Methodology
How LatencyCost.com sources its data, what is in and out of scope, the calculation framework behind every model, and the monthly refresh cadence that keeps it current.
Sources
Every cost anchor on the site traces back to one of the sources below. Where a number appears on a vertical page, it carries an inline citation; this table is the consolidated index.
| Source | URL / location | Refresh cadence | What we take from it |
|---|---|---|---|
| Akamai State of Online Retail Performance | akamai.com (retail performance reports) | Periodic (Akamai release schedule) | Conversion-rate elasticity per second of load time; bounce-rate at 1s / 3s / 5s thresholds |
| Akamai mPulse RUM platform documentation | techdocs.akamai.com | Reference (RUM methodology) | Real-user monitoring methodology framing for case-study interpretation |
| Google web.dev Core Web Vitals | web.dev/vitals | Quarterly (Google CWV announcements) | LCP, INP, CLS thresholds; authoritative source for the March 2026 LCP tightening |
| Google PageSpeed Insights / Lighthouse | developers.google.com/web/tools/lighthouse | On Lighthouse version bump | Per-metric scoring methodology that drives the SEO impact page |
| Chrome User Experience Report (CrUX) | developer.chrome.com/docs/crux | Monthly dataset refresh | Public field-data source for the SEO Impact page; aggregate site performance distributions |
| Cloudflare Radar | radar.cloudflare.com | Live / continuously updated | Independent global internet performance benchmarks for the CDN comparison |
| Catchpoint Internet Performance reports | catchpoint.com | Quarterly research releases | Third-party CDN latency measurements outside any single vendor's network |
| Cisco ThousandEyes Internet Reports | thousandeyes.com | Quarterly | Third-party internet performance benchmarks; complementary to Cloudflare Radar |
| HTTPArchive corpus | httparchive.org | Monthly run | Public performance corpus; aggregate cross-site Core Web Vitals distributions |
| WebPageTest documentation | webpagetest.org | Reference | Per-test methodology framing for the optimisation guide |
| Cloudflare CDN public pricing | cloudflare.com/plans | Monthly verification | Reference price per Cloudflare Free / Pro / Business / Enterprise tier |
| AWS CloudFront pricing | aws.amazon.com/cloudfront/pricing | Monthly verification | Per-GB and per-request CDN pricing; on-demand list price only |
| Fastly pricing | fastly.com/pricing | Monthly verification | Tiered CDN pricing; on-demand published rates |
| Bunny.net pricing | bunny.net/pricing | Monthly verification | Budget-tier CDN reference price for cross-vendor comparison |
| Moallemi and Saglam: The Cost of Latency in High-Frequency Trading | Columbia Business School working paper (publicly archived) | Reference (academic) | HFT loss-per-millisecond cost anchor for the /trading page |
| Aerospike / sreschool / Brendan Gregg P99 references | Multiple public sources | Reference | P99 tail latency math, sequential and parallel chain compounding for /api-latency |
| Deloitte / Google: Milliseconds Make Millions | blog.google / Deloitte 2020 mobile retail study | Reference (2020 study) | Mobile retail conversion-rate elasticity at 0.1s improvement granularity |
| ParksAssociates Cloud Gaming research | parksassociates.com | Periodic research releases | Cloud gaming latency threshold and churn-cliff data for /gaming |
| BLS OEWS wage tables | bls.gov/oes (Software Developer + SRE categories) | Annual | Engineering hour-rate anchor for incident response cost modelling |
In scope
- Page-load latency to conversion-rate models from Akamai, Google, Walmart, COOK, AutoAnything, and Mobify published research
- P99 microservice latency compounding math (sequential and parallel fan-out)
- Cloud gaming latency thresholds from ParksAssociates and academic survey research
- HFT latency cost economics anchored in the Moallemi academic paper
- CDN comparative latency from Cloudflare Radar, Catchpoint, and ThousandEyes
- Core Web Vitals thresholds from web.dev/vitals (Google authoritative source)
- CDN list pricing from vendor public pages
- Engineering hour-rate anchors from BLS OEWS
Out of scope
- Vendor-negotiated enterprise CDN pricing (Cloudflare Enterprise, Fastly Enterprise, Akamai Custom)
- Real-time HFT exchange colocation rack pricing (NYSE / Nasdaq private negotiation)
- Internal-loaded engineering hour rates beyond the conservative $75/h public anchor
- Geographic latency outside the major core latency corridors (NYC-Chicago, NYC-LON, LON-FRA)
- Per-browser RUM variance (mobile vs desktop is included; per-browser is not)
- Tier-1 transit backbone latency negotiated rates
Calculation framework
Conversion-loss model
Approximately 7 percent conversion loss per additional second of load time, aggregated from Akamai retail performance reports and the Deloitte / Google mobile retail study. Variance by site type is documented on each vertical page. The calculator uses the lower bound of the published range.
Bounce-rate elasticity
Google / SOASTA published thresholds: bounce-rate increases of roughly 32 percent at 2-3s, 90 percent at 3-5s, and 123 percent at 5-7s relative to the sub-1s baseline. Used in the quick-reference table on the homepage.
P99 compounding
Sequential chain: sum of per-service P99s. Parallel fan-out without hedging: maximum of per-service P99s, multiplied by the number of dependencies. With hedged requests and early cancellation, the effective latency approaches the slower service's P95 rather than its P99.
HFT loss-per-millisecond
Moallemi-anchored estimates: roughly $100M per millisecond for a major HFT firm, $4M per millisecond for a brokerage. Tabb Group and Aite Group derivative estimates within the same order of magnitude. The /trading page documents which estimate fits which firm size.
Cloud gaming churn elasticity
ParksAssociates research identifies a sharp churn cliff above the 100ms input-to-display threshold and a steeper cliff above 200ms. Subscription churn modelling on the /gaming page uses these thresholds with conservative monthly churn assumptions.
Engineering hour-rate anchor
$75 per engineer-hour for incident-response cost modelling on /api-latency. BLS OEWS-derived blended rate for US Software Developer plus SRE categories. Higher loaded rates exist; the published anchor stays conservative.
CDN price-to-latency tradeoff
CDN pricing from vendor public pages mapped against measured P50 and P99 from Cloudflare Radar / Catchpoint / ThousandEyes. The /cdn-comparison page presents both axes without scoring; the reader decides which side of the tradeoff fits their workload.
Refresh cadence
Monthly first-business-week verification pass against the eight vendor pricing pages (Cloudflare, AWS CloudFront, Fastly, Bunny.net) and the four authoritative threshold sources (web.dev/vitals, Cloudflare Radar, Catchpoint, ThousandEyes).
A single LAST_VERIFIED_DATE constant in src/lib/schema.ts drives the footer stamp, the Article schema dateModified across every page, and the verification badge on the /about and /methodology pages. There are no other date strings hardcoded into source.
Out-of-cycle refresh triggers:
- Google Core Web Vitals threshold updates (the March 2026 LCP tightening was the most recent example)
- CDN pricing-model changes by any of the four tracked vendors
- Major HFT cost research papers or Tabb Group / Aite Group rate updates
- Akamai or Google retail performance report releases
- Cloud gaming subscription pricing or platform churn data from ParksAssociates
Limitations
Calculator outputs are estimates from public industry data. Actual impact varies by site type, audience, and product category. The 7 percent conversion-loss-per-second figure is a portfolio average; commodity retail sites typically see a stronger elasticity than B2B SaaS sites.
Akamai and Google retail studies date from 2010-2020. We interpret them conservatively: where two studies disagree, the calculator uses the lower-bound elasticity. Re-running the Amazon / Walmart / Google studies on 2026 traffic would likely show stronger elasticity (mobile traffic share is higher than in 2010, and mobile conversion is more sensitive to load time than desktop), but in the absence of contemporary replications we stay with the published numbers.
HFT cost anchors are from public academic and industry research. Firm-specific economics, capital under management, and strategy mix mean any single firm's per-millisecond loss could be an order of magnitude higher or lower than the Moallemi-anchored figure.
Vendor pricing pages drift between verification cycles. CDN tiers occasionally rename or restructure (Cloudflare reorganised the Pro / Business / Enterprise gating in 2024); any layout change can introduce transient inconsistencies that the next monthly pass resolves.
Corrections process
Spot a stale price, a misquoted source, or a methodology gap? Email [email protected]. We aim to acknowledge corrections within 5 business days, ship the fix on the next monthly verification pass, and credit the reporter in the commit message unless they prefer otherwise.