Right-size, don’t ration: How we cut 42% from cloud cost

When teams hear “cut cloud cost,” they often hear “ration compute and suffer.” That’s the wrong framing. The biggest lesson from Northwind: most waste comes from inherited over-provisioning, not real workload.

1. Map cost by service

Start by breaking the bill down by service and environment tag. You can’t cut what you can’t measure.

• Tag every resource by service + environment
• Identify the top 5 hotspots
• Measure real load, not configured load

2. Right-size first, savings plans second

Right-sizing tells you what you actually need. Then lock in savings plans for the stable portion.

Rule: right-sizing shrinks demand, savings plans lock in price. Doing it backwards locks in price for capacity you don’t need.

3. Move batch to spot

Non-realtime workloads (batch, training, rendering) run fine on spot with a little retry logic. It’s the largest and safest saving.

Results

Northwind cut 42% of monthly cost, reduced p95 latency by 48%, and held 99.98% uptime — in 6 weeks.