Case study
Alan Fraud Detection Dashboard
Built for Alan, a French health insurer, this dashboard analyzes optical care claims (eyeglasses and contact lenses) to identify suspicious billing patterns. Four detection engines score each provider, and scores drive automated routing: auto-approve, manual review queue, or payment hold pending audit.
The brief
Challenge
Identify fraudulent optical care providers from billing data alone, without access to patient records or ground-truth labels, using only statistical patterns in the claims history.
Approach
What we made
Designed four rule-based detection engines targeting distinct fraud signals: billing spikes versus a provider's own rolling median, simultaneous glasses and contact lens billing, repeated identical euro amounts, and an abnormally high rate of round-number invoices. Scores are additive and provider-level, enabling transparent explanations for every flag.
- Four independent fraud detection engines: monthly billing spikes, dual-product co-billing, repeated exact amounts, and round-number bias.
- Risk scores 0–100 drive automated claim routing: auto-approve (<30), manual review (30–70), or payment hold (>70).
- 8 of 12 providers flagged across a dataset of 221 claims; 5 automatically held pending audit.
- Live demo deployed on Railway with CSV import and full review workflow.
Outcome
Results
8 of 12 providers were flagged with evidence-backed risk scores. 5 were automatically held. The system surfaces the exact rules triggered per provider, making audit decisions explainable and defensible.
Gallery
Visual snapshots
Click any image to expand.