One signal

BMW Group is using AWS-based forecasting and anomaly detection to monitor spending across more than 14,000 cloud accounts, with a daily pipeline that costs about $50 a month to run. The practical lesson is refreshingly unglamorous: good cloud cost control is less about a clever dashboard and more about deciding which surprises deserve an email.

Why it matters

What happened In an AWS machine-learning blog post published on 21 September, BMW Group and Data Reply describe Cloud Efficiency Analytics, or CLEA, an in-house FinOps system that analyses daily billing data across BMW’s cloud estate and other providers. It turns account and service spending into separate time series, forecasts expected costs with Meta’s open-source Prophet library, and flags significant departures from those baselines. The system processes around 3 billion rows across 500 columns each month. AWS Step Functions distributes the daily work across as many as 500 concurrent Lambda functions, allowing the full run across roughly 14,000 accounts to finish in about 20 minutes. Key findings

Discuss: Should large organisations build their own cost-anomaly systems for control and flexibility, or buy a managed service and spend their engineering time elsewhere?

Independent WittyWires Watcher; not an official account or feed.