Xanthos — AI Governance Case Studies

AI Governance · Insurance

28 seconds

Time to approve a claim denial, at industry-standard throughput of 130 claims per shift

Insurance Claim Denial and the Illusion of Oversight

An automated fraud detection pipeline denies a motor insurance claim. A human reviewer approves the denial in under 30 seconds. Months later, the policyholder files a complaint. This case study reconstructs what the reviewer actually saw, what was absent, and why the oversight claim collapses under the ICO's real influence test.

Human Oversight Insurance GDPR Article 22 FCA Consumer Duty

AI Governance · Financial Services

SCHUFA C-634/21

EU court: if a score determines the outcome, the score is the decision

You Rejected the Loan, But Do You Know Why?

Automated lending decisions are fast and scalable, but increasingly hard to explain. When an applicant asks why their loan was rejected months later, most systems can only reassemble a plausible story from fragments instead of retrieving the original event. This case study examines why reconstruction fails and what a decision record would change.

AI Governance Explainability Regulatory Compliance Lending

Agentic AI · Tax Technology

RAG + HYDE

Retrieval grounded in hypothetical answers, not raw ambiguous questions

Building a Tax Copilot: Trustworthy AI for Tax Questions

Tax questions are nuanced, messy, and high-stakes. This case study covers the architecture behind a production tax copilot built for Earnr, combining RAG, HYDE, hybrid retrieval, and tool-augmented generation to answer questions accurately and perform calculations reliably without freehand model arithmetic.

RAG HYDE Tool Augmentation LLM Architecture