Blog

From the fraud engineering team.

Practical writing on transaction risk, behavioral modeling, and the fraud patterns we are watching in the data. No product announcements, no thought leadership filler.

Abstract visualization of card-testing attack patterns
Abstract visualization of risk score threshold calibration
Risk Modeling
Calibrating Your Risk Score Threshold Without Burning Customers

Setting a block threshold at 0.7 feels safe until you see the false positive rate on your best customers. How to tune with precision.

· 7 min
Abstract visualization of device fingerprinting limitations
Detection Methods
Device Fingerprinting Has Limits. Here Is What It Misses.

Device signals are useful but degrade fast with VPN normalization and browser privacy modes. What the behavioral layer catches that fingerprinting leaves behind.

· 6 min
Abstract visualization of first-party fraud behavioral signal
Attack Patterns
First-Party Fraud Is the Hardest Type to Catch. Here Is the Signal Pattern.

The customer disputing their own transaction is using the same device, same IP, and the same behavioral fingerprint. The signal lives elsewhere.

· 9 min
Abstract visualization of chargeback data as training signal
Risk Modeling
Using Chargeback Data as a Training Signal for Your Risk Model

Every confirmed fraud chargeback is a labeled data point. How to wire that signal back into your scoring model without introducing lag.

· 7 min
Abstract visualization of BNPL installment fraud patterns
Attack Patterns
BNPL Fraud Patterns Are Different From Card Fraud. Treat Them That Way.

Installment-splitting exploits have a distinct temporal pattern that card-focused rules engines completely miss. What the data looks like.

· 8 min
Abstract visualization of synthetic identity fraud pattern
Attack Patterns
Synthetic Identity Fraud: How the First Transaction Reveals the Pattern

Synthetic identities pass KYC and look clean for weeks. The behavioral anomaly surfaces at the first high-value transaction attempt.

· 8 min
Abstract visualization of latency vs accuracy tradeoff in fraud detection
Engineering
The Latency vs Accuracy Tradeoff in Real-Time Fraud Scoring

Adding 40ms to your authorization path will cause measurable checkout abandonment. How to get the score in under 10ms without sacrificing model depth.

· 10 min
Abstract visualization of account takeover behavioral signals
Attack Patterns
The Three Behavioral Signals That Precede Most Account Takeovers

Credential stuffing leaves a trail before the takeover succeeds. Login timing entropy, device switch, and transaction type shift are the early indicators.

· 7 min
Split comparison visualization of rules-based versus ML fraud detection
Engineering
Velocity Rules vs Behavioral ML: When Each One Actually Works

Velocity rules are not obsolete. They are fast, auditable, and compliant. The question is what they leave on the table when attackers adapt.

· 9 min