Anodot by Glassbox

🎉 Breaking news: We’re excited to announce that Glassbox has acquired Anodot - the best-in-class anomaly detection engine! By combining Anodot within the Glassbox platform, our customers can: - Continuously detect anomalies without manual configuration or data science expertise - Identify early warning signs of friction, conversion drops, and performance issues - Proactively resolve digital experience issues before they impact business outcomes Together, Glassbox and Anodot help to deliver an exceptional digital experience. Read the full announcement here: https://lnkd.in/g_gDZWMS

New Case Study: How BioCatch Strengthened Fraud Detection with AI-Powered Monitoring In behavioral biometrics, stability is critical. Even small data shifts can distort fraud risk scores. BioCatch, which protects 196+ financial institutions and monitors over 10 billion sessions monthly, needed a way to track and validate the integrity of its fraud scoring models in real time. Using Anodot’s AI-driven anomaly detection, BioCatch now monitors: ✅ Fraud scores returned to customers ✅ Input features and components that influence scoring ✅ API and data pipeline changes that can silently skew results With Anodot, the team detects unexpected behavior within hours instead of weeks — preserving model accuracy, customer trust, and fraud detection reliability at scale. 📘 Read the full story: https://lnkd.in/dZ3qk--f #AI #FraudDetection #BehavioralBiometrics #Fintech #AnomalyDetection #DataMonitoring

Earlier today, Amazon Web Services (AWS) experienced a major disruption affecting thousands of online services worldwide. Many of our customers rely on AWS, and Anodot detected anomalies 20 minutes before AWS publicly acknowledged the issue, and nearly an hour before they classified it as major. The chart below shows how early detection made a difference: 🔹 6:51 AM UTC – Anodot detected the first signs of degradation. 🔹 7:11 AM UTC – AWS posted their first update. 🔹 7:51 AM UTC – AWS confirmed it as a major issue. 🔹 10:01 AM UTC – Recovery complete. While they couldn't prevent the outage, Anodot’s early alerts gave our customers a crucial head start, helping them prepare, mitigate impact, and communicate with their own users before most even realized it is happening. Wouldn’t you also want to know before everyone else? #anomalydetection #ai #awsoutage

I asked GPT-5: What are the leading solutions for business monitoring with AI? It did a good job:

🚀 Building Production‑Ready Anomaly Detection: Key Takeaways from Arun Kejariwal & Ira Cohen https://lnkd.in/dpYpD67Y A while back we recorded this session - which is still very much relevant today. In this deep‑dive panel moderated by Ben Lorica 罗瑞卡, two veterans of large‑scale AI based time‑series systems shared their technical playbooks: Arun Kejariwal: Architect and developer of Twitter’s real‑time anomaly detection open source Ira Cohen: Co‑Founder & Chief Data Scientist at Anodot, driving streaming ML for business metrics Some of the main topics discussed: 1. Framing Anomalies with Context An anomaly goes beyond statistical outliers by incorporating business metadata - like data‑center tags or promotion calendar events, and seasonality‑aware forecasts. This hybrid feature engineering ensures models distinguish true incidents from expected fluctuations. 2. A Hybrid, Unsupervised Pipeline At scale, manually labeled data isn’t feasible. Their solution combines: ⏺️ Forecast‑residual analysis for predictable cycles ⏺️ Density‑based clustering to capture multi‑dimensional shifts ⏺️ Online change‑point detection for continuous adaptation to concept drift 3. Scaling with Microservices & Streaming Architecture and in memory learning: A modular chain - Ingestion → Feature Store → Real‑Time Serving → Correlation → Root‑Cause Graph, runs each stage in containerized services connected via message queues. This design keeps end‑to‑end alert latency below 500 ms. 4. Correlation & Incident Grouping Instead of firing hundreds of isolated alerts, a correlation engine links related anomalies (e.g., traffic drop leading to cache errors and revenue impact). This clustering cuts noise by over 80% and surfaces coherent incidents. 5. Open‑Source vs. Production‑Grade Reliability While open‑source tools can assemble PoC pipelines, production demands: ⏺️ Real‑time SLAs (avoiding multi‑hour lag) ⏺️ Automated health checks and drift monitors ⏺️ Canary deployments and on‑the‑fly rollbacks These safeguards transform research prototypes into mission‑critical services. 6. Continuous Safeguards & Feedback Loops Embedding automated drift detection alongside user‑validated incident feedback keeps models accurate over time. Canary deployments and rollback triggers ensure any degradation is caught before SLAs slip. #anomalydetection #ML #AI #observability #monitoring