing capabilities. To streamline integration, Innovaccer adopted Amazon Bedrock, a fully managed service that simplifies building and scaling generative AI applications. This decision was driven by the need for model flexibility and efficient management, plus Innovaccer’s previous experience using Amazon Bedrock to build a dynamic internal chatbot.

Innovaccer also adopted an agentic architecture, using autonomous agents capable of iterative reasoning and validation to enhance its solution’s accuracy. By breaking down complex tasks into smaller, more manageable actions, the agents dynamically adapt their strategies, validate intermediate outputs, and significantly reduce inaccuracies. This approach facilitates greater scalability, efficiency, and trustworthiness in converting natural-language healthcare queries into precise insights.

To strengthen their generative AI skills, Innovaccer’s data science team participated in AWS Solution-Focused Immersion Days. These sessions provided hands-on experience with AI tools and guidance from AWS Solutions Architects, helping Innovaccer quickly get up to speed on new managed services. “The support we received was exceptional,” recalls Ghosh. “AWS not only offered technical guidance during development but also provided ongoing assistance to optimize our implementation.”

By using Amazon Bedrock, Innovaccer also gained the ability to switch between multiple LLMs for different types of queries. For example, the team used Anthropic Claude Haiku for simpler requests and Claude Sonnet for more complex analyses. This approach significantly reduced computational and token costs, optimizing the efficiency of AI integration. Moreover, the well-documented Amazon Bedrock interfaces made switching between models seamless, saving significant development time compared to using the previous API-based method.

To minimize the risk of misleading insights, Innovaccer is also exploring Amazon Bedrock Guardrails to help detect and filter any inaccurate outputs. “The flexibility and ease of use of Amazon Bedrock have been game-changers for our development process,” says Ghosh. “We can now optimize costs while ensuring reliable, accurate responses, which ultimately benefits healthcare teams and patients alike.”

Innovaccer released its AI-powered PHM, Population Health Copilot 2.0, in early 2025, marking a significant milestone in healthcare data management. By incorporating generative AI, Population Health Copilot 2.0 can extract insights for healthcare teams in under 60 seconds—a significant improvement from the previous process that took hours or even days. This rapid conversion automated the entire process through a conversational interface, boosting data analyst efficiency by 40−80 percent and freeing analysts to focus on more complex tasks.

One of the most impactful outcomes was a boost in accuracy. Since switching to Amazon Bedrock and utilizing Anthropic Claude, Innovaccer has seen the accuracy rate of text-to-SQL translations rise by 50 percent. This means healthcare teams can better rely on the insights generated, fostering greater trust in Innovaccer’s solutions. The solution has also delivered performance improvements, including a 20 percent increase in text-to-SQL conversion speed—not only saving time but also freeing backend systems to handle additional workloads, providing greater overall value to healthcare organizations.

Innovaccer has also achieved significant cost efficiencies while enhancing its solution’s performance. Routing simpler requests to Anthropic Claude Haiku rather than more complex models reduced computational and token costs by more than 90 percent and accelerated response generation by 10–20 percent. With automated data retrieval and analysis, healthcare teams can now make near real-time decisions, enabling more data-driven strategies and reducing cognitive load across multiple touchpoints. “Using generative AI on AWS, we’re empowering healthcare workers to detect early signs of health issues, deliver more personalized care, and enhance care coordination for long-term patient benefits,” concludes Ghosh.