I built an AI that reads denial letters and writes the appeal
Built Appealify — AI that reads your denial letter and writes your appeal. 85 million claims denied in 2024, only 0.3% appealed. Free to analyse, $24 one time to unlock three complete letters. Any denial, any country. Launching on Product Hunt today — would love feedback.
Trending on Indie Hackers
A one-week feature took two months, mostly spent keeping three systems in sync
The 85M denied claims stat is wild. Most people don't even attempt appeals because they assume it's a lost cause - but that 0.3% approval rate means anyone who actually tries has decent odds. The $24 unlock model is smart too. How are you handling the variability in denial letter formats across different insurers?
Exactly — the system is designed to feel final when it isn't. The gap between 85M denials and 262K appeals isn't apathy, it's the letter doing its job of discouraging people.
On format variability — Gemini 2.5 Flash handles this well. We don't parse structured fields or rely on templates. The model reads the full letter as unstructured text and identifies the denial grounds, policy references, and applicable rights from context. So whether it's a terse two-line BCBS rejection or a five-page Cigna clinical review decision, it extracts what matters.
The trickier challenge is jurisdiction. The same denial type has different appeal rights in different countries. That's something we're working toward in v2 — where the report will cite specific regulations by jurisdiction rather than generalising. For now the letters are strong but the legal citations are more general than I'd like.
Happy to share more about the approach if useful.