Imagine a monitoring system so smart that it not only alerts you to actionable problems but actively guides you to a solution—transforming reactive troubleshooting into proactive operation. Our presentation, "Machine Learning In Observability Systems: From Insights To Action," tackles today’s critical challenge: the deluge of data from digital ecosystems and the escalating costs of downtime. In a world where even a second's lapse in performance can cost thousands of dollars, traditional methods simply can't keep pace.

We dive into how cutting-edge machine learning techniques—with a focus on anomaly detection, forecasting to GenAI copilots enabling Root Cause Analysis (RCA) and agentic AI performing automated incident management—can revolutionize observability systems. By leveraging statistical methods, supervised and unsupervised algorithms, time series analysis, and deep learning architectures, our talk demonstrates how to filter out noise, correlate events at scale, and deliver actionable insights that ensure system resiliency and optimal performance.

This presentation aligns technical strategy with business imperatives, offering practical frameworks for reducing operational costs while boosting resilience amid growing data and fiscal challenges. Tailored for today's dynamic digital environment, our session promises a comprehensive roadmap that bridges theory and practice, providing attendees with the strategies needed to drive business innovation leveraging observability.

Devoxx

A conference talk by Nike's Observability Platform Engineering Director covering practical ML applications in observability systems at massive scale. Key topics include: the challenge of data volume and cardinality in telemetry (metrics, logs, traces), using ML for intelligent anomaly detection, building event-to-incident funnels to reduce alert noise (Nike processes ~2M alerts/day with a 15:1 noise-to-incident ratio), predictive pre-scaling for traffic spikes (shoe drops, Black Friday, Singles Day), using LLMs for natural language querying of observability data instead of proprietary DSLs like PromQL or SPL, AI-generated dynamic dashboards, automated postmortem generation, and AI-assisted runbook creation. The speaker emphasizes that organizational maturity and cultural readiness are prerequisites for AI-driven automation, and encourages experimenting with open-source ML tools immediately.

Machine Learning In Observability Systems: From Insights To Action by Panos Tsilopoulos