AI Meets APCD: Using Chatbots to Help Transform Healthcare Transparency

November 2025 – At the Annual Conference of the National Association of Health Data Organizations (NAHDO) in November, Onpoint Health Data’s Chief Technology Officer, Devon Holgate, and Data Analytics Manager, Dominic Gayton, explored one of the most recent and promising intersections of technology and health data: Using artificial intelligence (AI) assistants to enhance public access to the valuable information available from an all-payer claims database (APCD).

From Dashboards to Dialogue

For years, public reporting tools have consisted largely of dashboards to help users explore trends in healthcare cost, quality, and use. While dashboards remain powerful for analysts and some members of the public, their abundance of slicers and stratifiers can be overwhelming for others.

Onpoint’s latest offerings change that dynamic by supplementing dashboards with an AI assistant that offers a natural-language chat experience. Instead of setting multiple filters and parameters, users can simply ask, “What is the most expensive drug in the state?” and receive a clear, conversational answer like: “The most expensive drug by total spend is Eliquis, with $1.4 billion in spending in 2023. Would you like to know the most expensive drug based on the average dollar amount paid per claim?”

By enabling intuitive exploration of APCD data, Onpoint’s chatbot model lowers barriers and elevates engagement, empowering more people to understand and act on valuable healthcare information.

How It Works: Foundation Models & RAG

In their NAHDO presentation, Holgate began by exploring how modern AI assistants can be built leveraging leading AI tools such as ChatGPT (from OpenAI), Claude (from Anthropic), and Gemini (from Google). These large language models (LLMs) are trained on massive amounts of text to understand and generate natural-sounding language that can be leveraged to perform a wide variety of tasks. 

On top of these models, developers can add retrieval-augmented generation (RAG) frameworks that access specific documentation and relevant source materials for focused answers. Through a process known as “tool calling,” these RAG agents can be configured to:

• Write and execute database queries
• Retrieve information from vector stores or application programming interfaces (APIs)
• Search the web
• Summarize and synthesize results for users

When connected with other agents, these specialized RAG frameworks can coordinate to plan tasks, query data, and summarize insights, allowing for more accurate and context-specific responses instead of generalized answers.

From RAGs to Richness

Gayton showcased how this approach can enhance APCD-based transparency reporting. While traditional dashboards can display detailed information visually, AI chatbots do so conversationally, bringing accessibility, interactivity, and personalization to the forefront.

Users can ask follow-up questions, explore tangents, and interact with data in plain English. The AI assistant’s conversational approach offers an easy entry point for any user to explore complex data, bridging the gap between technical complexity and public understanding and democratizing access to health insights.

Building Responsible AI for Public Use

With innovation comes responsibility, especially when working with data related to healthcare. Three best practices for securely and effectively deploying AI assistants in public transparency settings include:

1. Establish guardrails. Prevent chatbots from straying off-topic or accessing restricted content by implementing strict domain and content boundaries.
2. Automate evaluation. Continuously test and monitor chatbot responses using methods like “LLM-as-a-judge” to assess the quality of the chatbot’s answers and ensure accuracy, consistency, and reliability over time.
3. Design for transparency. To ensure that users understand any data limitations and constraints, incorporate clear and prominent disclaimers directly into the chatbot’s user interface rather than trusting the AI model to generate them accurately.

Looking Ahead

AI assistants represent the next step in making health data more usable, equitable, and transparent, helping the public engage directly with the data that shapes healthcare policy and personal decision-making, Holgate noted during the presentation’s conclusion.

AI assistants are becoming easier to build as major technology companies deploy supporting frameworks and likely will become a standard expectation of users over time.

Onpoint continues to explore how AI can responsibly advance health data accessibility while maintaining the rigor, accuracy, and trust that have long defined our work with APCDs and other health data initiatives nationwide.