Ambetter, UHC, Aetna — How AI Agents Navigate Payer IVR Systems

Building an AI voice agent for insurance follow-up sounds straightforward until you actually try to call Aetna. Then you discover that their IVR has 6 layers, asks for a 10-digit claim number that has to be entered in a specific format, puts you on hold for 18 minutes, and sometimes transfers you to a department that can't see the claim you're asking about.

Multiply that by 20+ payers, each with their own system, and you understand why IVR navigation is the hardest part of building production-grade follow-up AI.

How We Map Payer IVR Systems

Before writing a single line of agent code, we do a full IVR tree mapping for each payer. This involves making hundreds of test calls across different scenarios, documenting every prompt, every decision point, every authentication requirement, and every failure mode.

The output is a payer-specific navigation graph — essentially a decision tree that tells the AI exactly what to say or press at each step, including fallback paths when the IVR behaves unexpectedly.

The Major Payers: What We've Learned

UnitedHealthcare

UHC has one of the more complex IVR systems but also offers robust provider portal APIs for practices with high call volume. The IVR requires NPI, TIN, and date of birth for authentication. Average hold time: 14–28 minutes. Key challenge: UHC sometimes requires verbal confirmation for certain claim types.

Aetna

Aetna's system is relatively stable and predictable — good for automation. Authentication uses NPI + claim number. The main challenge is routing: Aetna has separate queues for commercial vs Medicare Advantage that require different navigation paths.

Cigna

Cigna has invested in provider-friendly automation — they have a dedicated provider line with shorter wait times and an eviCore integration for prior auth. Claims follow-up IVR is manageable. Average hold time: 8–15 minutes.

Ambetter

Ambetter is the challenging case. They have an explicit policy against virtual agents and use voice biometrics to detect them. Our approach: the AI handles all prep work, identifies the claim, drafts the talking points, and flags it for a human caller who can complete the call in under 3 minutes with full AI support. Still cuts follow-up time by 65%.

Handling Hold Times

Hold time is where AI voice agents have their most obvious advantage. A human billing specialist sitting on hold for 22 minutes is losing productivity. An AI agent sitting on hold costs essentially nothing — and can be doing other work while waiting.

Our agents use a dual-channel architecture: one thread holds the payer line, another thread continues processing the AR worklist. By the time the payer rep picks up, the agent has already worked 3–4 other claims.

When Calls Go Off-Script

The hardest part of IVR navigation isn't the predictable path — it's the edge cases. What happens when the payer says the claim isn't in their system? What happens when there's a system outage? What happens when you get transferred 3 times and end up back at the main menu?

Production-grade agents need explicit handling for all of these scenarios. Each one requires a specific escalation path — either a retry logic, a human handoff trigger, or a documentation action that flags the claim for manual review.

The Integration Layer

IVR navigation is only useful if the information extracted from the call makes it back into your billing system accurately. Our agents generate structured call summaries that map directly to your EHR or PM system fields — no manual transcription, no interpretation, no data entry errors.