Blog

AI agents, RAG, and integrations — notes from production.

Voice AI for Retail Bank Servicing: Balance, Transfers, and Card Actions

Retail bank servicing calls are mostly balance checks, transfers, and card actions - a strong voice AI fit with authentication built in.

Voice AI for Collections Calls: What Regulation Allows

Collections calling is one of the most regulated calling contexts - disclosure, timing limits, and heavy escalation requirements. Not legal advice.

Voice AI for Insurance First Notice of Loss (FNOL) Intake

Insurance FNOL calls are high-volume and mostly structured data capture - a strong voice AI fit, with a clear escalation line at injury or dispute.

Synthetic Identity Fraud: Why It's Hard and What Signals Catch It

Synthetic identity fraud blends real and fake elements, so each field checks out alone. What catches it is the combination, not any single field.

Real-Time Sanctions and Watchlist Screening Alongside Fraud Scoring

Sanctions screening is a deterministic compliance check, not a fraud score. How to run it alongside fraud scoring without blending the two into one number.

Fraud Detection for a High-Growth Startup: Buy Now, Build Later

For a high-growth fintech, fraud detection is a growth path, not a one-time call - buy now, instrument everything, build the custom layer later.

Scaling Fraud Scoring From Thousands to Millions of Transactions a Day

What breaks as fraud-scoring volume grows: feature-store latency, the LLM cost line, the analyst queue. The decision architecture stays the same.

Building Fraud Detection In-House at a Small Fintech

At small scale, buying fraud detection is almost always right. Building makes sense only for a thin, proprietary layer on top - here's the honest version.

Fraud Model Governance for a Regulated EU Financial Institution

Fraud model governance sits above the audit trail: a model inventory, independent validation, named owners, a review cadence. Not legal advice.

AI Fraud Decisions Under US Fair-Lending Rules

Fraud models can trigger US fair-lending scrutiny without seeing protected characteristics, through proxy variables. Not legal advice.