Claimback predicts whether your complaint will win money back, shows what's weakening it, and helps you rewrite it with only your own facts.
Live demo
Type what happened, or load a fictional demo story. Nothing you type is stored.
Your score, the sentences that help or hurt, and what's missing will appear here.
Baseline model, running in your browser
Issue type:
How it works
Money complaints sent to the US Consumer Financial Protection Bureau, 2018–2026, each labelled with whether the person got money back.
Every company name becomes [COMPANY], so the model learns from how the story is told, not who it's about.
A language model fine-tuned to predict the chance of money back and the issue type, with a fast baseline model as backup.
An LLM restructures the complaint. Any rewrite that adds a new number or claim Sam never gave is rejected.
We test on the future, not the past: the model never saw a single 2025 or 2026 complaint while training.
Results
Honest limit: only about 6% of recent complaints got money back. The score shows how strong a case is. It is not a promise.