A SaaS founder asks the obvious question: “Is our cloud bill reasonable?” The honest answer is that the bill itself is meaningless. What matters is what the bill buys per paying customer, and how that number moves with growth.
What’s happening
Cost-per-customer (sometimes “infrastructure CAC adjacent”) is the single metric that tells you whether infrastructure is scaling sub-linearly with revenue. Most SaaS teams either do not compute it, or compute it once and never refresh it. The result is a class of surprises: a free-tier funnel that has quietly become a majority of compute spend, a feature flag that doubled per-request memory and broke the unit economics, or a region expansion that added fixed overhead without adding customers in that region.
The math itself is simple. The work is identifying which costs are per-customer, which are fixed, and which are growing on a different curve entirely.
How to verify
Take the monthly cloud bill and decompose it into three buckets:
- Per-tenant variable. Compute, request egress, storage, and observability that scale with tenant activity. This is what cost-per-customer is actually measuring.
- Fixed platform. Control plane, CI runners, shared databases that do not scale linearly with customer count, security tooling, base monitoring. These get amortized across the customer base.
- Growth investment. Pre-production environments, new region footprints, experimental services. Exclude from per-customer math; track separately.
Compute is (bucket 1 + (bucket 2 / paying customers)) / paying customers per paying customer per month. Track it monthly. The slope matters more than the absolute number — a flat or declining line is healthy; a rising line means infrastructure is growing faster than the customer base.
The fix
Get tagging discipline first. Every billable resource needs a tenant or shared tag so the bill can be split. Wire the AWS Cost and Usage Report (or GCP billing export) into a warehouse and join it against the customer count by day.
Then look for the outliers. The two most common findings on first analysis: free-tier and trial accounts dominate bucket 1; a small number of enterprise accounts have customer-specific overhead that should be priced separately. Both are addressable, but only after the math is visible.
Stack Harbor builds the tagging and reporting pipeline as part of Cloud Compute onboarding. If your unit economics feel opaque, book a SaaS cost review.