A rising CDN bill does not always mean traffic growth is the only problem. This monthly review workflow helps you separate cacheable and uncacheable requests, find oversized assets and inefficient image variants, measure origin pulls, and estimate savings before changing production traffic rules.
Overview
CDN bandwidth cost optimization starts with understanding what the bill actually measures. Depending on the provider and plan, charges may include delivered bandwidth, request volume, regional egress, image transformations, cache storage, origin fetches, or add-on services. A useful review therefore begins with the provider's current billing definitions rather than assuming that every increase is a cache problem.
The central question is: which traffic is being delivered, how often is it being delivered, and why is it reaching the CDN or origin in that form? A popular campaign can increase legitimate traffic. However, the same bill may also reveal uncompressed JavaScript, oversized images, cache-busting query strings, low cache-hit ratios, bot activity, repeated downloads, or an image transformation rule that creates too many variants.
Review costs alongside performance. Removing caching from dynamic pages may reduce a particular type of cache usage but increase origin load and time to first byte (TTFB). Conversely, improving cache-control headers, compression, or asset formats can reduce bandwidth while preserving or improving delivery speed. The best change lowers avoidable usage without weakening reliability, freshness, security, or user experience.
For a repeatable process, export one complete billing period of CDN data and compare it with the same period's request, cache, origin, analytics, and deployment data. Keep a record of assumptions, changes, and results so the review becomes a trend rather than a one-time investigation.
How to estimate
Use a simple cost model before making a configuration change:
Estimated CDN cost = delivered bandwidth × bandwidth rate + requests × request rate + transformation usage × transformation rate + other applicable charges
Not every provider uses all of these components, and some plans combine or waive certain fees. Replace each rate with the value shown in your current contract or billing dashboard. If a charge is included in a subscription, record it as included rather than assigning an invented unit price.
To estimate a potential saving, first calculate the volume that could be removed or avoided:
Potential bandwidth saving = current delivered bandwidth × avoidable percentage
Then apply the relevant rate:
Potential variable-cost saving = potential bandwidth saving × applicable bandwidth rate
Run the calculation separately for each cause. For example, oversized images should not be combined with low cache efficiency when both changes have different implementation risks. Also account for new costs. A resizing service might reduce image bytes but add transformation requests or storage usage. A tiered cache may reduce origin egress while adding a feature charge. The estimate should show both sides.
Use a confidence label for each estimate:
- High confidence: based on a measured asset size, request count, and current billing rate.
- Medium confidence: based on representative traffic samples or an incomplete dashboard export.
- Low confidence: based on a proposed rule whose effect has not yet been tested.
After implementation, compare actual usage with a comparable period. If traffic volume changed substantially, normalize the result by requests, sessions, page views, or another stable business measure. This prevents a successful traffic campaign from being mistaken for a failed optimization.
Inputs and assumptions
Collect the following inputs before diagnosing why CDN bandwidth is high:
| Input | What to inspect | Likely question |
|---|---|---|
| Delivered bandwidth | Total bytes, ideally by hostname, path, region, and content type | Which resources account for most of the transfer? |
| Request volume | Requests by path, status, method, device, and geography | Are many small requests creating cost or load? |
| Cache performance | Cache-hit ratio, miss ratio, bypasses, stale responses, and revalidations | Why are repeat requests reaching the origin? |
| Origin pulls | Origin bandwidth, fetch count, response time, and error rate | Is the CDN shielding the origin as intended? |
| Asset characteristics | File size, compression, format, dimensions, and cache lifetime | Can the same content be delivered with fewer bytes? |
| Transformations | Image variants, requested dimensions, formats, and failed transformations | Are too many variants being generated? |
| Billing terms | Current rates, included allowances, minimums, and regional rules | Which measured activity is actually billable? |
Separate traffic into at least four groups: cacheable static assets, cacheable public pages or responses, private or personalized traffic, and deliberately bypassed traffic such as authenticated requests. The correct cache policy depends on whether a response contains user-specific or sensitive data. Never make a resource public merely to improve a cache-hit ratio.
Review cache-control headers and cache keys together. A long browser lifetime does not guarantee edge caching, and a CDN rule cannot safely fix an origin response that includes private data. Query strings, cookies, authorization headers, device dimensions, language, and unnecessary cache-key variation can all split otherwise reusable content into separate objects.
For static assets, check how to cache static assets with versioned filenames or controlled query parameters. For text resources, compare compression settings and compatibility using the Brotli versus Gzip CDN guide. For media, inspect format, dimensions, quality, and variant counts with the CDN image optimization guide.
Worked examples
Example 1: Oversized image delivery
Assume a monthly report shows that product images deliver 240 GB. A sample of real requests indicates that 25% of those bytes are avoidable because desktop-sized images are being sent to smaller screens. Under the provider's current bandwidth rate, estimate the affected volume as:
240 GB × 25% = 60 GB potentially avoidable
Do not treat 60 GB as guaranteed savings. Test responsive dimensions, modern formats, and an appropriate quality policy on a representative set of URLs. Then check whether image transformations introduce new request or processing charges. The related image optimization guide can help structure that test.
Example 2: Low cache efficiency on public assets
Suppose a JavaScript bundle is requested 500,000 times in a month, but deployment logic gives every request a different query-string value. The file content is effectively identical, yet the CDN sees multiple cache keys. If a cache audit shows that 40% of the bundle traffic is being fetched or revalidated unnecessarily, estimate the potentially affected request volume as:
500,000 × 40% = 200,000 requests for investigation
The cost impact depends on the bundle size, request pricing, origin-fetch behavior, and the provider's cache rules. A safer improvement may be to use content-hashed filenames, set an explicit immutable policy for versioned files, and keep deployment invalidation targeted. Validate cache purges and hit behavior after deployment rather than assuming the header change worked.
Example 3: Origin shielding decision
Assume a public site has a high volume of edge misses concentrated across several regions. Before adding another caching layer, compare current origin pulls, origin bandwidth, origin response time, and the provider's applicable feature charges. Estimate the avoided origin volume from measured repeated requests, then compare that value with the operational and financial cost of origin shielding or tiered caching. The origin shielding guide provides a framework for this decision.
Do not judge the change by CDN bandwidth alone. A successful configuration may leave edge delivery nearly unchanged while reducing origin load, improving consistency, or protecting the application during traffic spikes.
When to recalculate
Make CDN cost review a monthly operational task, with a deeper comparison after material changes. Recalculate immediately when the provider changes pricing, billing definitions, included allowances, regional treatment, transformation fees, or request accounting. These inputs can change the answer even when application traffic is stable.
Also revisit the model after a marketing campaign, product launch, major release, domain or DNS change, CDN migration, image-pipeline change, compression change, cache-rule update, or origin move. Review the first complete billing period after each event, then compare with a representative baseline. A traffic-growth plan should include projected requests, average response size, cache-hit assumptions, origin bandwidth, and a conservative range for variable charges.
Use this monthly checklist:
- Save the current invoice and provider pricing documentation.
- Rank bandwidth by hostname, path, content type, and region.
- Separate public cacheable, private, bypassed, and failed traffic.
- Review cache-hit ratio, origin pulls, revalidations, and purge activity.
- Inspect the largest assets and the most frequently requested assets separately.
- Check image transformation counts and the number of generated variants.
- Estimate each proposed change using current billing inputs.
- Test one low-risk change, document the expected result, and monitor it.
- Compare actual savings, performance, origin load, and error rates.
For broader diagnosis, pair this process with a website cache audit checklist, a repeatable CDN performance testing workflow, and the DNS and CDN routing guide when regional TTFB changes. A disciplined review does more than reduce CDN egress cost: it creates a current, evidence-based view of how your infrastructure delivers every byte.