Back to Glossary

Fraud & AML

What is a merchant payment hold for fraud score?

A merchant payment hold for fraud score is a temporary suspension of transaction processing when automated fraud detection models assign risk scores above predefined thresholds, typically 70-85 on a 100-point scale, requiring manual review before payment release.

Why It Matters

Fraud holds protect merchants from chargebacks that average $3.60 per dollar of fraud loss when including fees and operational costs. Effective hold strategies reduce false positives by 40-60% while maintaining fraud catch rates above 85%. However, excessive holds damage customer experience and can decrease conversion rates by 2-8%. Balance is critical since merchants face regulatory scrutiny under PSD2 and similar frameworks that mandate reasonable fraud prevention without excessive customer friction.

How It Works in Practice

  1. 1Score each transaction using machine learning models that evaluate 200+ data points including device fingerprints, velocity patterns, and behavioral anomalies
  2. 2Trigger automatic holds when fraud scores exceed merchant-defined thresholds, typically 75+ for high-risk and 65+ for medium-risk merchants
  3. 3Route held transactions to fraud analyst queues with priority ranking based on transaction value and risk severity
  4. 4Review transactions within SLA windows, usually 2-4 hours for high-value payments to minimize customer impact
  5. 5Release approved payments immediately or decline with detailed reason codes for merchant reconciliation
  6. 6Update model parameters based on hold outcomes to improve future scoring accuracy

Common Pitfalls

Hold thresholds set too low generate excessive false positives, creating analyst backlogs and customer complaints that violate PCI DSS customer experience requirements

Insufficient analyst coverage during off-hours can cause legitimate high-value transactions to expire, particularly problematic for international merchants subject to multiple time zones

Missing integration with chargeback data feeds prevents models from learning, causing fraud scores to drift and hold effectiveness to degrade by 15-25% quarterly

Key Metrics

MetricTargetFormula
Hold False Positive Rate<5%Approved holds after review / Total holds placed × 100
Hold Review Time<2 hoursAverage time from hold placement to analyst decision

Related Terms