Lilac Software recently hosted a webinar that brought together four leading practitioners who are actively deploying agentic AI to improve operations, outcomes, and member experience at health plans. The discussion covered everything from real world implementation challenges to organizational change strategies and early ROI signals.
What emerged was a clear-eyed view of what it really takes to make agentic AI work in the complex, regulated world of healthcare. Below is a distilled summary of the key lessons shared. Each lesson is hyperlinked to a point in our conversation where the topic is covered. Collectively these nuggets can help other health plans and innovators accelerate their own journeys of going beyond analytics to unlock the potential of this powerful new technology.
1. Start with Education and a Culture Shift
Adopting AI isn’t just a technical lift—it’s a cultural one. Success depends on demystifying the technology for business teams and creating space for experimentation. That means investing in education, internal champions, and initiatives like hackathons that bring diverse teams together around real problems. The most effective AI rollouts happen when everyone—not just engineers—feels empowered to ideate, test, and refine AI-powered workflows.
2. Prioritize Impact Over Volume of Work
It’s tempting to start AI adoption by targeting high volume workflows. But that often introduces complexity too early. Instead, focus on high impact teams or processes—small groups that influence strategy, care delivery, or compliance. These teams may be underserved by traditional tools but are primed to benefit from intelligent automation. Equipping them with agentic AI creates outsized value and strong internal proof points.
3. Clean Data Is Half the Battle
In healthcare, the biggest barrier to automation isn’t related to an agentic system–it’s the data. Claims, contracts, and medical records often arrive as unstructured PDFs, scanned documents, or bundles of faxes. Turning this information into machine-readable inputs requires serious investment in preprocessing. Organizations that have a foundational investment in analytics can leverage this to power intelligent agents. For plans that don’t have the analytics infrastructure in place for a use case, there is an opportunity to bypass the step of building out sophisticated analytics but there needs to be a well thought out strategy for ingesting documents and structuring data to support agentic workstreams.
4. Focus on Business Outcomes, Not Features
Having Agentic AI deliver “just” time savings isn’t enough to prove ROI. It’s important for plans to measure and report on gains in terms of throughput, faster turnaround times, and reduced administrative load—particularly for member services and clinical review teams. But the most meaningful metrics tie back to mission. For example, if AI reduces post-call documentation, the ROI to highlight is member advocates spending more time with seniors, which clearly connects to improving member impact. In that case, improving efficiency is secondary (but still an essential part of the impact story and worth highlighting).
5. Design for Safety, Scale, and Human-in-the-Loop
Building AI agents that interact with humans requires nuance in understanding emotions and context and includes rigorous safeguards that take those into consideration. This is true in all industries but in healthcare, a small error can have material consequences. That means combining structured prompts, decomposed workflows, and fallback mechanisms to ensure reliability. In most places AI should assist—not replace—human decision-makers. Use cases with clear data provenance and bounded scope make strong starting points.
6. Reimagine Product Thinking
Traditional software development focuses on features. But agentic AI reframes the product as an outcome. What matters most is not the interface, but the work product an AI agent can deliver. Teams should focus on enabling meaningful end-to-end results—like pre-populated audit summaries, drafted care plans, or validated claims—rather than just embedding AI into UI components.
7. Choose the Right Tools for the Job
All AI is not built the same and good at the same things. Not every task requires a massive foundation model. Sometimes AI isn’t even the right tool for the challenge. Traditional NLP or OCR might be more than enough and easier to implement. Emerging best practices involve routing requests through different models based on cost, complexity, and precision needs. Organizations are also moving beyond simple prompts to orchestrated agent workflows that combine reasoning, memory, and external tools for more advanced tasks.
8. Be Realistic About AI’s Limitations
Large language models may appear intelligent, but they require precise instructions and can fail in unexpected ways. They’re prone to hallucination, avoid saying “I don’t know,” and often need structured output formats to stay accurate. Decomposing complex prompts into smaller steps dramatically improves reliability. Teams should plan for prompt engineering, testing, and continuous iteration.
9. Engage Business and Technical Teams Alike
Driving AI at health plans is not the sole domain of engineers and data scientists. Increasingly the best ideas for where AI can be impactful is coming from frontline teams who understand the gaps in care, compliance, or operations. The role of technologists is increasingly to enable—not prescribe—the use of AI. This bottom-up momentum is critical for scaling AI across the organization. There are many tactics to foster that at a health plan, including cross-functional hackathons and teach-in sessions.
10. Lead with a Mission Aligned Manifesto
Every AI initiative should be grounded in the organization’s mission. For health plans, and especially for the likes of Medicare Advantage and Medicaid that serve high risk populations, the core of the mission relates to positively impacting members. Of course, being able to focus on members requires starting with a strong business and being in compliance. Framing specific goals like improving member experience, accelerating care transitions, or reducing audit burden, aligning AI with the human-centric values of your plan builds broader buy-in and creates stronger impetus for change.
—————————————-
Lilac Software is building agentic AI capabilities into our platform to help health plans automate complex operational workflows.
If you’re interested in learning how these systems can transform your plan’s operational efficiency, reach out here to start a conversation with the Lilac team.
