From Vision to Value: Scaling AI at HealthEdge 

At HealthEdge, our AI Team was created with a clear mission: to accelerate the innovation and adoption of AI technologies that deliver real value for both our customers and our internal teams. Like any transformative technology initiative, we faced a pivotal question early on: should AI capabilities be decentralized into product teams to maximize speed and innovation, or centralized to establish standards and eliminate redundancy?

The answer wasn’t one-size-fits-all. We chose a hybrid approach, balancing autonomy and alignment, by defining three distinct roles for our AI Team: Enablement, Platform Development, and End-to-End Solutioning. We also developed a simple decision-making framework to determine when and how our team engages in each role. This structure enables us to scale AI effectively, maintain quality, and quickly leverage cutting-edge AI tooling and methodologies across the organization.

1.Enablement: In the Enablement role, the AI Team guides stakeholder teams in applying AI technologies. We might suggest no-code solutions such as Claude Desktop in combination with MCP (Model Context Protocol) tools to automate a simple but time-consuming operational process. For a more AI-ready stakeholder team, we might offer architectural guidance on how to set up a multi-agent system using LangGraph with the appropriate handoffs, evaluations, and guardrails.

At HealthEdge, the AI Team plays the Enablement role by providing chat support and regular “office hours” to users in our Claude Pilot Program. We share best practices and reusable templates for concepts such as prompt engineering and context management. We’ve also partnered with HealthEdge’s Learning & Development team to centralize learning resources and present about AI innovations to the entire organization.

2. Platform Development: The AI Team’s core contribution is developing a scalable, robust platform of reusable AI components that provide value across the business. This includes core features of a generative AI system, such as multi-agent architectures, tools, and RAG (retrieval augmented generation), as well as supportive functions like logging, traceability, evaluations, and guardrails. It also includes building out common use cases such as information summarization or Q&A. Individual product teams then configure or combine these components to fit their own needs.

For example, the AI Team built the Claims Summarizer platform as a flexible tool that delivers consistent value across different products. Product teams define their own configurations to achieve uniform results despite varying applications. A claims review analyst can use the Claims Summarizer to quickly assess key claim details before adjudication in our flagship HealthRules Payer product. Similarly, a care manager can leverage the tool to understand a member’s medical history in GuidingCare before determining next steps for care.

3. End-to-End Solutioning: Occasionally, it is necessary for the AI Team to build a complete end-to-end solution beyond just providing functional components to product teams. This can be mandated for high-priority, complex use cases where AI expertise is required for successful delivery of value. Complexity may entail highly networked multi-agent architectures leveraging a broad range of tools or sensitive outputs, necessitating robust evaluations and guardrails. End-to-end solutioning is also a good opportunity for the AI Team to showcase what is possible with AI technology while simultaneously building out the platform to allow other teams to follow the pattern.

At HealthEdge, the AI Team took ownership of an automated document extraction workflow for prior authorization. This involved using OCR to extract key data fields from various prior authorization forms and leveraging AI to map them to internal elements. The large variety of form templates and the lack of one-to-one mappings of data fields made this complex use case a good candidate for the AI Team to take on end-to-end. The project also had a high business impact, with the potential savings of automating the processing of hundreds of thousands of prior authorization forms annually. Given that errors could lead to increased operational costs and delays in care, the AI Team’s thoughtful architecture and thorough evaluations were critical to its success.

CARBS: AI Team’s Role Decision Framework

Given the high demand for the AI Team’s expertise across a large number of initiatives, it was necessary to develop a framework for determining which of the three roles the team would play for a given project, conveniently fitting the acronym “CARBS”:

• Complexity: how much AI expertise does the project require?
• AI Readiness: how much AI expertise does the stakeholder team have?
• Risk: how sensitive is the output (due to privacy, regulatory, or clinical concerns)?
• Business Impact: how much value does the project bring to the organization?
• Scalability: how reusable is the solution across the organization?

When the CARBS framework is combined with the AI Team’s various roles, we maintain quality and avoid repeated work while scaling our impact to the organization.

Why This Matters

The AI Team is designed to collaborate with, not replace, our domain experts. Domain teams continue to own their products, define user needs, and validate success criteria. The AI Team acts as a multiplier, providing them with tools and infrastructure that they might not otherwise have time or expertise to build themselves.

The AI Team’s approach ensures we can move fast without sacrificing quality, avoid redundant work, and scale innovation efficiently. Whether we’re enabling teams with the right tools, building reusable AI capabilities, or delivering complex solutions, our focus is always on turning AI’s potential into tangible results for HealthEdge and the people we serve. For more information about HealthEdge and our AI strategy, visit our blog series found here on our website.