Cutting CDN Spend Without Breaking the Reader Experience

The problem

The publisher was on a Cloudflare plan billed partly by request volume. The bill kept climbing while traffic stayed flat — a clear sign that something was paying for noise instead of value.

A naive cost-cut would just disable features and degrade the reader experience. The interesting question was: which requests are actually worth what they cost?

The approach

Three weeks of measure-first, change-second.

Logpush + custom dashboards to break cost down by hostname, path, status code, content-type and origin pull rate. Suddenly "the CDN is expensive" became "these 6 endpoints account for 41% of the bill and 0.3% of pageviews."

Cache Rules audit — found the legacy assets pipeline was bypassing cache entirely because of stale `Cache-Control: private` headers from a long-removed plugin. Fixed at the edge with a Cache Rule, kept origin honest.

Worker for stale-while-revalidate on a category of requests where the origin was slow to respond and we were paying for retries.

Killed dead endpoints: 4 paths that had been deprecated for over a year were still being hit by old apps. Returned a permanent redirect at the edge instead of a full origin pull.

Plan review — once the actual usage shape was visible, we renegotiated the plan tier. Same SLA, ~38% less.

The outcome

Monthly CDN spend dropped ~38%. TTFB and cache hit ratio held flat or improved. The savings paid for the engagement many times over in the first 6 months. Editor and reader experience untouched.

More importantly: the team now has dashboards that show cost per route, per content type and per ASN. Future cost growth is visible before it becomes a bill shock.

Why it matters

FinOps for publishers isn't about finding cheaper providers. It's about understanding which traffic is profitable and which is just noise the CDN happily charges you for. Most opinion sites have 10-20% of their bill sitting in noise nobody is watching.