FinOps: Anomaly Detection in Cloud Spend
A SaaS company spends €2.4M a year on cloud across 900 services and 60 teams. Several times a year someone leaves an oversized database running or a job in a loop, and the bill shoots up by €40,000 before anyone notices. Incidents have never been labelled: each one was different.
Where the money comes from
What to do now
- 1Tag cloud spend by team and service.
- 2Activate anomaly detection with alerts in euros.
- 3Define owners and a 24h response per team.
For your committee
One page with the decision, the impact, the risks, the steps and the KPI to track.
The print dialog opens: choose "Save as PDF".
Watch out: more volume is not always more profit
Cutting cloud spend blindly can degrade the product and cost customers. Separate anomalous spend from spend that grows because the business grows (cost per customer).
KPI for the committee
Cloud cost per customer and euros of anomalies detected.
Detail for your data team
Proposed technique: Isolation Forest
Why: There are no 'incident' labels and each case is different, so a supervised model cannot learn. Isolation Forest detects what is unusual compared with normal behaviour: anomalies are isolated with very few random cuts.
Cost of getting it wrong: A false negative costs tens of thousands of euros; too many false positives lead engineers to mute the alert channel.
Data needed: Cloud billing by hour, service, team and region (last 18 months).; Usage metrics: CPU, storage, network traffic, number of instances.; Calendar of deployments and launches (to explain legitimate peaks).