Compare font resources under stated assumptions
Set a synthetic font allowance of 90 kB and three files, then compare it with the 240 kB six-file starting example under stated transfer assumptions. The resulting waterfall is arithmetic, not a browser measurement. Approve the reduction only if required characters, readable fallback and the buying controls still work.
Worked example — illustrative data.
| Font resource | Before size | Before serial interval | Proposed size | Proposed serial interval |
|---|---|---|---|---|
| Body regular | 50 kB | 0.20–0.70 s | 35 kB | 0.20–0.55 s |
| Body bold | 50 kB | 0.70–1.20 s | 35 kB | 0.55–0.90 s |
| Display regular | 45 kB | 1.20–1.65 s | 20 kB | 0.90–1.10 s |
| Display bold | 45 kB | 1.65–2.10 s | Removed | — |
| Body italic | 25 kB | 2.10–2.35 s | Removed | — |
| Display italic | 25 kB | 2.35–2.60 s | Removed | — |
| Total | 240 kB / 6 files | Ends 2.60 s | 90 kB / 3 files | Ends 1.10 s |
Synthetic budget: at most 90,000 transferred font bytes across three files. Assumptions: decimal kB; all fonts discovered at 0.20 s; serial transfer at 100 kB/s; no extra latency, caching, TLS, decoding or parallelism. Each duration equals size ÷ 100. The modeled difference is 150 kB and 1.50 s; it is not a measured site improvement or a browser-loading prediction.
Inventory required characters, languages, styles and licensing before selecting files. Replace these assumed sizes with actual transferred-byte captures in a controlled staging comparison. Review fallback readability, product names, prices and selection controls during slow or failed font loading. Keep field metrics and browser traces separate from this arithmetic model.
Preserve the language and layout task in the font comparison
Run the same product page with both font sets under matching conditions, including accented names, long labels and purchase controls. Record request timing and visible layout behavior separately. The serial intervals in the table are synthetic; a real browser can schedule requests differently. Keep a fallback-font capture so a smaller download cannot conceal unreadable text or shifted controls.
Decision: Approve the smallest font set that preserves the intended reading and interaction tasks. Publish measured before/after values only after actual captures exist and the conditions are recorded.
Subsetting can remove required accented characters: A reduced character set can drop accented names or required language glyphs. A smaller file is not a successful result if customers see missing characters or inconsistent fallback in essential content.
Continue with Guides for stores, software teams and local businesses, Technical SEO services and How to prioritize website speed improvements.
For loading, responsiveness and layout-stability metrics, see Google Web Vitals documentation (checked November 2026).