Case Studies

How anomaly detection caught two cost issues in one week — a CloudWatch log burn and a Vertex AI billing error.

AWS CloudWatch

~$800 saved

CloudWatch log burn from UAT data warehouse

What happened

A UAT data warehouse hit a disk space limit. The application started writing "no space left" errors to CloudWatch with every operation — including full stack traces — multiple times per second. The UAT log group had no alerting configured.

CloudWatch costs climbed to roughly $800 per day. Within two hours, StackSpend sent an anomaly alert. The team investigated, identified the log source, and resolved the underlying space issue. Log volume returned to normal and costs dropped.

Anomaly detection caught a runaway log pattern that internal alerting missed, in time to stop the burn.

Google Vertex AI

Billing corrected

Vertex AI cost spike — billing error

What happened

Vertex AI costs had been close to zero for days. Overnight, reported spend spiked with no change to usage or configuration.

StackSpend flagged the spike as an anomaly. The team raised it with Google. Google confirmed a billing error on their side and corrected the charges.

Daily visibility and anomaly alerts made an incorrect bill visible quickly, so it could be disputed and fixed.

Catch cost anomalies before they scale. Get daily visibility across cloud and AI spend.

Sign up
Case Studies — StackSpend