Six things, ranked by impact:

Six things, ranked by impact: 1. Right-sized requests using 2 weeks of actual usage (most pods requested 3-4x what they used) 2. Moved stateless workloads to Spot via Karpenter with diverse instance types 3. Scaled dev/staging to zero outside 9am-9pm IST 4. Switched to Graviton (arm64) for 70% of services, just a rebuild 5. Cleaned up orphaned EBS volumes and old snapshots (₹40k/month!) 6. Set log retention to 14 days on noisy services #1 alone was 15%. Nobody wants to lower their requests until you show them the graph.