Juro convenes a tribunal of four AI advocates to argue a denied claim, for and against. A human delivers the ruling, grounded in real law, with every word on the record.
Case DN-2026-04471, an employer plan under ERISA. An MRI was auto-denied as not medically necessary. The question is whether the record meets the plan's own criteria. Advocate, begin.
The denial fails the plan's own rule. SPD §4.2 grants a red-flag exception for a progressive deficit, and the 06-02 exam shows a new, worsening foot-drop. ACR criteria independently call for an MRI here. This imaging is necessary.
On the claim as filed, the exception wasn't shown. Code ST-06 fired because no conservative-care window and no neuro findings reached the reviewer. On that submission, the denial was defensible.
In the US, nearly 1 in 5 in-network claims is denied, and fewer than 1 in 100 is appealed. In India, about 1 in 9 health claims is rejected, and the insurance ombudsman took 31,490 health complaints in a single year. In the UK, two thirds of disability denials get overturned once a judge actually looks. The denials usually aren't right. They just go unchallenged.
Different countries, different appeal systems, different years. The shared lesson holds: a large share of denials do not survive review. Sources: KFF, UK Ministry of Justice, India Insurance Ombudsman.
Real people, real denials, on three continents. Each one was a single rule away from a different ending. These are documented cases. For the ones still in court, we keep to the wording the filings use.
After he broke a leg, an algorithm called nH Predict cut his rehab coverage at 19 days, against his own doctor's plan. The lawsuit says only about 0.2% of patients ever appeal.
His family paid roughly $150,000 before he died.
What Juro does. Cite Jimmo v. Sebelius, which bars ending skilled care just because a patient has stopped improving.
Estate of Lokken v. UnitedHealth (alleged) · StatNews
Cigna's PxDx system flagged a $350 blood test and a medical director signed the denial in about 1.2 seconds, without opening the file. The appeal took seven months.
An external reviewer finally ruled the test medically necessary, and Cigna paid.
What Juro does. Hold a 1.2-second batch denial against the duty to run a thorough, fair investigation of each claim.
ProPublica
Niva Bupa rejected his overseas cancer-treatment claim by pointing to an unrelated asthma history it said he hadn't disclosed. A consumer court found a deficiency in service.
He carried about ₹66.5 lakh in treatment costs before the court ordered the insurer to pay.
What Juro does. Cite the rule that a non-disclosure must be material to the condition claimed. Asthma is not material to colorectal cancer.
Mumbai Suburban Consumer Commission · Free Press Journal
Her disability payment (PIP) was wrongly stopped. A coroner later found 28 separate failures in how the claim was handled.
The coroner found the resulting financial distress was the predominant factor in her death.
What Juro does. Catch a benefit stopped on a wrongly-decided 'no good cause' finding, despite a recorded need for support.
Coroner's inquest, 2021 · Disability News Service
Argues the claim is valid, and pulls the records and policy language that support coverage.
Stress-tests every weakness: exclusions, missing documentation, the insurer's strongest case.
Grounds both sides in fact: the exact document or statute that settles each dispute.
The AI weighs both rails and recommends. A human delivers the binding ruling.
cross-model panel · three debaters on Haiku 4.5 · chair on Sonnet 4.6
Case DN-2026-04471, an employer plan under ERISA. An MRI was auto-denied as not medically necessary. The question is whether the record meets the plan's own criteria. Advocate, begin.
Today, fewer than 1% of denials are ever appealed. Of the ones that are filed, about a third get overturned. The gap isn't whether people are right. It's whether anyone has the time to argue. Juro closes the distance from months to minutes. And this is not a one-click appeal letter. It's a full hearing, both sides argued and on the record, which is what holds up when a decision is contested.
Paste the denial letter, or connect a claims feed and let denials land in the queue as they come. No new format to learn. The letter you already have is the input.
Four advocates argue the claim, for and against, against the plan's own rules and the law that sits above them. An appeal otherwise takes months. Here a ruling lands while you're still on the call.
Not a summary. A finished appeal letter with the citations already set in it, the statute, the policy clause, the chart date, each one tied to the line it backs.
Each turn, each exhibit, each ruling is hash-chained into one root for the file. If anyone asks how the decision was reached, the answer is the record itself.
How it pays. Juro earns a share of the claims it recovers, and runs as a reviewer console and API for the hospitals and patient-advocacy firms already spending to fight denials by hand.
A challenge isn't a plea. It's an assertion of a right you already hold. Every ruling cites its authority.
The four advocates are not a script that calls one function after another. Each is a separate agent that registers on Band and sits in the same room. A denied claim drops in, the Adjudicator gets mentioned, and from there the agents argue by answering each other in that one shared room. The hearing is the room's own record, and we export it straight into this interface.
The hackathon asked for agents that genuinely work together, not a pipeline wearing four hats. Putting them in one room on Band, where the Scrutinizer can read the Advocate's argument and push back on it, is what makes the debate real.
The panel even spans models on purpose: the three debaters run on a fast, light model so the back-and-forth stays quick; the chair runs on a heavier one, because weighing both sides and writing the ruling is the call that carries the most reasoning. Different engines, one room — the kind of mixed company Band exists to coordinate.
A tribunal is only worth trusting if it runs cleanly and can't be quietly rewritten. We engineered for both: a debate that can't stall, and a record that can't lie.
Every handoff between agents is forced in code, in a fixed order. A model can argue however it likes; it cannot skip a voice or break the chain.
On Band an agent only speaks by calling a tool. If one reasons out its ruling but forgets to post, we capture the text and post it for it. No turn is ever lost.
The debaters run on a fast model; the chair on a heavier one, because weighing both sides and writing the ruling carries the most reasoning.
Each turn is hashed together with the one before it. The whole hearing collapses to a single root.
Change one word, anywhere in the hearing, and the root no longer matches. Every argument stays pinned to the exact record it cites. The whole thing is verifiable, and tamper-evident by construction.
Health insurance is the wedge, because the rules are already written down. The wedge alone rides on a $150B-plus denial-management market and $4.9 trillion of US health spend, with about one in five claims denied. But the machinery is the same wherever a denial hides behind a rulebook. A different statute loads in; the four roles, the relay, and the sealed record do not change, and every rulebook after the first widens the floor.
Benefits stopped on a contested capacity finding.
Same tribunal. The same Advocate and Scrutinizer, argued from the medical record.
A claim denied for “insufficient nexus” to service.
Same tribunal. Evidence pins each finding to the file; a human signs the rating.
Care delayed by an automated “not medically necessary.”
Same tribunal. Necessity argued against the plan’s own criteria, on a clock.
A claim refused over “misconduct” or availability.
Same tribunal. Both sides heard against the statute, not a call-center script.
A payout cut by an estimating algorithm.
Same tribunal. The Scrutinizer in reverse: stress-test the insurer’s number.
Surprise bills and quietly denied refunds.
Same tribunal. A small, fast hearing where none existed before.
JURO · BOTH SIDES HEARD · A HUMAN ALWAYS RULES · EVERY DECISION AUDITABLE