What AVA can (and can't) do in a healthcare denial
AVA is InsuredAudit's Claude-backed advisor for denied claims and authorizations. It's genuinely useful, and it's deliberately boxed in. Here's exactly where the line sits, and why.
What AVA does
When a denial can potentially be overturned under the plan's own published criteria, AVA:
- Compares the denial against the applicable government or plan guideline and identifies whether the record on file appears to satisfy it.
- Drafts specific, citable appeal language, referencing the exact rule, citation, and the facts already in the member's record.
- Suggests alternative procedure or diagnosis codes and documentation that could strengthen a resubmission, when the underlying facts support it.
- Explains a denial in plain English, translating CARC/RARC codes and clinical shorthand into a narrative a member can actually read.
What AVA never does
AVA does
- Draft appeal language for a human to review
- Cite the specific rule and explain the comparison
- Suggest documentation to gather
- Translate codes and jargon into plain English
AVA never
- Files an appeal on its own
- Determines medical necessity
- Decides the outcome of a case
- Acts without a human reviewing its output first
Why the line is drawn exactly here
This isn't caution for its own sake. It's a direct response to how AI-generated conclusions get treated in medical and legal proceedings. If a large language model were the thing determining medical necessity, every one of its outputs would face the same scrutiny any expert testimony faces under evidentiary standards like Daubert: what's the model's error rate, is its reasoning reproducible, was it trained on relevant, unbiased data? Those are hard, sometimes unanswerable questions for a general-purpose language model.
Deterministic rules, published clinical criteria, and a qualified human reviewer don't have that problem. Their reasoning is inspectable and their citations are checkable against a real published document. So InsuredAudit's platform keeps AI on the acceleration side of the line, drafting, summarizing, translating, and keeps every actual determination (medical necessity, appeal outcome, tier decision) on the deterministic-rules-plus-human-signoff side.
What this means if you use AVA's suggestions
Anything AVA drafts, an appeal letter, a re-coding suggestion, is a starting point, not a final answer. It's built to save the time of finding the right rule and drafting the first pass at citing it correctly, not to replace a provider's or reviewer's judgment about whether it actually applies to your specific situation.
See AVA's output on a real denial
The demo shows AVA's actual suggested appeal letter, side by side with the guideline it cites.