How recommendations work
Data inputs
Costimizer generates recommendations from:
- Billing data — hourly/daily cost from CUR, Azure Cost Management, GCP BigQuery exports
- Resource inventory — instances, volumes, buckets, and services discovered via cloud APIs
- Utilization metrics — CPU, memory, network (where available from cloud monitoring or K8s collector)
- Policy context — TTL rules, tagging gaps, budget thresholds
Recommendation lifecycle
Data sync → Analysis engine → Recommendation card
↓
User reviews detail
↓
Action in cloud / archive
Analysis categories
| Signal | Example recommendation |
|---|---|
| Low utilization | Rightsize EC2 instance |
| Zero utilization | Terminate idle instance |
| Duplicate objects | S3 duplicate cleanup |
| On-demand steady usage | Purchase RI/SP |
| TTL exceeded | Review or extend resource |
| Off-hours usage | Apply power schedule |
Timing
- Most recommendations require 7–30 days of usage history
- Anomaly-based suggestions may appear within 24–48 hours
- RI/SP recommendations improve with 12+ months of stable usage patterns