AI SRE at RH: 75% Less Alert Noise, Faster Root Cause
RH's SVP of Technology on bringing in NeuBird, the AI SRE: alert noise down about 75%, a 56-minute early warning, and a Vault outage resolved fast. Watch now.
RH, the luxury lifestyle brand, runs e-commerce around the clock across North America, the UK and Europe. Moving from a monolith to microservices took RH from one release every two months to two a week, but it also brought observability noise and longer investigations. At FLOCK '26, Nagesh Bidhurakanthu, SVP of Technology at RH, talks with Francois Martel, Head of Field Engineering at NeuBird, about what changed when RH brought in an AI SRE.
The slowest part of every incident call was agreeing on what the problem was. NeuBird investigates on demand, querying RH's telemetry in place with no heavy streaming pipeline, so the team starts from a defined problem. In a web-service incident the team first suspected Redis; NeuBird traced the first signs of MongoDB cluster latency to 56 minutes before the primary alerts fired. In a Vault outage, it pinpointed a mis-scoped change that landed 400 milliseconds before the first errors and recommended the fix, and RH recovered in about 40 minutes.
Alert volume dropped from about 4,000 to about 1,000, and the proof of concept turned up cloud cost savings the team had not gone looking for. Nagesh closes on where incident response goes next: stakeholder updates sent automatically, and remediation that NeuBird stages for an engineer to approve, behind configurable human approval gates.
In this conversation
Why microservices made incidents harder to frame
Cutting alert volume from about 4,000 to about 1,000
The Vault outage: root cause and a recommended fix
A 56-minute early warning in a web-service incident
How root cause changes the first 10 minutes of an incident
What's next: remediation behind configurable human approval gates
Related Resources
See NeuBird in action
Root cause in minutes, not war rooms.
