Fraud & Dispute Prevention

RefundKit's dispute risk engine scores refund requests for chargeback probability. It evaluates multiple signals — velocity patterns, amount anomalies, customer history, and behavioral patterns — to produce a risk score and recommendation.

Assess Dispute Risk

const { data: risk, error } = await rk.disputes.getRisk({
  transactionId: 'ch_1N4HbSKz9cXRvFYr',
});

console.log(risk.riskScore);        // 0-100
console.log(risk.riskLevel);        // 'low' | 'medium' | 'high' | 'critical'
console.log(risk.recommendation);   // 'approve' | 'review' | 'deny' | 'preemptive_refund'
console.log(risk.signals);          // Array of detected signals

Risk Levels

| Level | Score Range | Description | |-------|------------|-------------| | low | 0-24 | Normal transaction, safe to process | | medium | 25-49 | Minor flags, monitor closely | | high | 50-74 | Multiple risk signals, recommend review | | critical | 75-100 | Strong fraud indicators, recommend denial |

Signal Types

The engine evaluates four categories of signals:

| Signal | Weight | Description | |--------|--------|-------------| | velocity_spike | 35% | Unusual number of refund requests in a short period | | amount_anomaly | 25% | Refund amount significantly deviates from typical values | | customer_history | 20% | Customer's refund history exceeds normal patterns | | pattern_match | 20% | Transaction matches known fraud patterns |

Recommendations

| Recommendation | Action | |----------------|--------| | approve | Safe to process without additional review | | review | Route to the approval queue for human review | | deny | Block the refund, high fraud probability | | preemptive_refund | Issue a proactive refund to avoid a chargeback |

Using the Risk Engine Locally

import { DisputeRiskEngine } from '@refundkit/sdk';

const engine = new DisputeRiskEngine();

const risk = engine.evaluate({
  transactionId: 'ch_1N4HbSKz9cXRvFYr',
  amount: 15000,
  customerRefundsLast30d: 8,
  averageOrderValue: 5000,
});

Flag a Transaction

Manually flag a transaction for review:

const { data, error } = await rk.disputes.flag(
  'ch_1N4HbSKz9cXRvFYr',
  'Customer opened 5 refunds this week',
);

MCP Tool

AI agents use the refundkit_get_dispute_risk tool to assess risk before processing refunds.

Next Steps

Approvals — Risk scores can trigger approval workflows.
Webhooks — Get notified when transactions are flagged.
Eligibility — Policy rules work alongside risk scoring.