Automated eligibility checks
Electronic eligibility is checked first, in seconds, through a real third-party eligibility API. No phone call unless it's actually needed.
AI benefits verification
Less time on hold. More time for patients.
Meet your AI insurance verification assistant. Coverage Agent checks eligibility, follows up on missing benefits, validates coverage information, and keeps your team informed.
Case CA-2047 · Today, 9:14 AMVerified
- Procedure
- Extraction · D7140
- Insurer
- Harborview Dental (HD)
- Eligibility check
- Active, benefits partial
- Insurer call
- 5 benefits confirmed
- Validation
- All fields matched
- Patient estimate
- $80.00
The problem
One coordinator. Hours on hold, every week.
Most payers don't offer a usable API, so verifying an upcoming surgery usually means a phone call, and the number that comes back drives the patient's estimate and deposit.
- of checks need a phone call
- 75%
- on hold, per call
- ~15 min
- insurer calls a day
- 20+
- a week spent on hold
- ~25 hrs
Figures as reported by the practice's coordinator.
Step 2 · Called the insurer
Watch it make the call
When electronic eligibility comes back incomplete, Coverage Agent phones the insurer, asks for each benefit and reads every number back before accepting it.
Simulated insurers. No affiliation with any carrier.
Read the full call transcript
Product
Built for the coordinator behind every surgery
Coverage Agent handles the verification workflow end to end, with safety rules enforced in code, not left to a model's judgment.
Intelligent follow-up for missing information
When electronic data is incomplete, the agent follows up through a simulated insurer phone call to fill in exactly what's missing.
Deterministic benefit validation
Every value the agent reports is checked field-by-field against the authoritative source in plain, auditable code, never just trusted.
Transparent financial estimates
Patient-portion math runs in deterministic code, not the model, and every estimate is clearly labeled with its basis, never a silent guess.
Human escalation when verification fails
Unknown, incomplete, or conflicting information is never guessed. It's handed to a coordinator with a plain-language reason and checklist.
Complete audit trails
Every case keeps a full tool-call timeline and (when applicable) a call transcript, so any outcome can be reviewed end to end.
Step 3 · Submitted benefit reading
Nothing is written until every field matches
The agent reports what it heard. Plain code, not the model, compares each value with what the insurer confirmed, and a single mismatch locks the case.
All five benefit fields the agent reported matched what the insurer confirmed: plan active, procedure covered, plan pays 80%, $50.00 deductible left, $1,000.00 annual maximum left.
How it works
One workflow, six steps
This is a working prototype: electronic checks run against a real test-mode API, and the insurer phone step is simulated rather than a live call. No production clinic integration exists yet.
Clinic appointment
An upcoming appointment needs its insurance benefits verified before the visit.
Bedrock agent
An Amazon Bedrock model (Claude Haiku 4.5) chooses the next action. It never does the math and never sees patient identifiers.
Stedi eligibility API
Electronic eligibility is checked first, in test mode against a real third-party API, the cheapest and fastest path.
Simulated insurer follow-upSimulated
When electronic data is incomplete, a simulated phone call fills in what's missing, not a real call to a real insurer.
Benefit validation
The model's reading is checked field-by-field against the authoritative source in deterministic code, never just trusted.
Case result and audit trail
A verified estimate is recorded, or the case is flagged for a human. Either way, there's a full, reviewable trail.
See it run
Watch a replay of a real, verified Coverage Agent execution, or browse the read‑only case dashboard.