PayU Overwatch

PayU Overwatch is a real-time monitoring and anomaly detection platform designed for ensuring smooth payment operations across the PayU transactional ecosystem. Its main features include observing transactional metrics and early issue detection.

Core Objectives

The platform has four clearly defined goals:

1. Proactive Detection: Quickly identify anomalies in transaction performance within minutes.
2. Intelligent Correlation: Provide root-cause insights by linking anomalies across merchants, banks, and payment modes.
3. Automated Communication: Send structured alerts via webhooks, Teams, or Email.
4. Continuous Learning: Adapt dynamically to historical data and trends by learning over time.

How It Works

This section breaks down the operational workflow of PayU Overwatch into four steps:

1. Data Collection: Ingest live transactional and performance data from PayU's ecosystem.
2. Anomaly Detection: Use machine learning and rule-based logic to detect deviations in metrics such as success rates or error rates.
3. Alert Generation: Generate alerts with context (e.g., affected entities, duration, severity) upon detecting anomalies.
4. Notification Delivery: Send alerts via webhook callbacks to configured destinations for automated responses.

Benefits

• Early Visibility: Detect and mitigate potential issues before merchants notice them.
• Operational Efficiency: Reduce manual workload and improve mean time to detect (MTTD).
• Automated Response: Facilitate incident management workflows.
• Enhanced Transparency: Equip stakeholders with clear, data-backed information about system health.

Key Monitoring Capabilities

1. Transaction Success Rate Monitoring

• Continuously tracks success rate metrics across merchants and banks.
• Detects deviations and performance degradation.
• Triggers alerts for irregularities based on historical trends.

2. Failure Pattern Detection

• Identifies sudden spikes in failure rates across merchants, acquirers, and product modes.
• Helps trace the root causes of failure (e.g., gateway downtimes or configuration problems).

3. Real-Time Anomaly Detection

• Uses statistical and behavioral models to detect unusual patterns.
• Captures issues before merchants escalate them.
• Contextualizes alerts with stats like srt_drop_abs and srt_drop_rel.

4. Multi-Entity Monitoring

Allows alerting at multiple levels:

• Merchant level
• Acquirer/Bank level
• Card Scheme level
• Product level

Supports selective alert subscriptions for specific entities.

5. Severity Classification

All alerts are rated with a criticality score (0–100):

• 0–30: Low Severity
• 31–60: Medium Severity
• 61–100: High Severity

Enables prioritization of investigations.

6. Continuous Health Tracking

• Tracks alert duration.
• Resolves alerts if performance normalcy is restored automatically.
• Enables performance benchmarking with correlations to other system metrics.

Alert Delivery Methods

The platform supports three methods for delivering alerts:

• Email
• Teams
• Webhook

📨 Webhook Alerts

For additional details about webhook alerts implementation and configurations, refer to: Webhook Alerts Documentation

Use Cases & Scenarios

1. Merchant Performance Degradation

Scenario: A merchant's transaction success rate decreases compared to historical data.
Webhook Trigger: Metric: "success_rate" and a criticality score ranging between low, medium, and high.

2. Early Warning for Zero Success Rate

Scenario: All merchant transactions fail, leading to a zero success rate.
Webhook Trigger: Metric: stats.zero_srt: true and criticality_score ≥ 60.

3. SLA Compliance & Reporting

Scenario: Teams require data on uptime and anomaly durations across merchants/products.
Webhook Trigger: Notifications for anomaly start and end times (started_at, ended_at) are utilized.

4. Intelligent Alert Routing

Scenario: Dedicated teams manage specific merchants or product lines.
Webhook Trigger: Payload includes entity_type and product information for routing alerts.

Support & Resources

For further assistance, contact: [email protected].