ML-Based Anomaly Detection
Quick Reference
What & Why
Definition
>= 60% of critical metrics use ML anomaly detection (DeepAR, ARIMA) for dynamic thresholds instead of static alerts.
Business Value
Provides objective Go/No-Go deployment decisions and prevents 80% of error budget violations through SLO-driven alerting and error budget policies Achieving >= 60% critical metrics have anomaly detection is a key milestone toward this goal.
Context
This capability is part of the Acceleration milestone's focus on scale automation, embed compliance, improve speed & reliability. Essential for teams targeting MTTR, CFR improvements.
Success Criteria
>= 60% critical metrics have anomaly detection
Measurement
Anomaly detection coverage across metrics
Evidence
- Anomaly detection model configs
- Anomaly alert examples
- False positive rate < 10%
In Practice
Real-World Implementation
Teams train ML models on metric history (7-30 days), detect anomalies based on seasonal patterns, normal variance. Alert on deviations > 3 sigma.
Concrete Example
Implementation Guide
Implementation Steps
Follow the measurement approach: Anomaly detection coverage across metrics
For detailed step-by-step guidance, refer to the SLO-Driven Observability & Error Budgets Implementation Kit.
Resources
Implementation Kit
SLO-Driven Observability & Error Budgets KitTemplates
Browse all templatesRelated Resources
View learning pathsRelated Capabilities
Prerequisites
Implement these first
Enables
What this unlocks
Complementary
Often adopted together, from the SLO-Driven Observability & Error Budgets epic
Troubleshooting & FAQs
Common Issues
Issue: Target metric not improving
Solution: Verify measurement is accurate, check if prerequisites are fully implemented, review evidence artifacts for completeness
Issue: Team resistance to adoption
Solution: Start with pilot team, demonstrate value with metrics, provide training and support during transition
Issue: Inconsistent implementation across teams
Solution: Create shared templates and guidelines, establish regular sync meetings, use automation to enforce standards
Frequently Asked Questions
Can we implement this before completing prerequisites?
While possible, it's not recommended. Prerequisites ensure foundational practices are in place, making this capability more effective and easier to adopt.
How long does implementation typically take?
Most capabilities can be implemented within 90 days when tackled as part of the Acceleration milestone. Individual timelines vary based on team size and existing practices.