The fastest way to cut a delivery bill is almost never the renewal call — it is configuration, because most estates are quietly paying for cache misses, uncompressed bytes, oversized images and logs nobody reads. Here are ten levers ranked by typical payback speed within a quarter, with the honest note on each about who it works for. Most estates find double-digit percentage savings in the first six without a single vendor conversation.
Ground rules: measure, then move
Two rules keep cost work honest. First, decompose the bill before touching anything — the fifteen-minute decode from the invoice guide tells you whether your money is in bandwidth, requests, features or logs, and the ranking below reorders itself accordingly (a video estate lives in levers 1–6; an API estate may find lever 5 is half the bill). Second, one lever at a time with before/after measurement, because several of these interact and you want to know what actually paid. Every saving below shows up on the next invoice, which is the scoreboard.
Levers 1–3: serve fewer origin bytes
1. Raise cache-hit ratio. Every recovered miss saves origin egress (billed by your cloud at rates that dwarf CDN rates) and often CDN-internal fees. Cache-key fragmentation, needless Vary, short TTLs on stable objects — the ten levers inside this lever are their own guide, and on most estates this is the largest, fastest win available. 2. Fix origin economics directly. If origin egress is a big line, move the heavy object classes to zero-egress storage — the arithmetic and setup from the storage guide applies to single-CDN estates identically; payback is typically months. 3. Shield and collapse. If shields are off or partial, the origin is paying retail for what a mid-tier would consolidate — shield configuration is an afternoon, and on chatty estates the origin-side saving is immediate.
Levers 4–6: serve fewer bytes, full stop
4. Compression to modern settings. Brotli (and Zstandard where supported) at sane levels on all compressible types, verified actually serving — estates routinely ship gzip-or-nothing years after better was available; the compression guide has the settings, and text-heavy estates see meaningful billed-GB drops. 5. Put images on a diet. Modern formats, right-sizing to display dimensions, sensible quality — images are most estates’ largest byte class and the image guide’s checklist commonly cuts image bytes by a third or more; weigh optimizer add-on fees against the bandwidth saved, per the invoice decode. 6. Shrink the request count. Where the requests line is material: bundle micro-assets, kill polling that could be event-driven, extend client caching so repeat visitors stop re-validating, and check that error storms and retry loops are not manufacturing billable requests — the requests-to-GB ratio from your invoice review is the tell.
Levers 7–8: pay for less machinery
7. The log diet. Full-take logs to expensive storage, on an estate that queries one percent of them, is pure burn: sample intelligently, trim fields, tier the storage — the logging guide’s framework routinely cuts this line by half or more with zero capability loss for how logs are actually used. 8. The feature audit. Walk every metered feature line to the config that drives it and ask whether it still earns its meter: the image optimizer processing images your build pipeline already optimized, edge functions invoked on paths that no longer need them, real-time analytics nobody opens, dedicated certs on hostnames that could share. Features accrete; the audit is annual at minimum — fold it into the annual review — and each retired meter is a permanent saving.
Levers 9–10: pay better prices
9. Fix the commit fit. If the invoice shows chronic overage, or a chronic gap below commit, the commit is mis-sized and money is leaking at whichever penalty applies — re-run the sizing maths against current reality and use the mid-term-upgrade clause if you negotiated one; this is the cheapest “negotiation” there is, because the case is arithmetic. 10. Bring leverage to the rate. The full renegotiation — market-checked rates, a credible second-platform option or an active split whose ratio can move, timed to the contract calendar — is the biggest single lever on the sheet and the slowest, which is why it is ranked last for a this-quarter horizon: start it this quarter, bank it at renewal, with the renewal guide as the playbook. The quarter’s honest sequence: levers 1–8 now, 9 as soon as the clause allows, 10 on the calendar — and every saving documented, because the next budget conversation goes very differently when delivery walks in having cut its own bill.