Highlights
As healthcare organisations move from AI experimentation to enterprise-scale adoption, the focus is shifting from technology capabilities to operational transformation. Health plans are no longer simply testing the waters with agentic AI: conversations have moved on to where to start; which use cases to prioritise; how to ensure compliance and quality; how to incorporate human oversight; and on navigating the transition to an AI-human operating model without compromising business continuity.
Agentic AI can execute operational workflows from end to end rather than merely assist with individual tasks. The challenge is no longer proving the technology, but turning a working capability into a reliable, governed operation at scale.
A blend of AI, human expertise, governance and operational excellence can transform payer operations into a scalable model. Here is practical guidance on how health plans can successfully transform, scale and realise value from agentic AI-driven operations.
Deriving value from AI requires a plan for putting it to work. Here is that plan, in five steps.
1. Map the value chain, then prioritise by pain and cost
Start with the whole operation and not just a single use case. Divide the member and provider journeys into their constituent parts, and rate each part on how much friction it creates and the costs to run. Start with the parts that score high on both. Pick your top five. This will give you a simple priority chart you can build for your own operation, and this ensures your budget is mapped to operations where the value actually lies.
2. Build a target-state working model, or agentic twin, for each top function
An agentic twin is a live, working model of the operation, running rules on real cases, for the team to rehearse on before committing real volume and data. Built to production standard from the beginning, the twin grows into the live operation instead of being discarded after the pilot. A twin must be:
3. Design the operating model, not just the automation
The value results from operational redesign, and not from merely installing a tool. It’s not about applying AI to automate a few tasks, while keeping the same cost structure, which will only result in unsettling people without any significant gain. Decide up front how human-AI teaming can run the operations. Identify what should AI decide or escalate, who handles exceptions, and carve out the higher-value human roles.
4. Move the work over gradually and reskill the people who stay
Adoption at scale doesn’t mean acting with haste. Move volume into the AI-first model a piece at a time. As each twin earns confidence – from shadow, to supervised, to running on its own where safe. In parallel, plan for the workforce. Train the human group deeply for higher-value work: judgment, oversight, and handling exceptions.
5. Instrument first, then scale across the chain
You can only scale what you can see. Put observability in place from day one: every decision must be traceable, guardrails checked, outcomes fed back. Prove the numbers on the first five functions, then reuse the same platform and operating model across the rest of the chain. Because the foundation already exists, each new area comes online faster and cheaper than the last.
Here’s how an agentic twin creates value across the value chain.
Claims – pend processing
About one in five claims falls out of auto-adjudication into a manual queue. An agentic twin works the queue end to end, checks the systems, clears clean cases, and sends only exceptions to a human. Meanwhile it learns which problems should be fixed upstream.
Appeals
Appeals typically fall under high volume work with non-negotiable and stringent deadlines, and they are heavily regulated. A twin assembles each case, checks it against policy and regulation, and drafts a determination tied to the governing rule. However, a human still decides every appeal.
Cost of care
Risk ranking typically happens on an annual basis on old claims therefore the problems end up being identified late. A twin keeps risk current, focuses on members who will respond, and checks every recommendation against clinical guidelines.
Member service
Another high-volume process, which is also expensive, and usually spans multiple aspects. A twin follows the conversation, pulls together member, claim, and plan details, and resolves routine issues. It assists a human with the rest.
Operations
With the twin becoming a part of the workforce, hiring, training, quality assurance, and operations strategy needs to be revisited to fortify the operations.
Deriving value from AI requires a disciplined approach to find where the value is, build something that can be trusted, redesign the operation around it, move the work over, and scale once proven value is realised.
Special thanks to Judith Andre, Healthcare Industry Advisor, TCS, for her thoughtful contribution.