Ideally, every team would have access to clean "Cost per Customer" data updated daily. In practice, Platform Engineers often lack direct access to revenue databases, CRM systems, or business analytics platforms. The data exists, but it lives behind organizational boundaries or requires complex data pipeline work to access.
In these situations, use Engineering Proxy Metrics — technical measurements that correlate strongly with business activity.
What Makes a Good Proxy?
A good proxy metric has two properties:
1. High correlation with business output: When business activity increases, the proxy increases proportionally.
2. Easy to measure from infrastructure data: It can be derived from logs, metrics, or cloud billing data without needing external data sources.
Proxy Examples by System
| Business Metric (Ideal) | Engineering Proxy (Practical) | Data Source |
|---|---|---|
| Cost per Order | Cost per POST /api/checkout request |
API Gateway logs |
| Cost per Active User | Cost per authenticated session | Auth service metrics |
| Cost per Video Streamed | Cost per GB delivered via CDN | CloudFront/CDN logs |
| Cost per Search | Cost per Elasticsearch query | ES slow query logs |
| Cost per ML Prediction | Cost per model inference call | Model serving metrics |
The Proxy Workflow in Practice
Suppose you run an e-commerce platform and want to track cost efficiency without access to the orders database. A practical approach using AWS: enable API Gateway access logging, query CloudWatch Logs Insights for the daily volume of checkout requests, and separately pull the monthly tagged cost for the checkout service from Cost Explorer, filtered by the service's cost-allocation tag.
Dividing the monthly tagged cost by the monthly checkout request count gives you "Cost per Checkout Attempt" — a proxy that engineering teams can optimize against without needing access to the finance data warehouse.
Proxy Limitations
Proxies are approximations. A "Cost per Checkout API Call" metric will be inflated if your error rate is high (failed checkouts still cost money but produce no business value). Always track proxy metrics alongside quality metrics (error rate, success rate) to avoid optimizing the wrong thing.