The business case for website speed

Measurably, at increments too small to see. Deloitte and Google studied 37 brand sites across 30 million user sessions for their 2020 report Milliseconds Make Millions and found that a 0.1-second improvement in mobile site speed was associated with retail conversion rates 8.4% higher and average order value 9.2% higher. Travel-site conversions ran 10.1% higher; lead-generation bounce rates 8.3% lower. The increment behind those figures lasts about as long as a blink.

The pattern predates Core Web Vitals. Akamai's 2017 State of Online Retail Performance report, built on roughly 10 billion retail site visits, found a 100-millisecond delay in load time associated with a 7% drop in conversion rates, and a two-second delay associated with bounce rate, the share of visitors who leave after viewing a single page, up 103%.

Visitors also leave earlier than the old three-second rule of thumb suggests. Google and SOASTA's machine-learning analysis of 11 million landing pages (2017) put the probability of a bounce 32% higher when mobile load time grows from 1 second to 3 seconds, and 123% higher at 10 seconds. Companion research from Google found 53% of mobile visitors abandon a page that takes longer than 3 seconds to load. For retail and e-commerce sites, each abandoned session is a product page shown to no one.

Core Web Vitals are three numbers Google collects from the Chrome browsers of real visitors to your site. They measure the three delays a person can feel: how long the main content takes to appear, how long the page takes to react to a tap, and how much the layout jumps around while loading. A page passes when at least 75% of real visits hit the good threshold on all three.

• Largest Contentful Paint (LCP) measures how long the largest visible element, usually the hero image or main headline, takes to render. Good is 2.5 seconds or less; over 4 seconds is poor.
• Interaction to Next Paint (INP) measures how long the page takes to respond visibly after a tap, click, or key press. Good is 200 milliseconds or less; over 500 milliseconds is poor. INP replaced First Input Delay as a Core Web Vital in March 2024.
• Cumulative Layout Shift (CLS) is a unitless score for how much content unexpectedly moves while the page loads, the effect that makes you tap the wrong button. Good is 0.1 or less; above 0.25 is poor.

Field data versus lab data

PageSpeed Insights, Google's free testing tool at pagespeed.web.dev, reports two kinds of results, and the distinction settles most arguments about speed. The top section is field data: measurements from real visitors' Chrome browsers, aggregated in the Chrome User Experience Report (CrUX) over a rolling 28 days. Google grades your site on this. The Lighthouse score below it is lab data, a single simulated run on standardized hardware. Lab data diagnoses problems; field data delivers the verdict. A 95 Lighthouse score with failing field data still fails.

Passing is less common than the tooling's green circles imply. Per the HTTP Archive's Web Almanac 2025, which draws on CrUX data across millions of sites, only 48% of websites deliver a good Core Web Vitals experience on mobile, and 56% on desktop. A site that merely passes beats roughly half the web.

Speed statistics circulate freely; controlled evidence is scarcer. The table below labels each result by rigor. In an A/B test, half of visitors get the fast version and half get the slow one, with everything else identical, so any revenue difference can only come from speed. Before-and-after comparisons and cross-site observational studies are weaker: redesigns, seasonality, and site mix can leak into the numbers. Read the A/B rows as causal evidence and the rest as consistent correlation at very large scale.

Two details stand out. Vodafone's test held everything constant except rendering speed, so its 8% sales lift prices a 31% LCP improvement in isolation, and Rakuten 24's experiment did the same for a full Core Web Vitals pass. The before-and-after rows come from Google's performance documentation and business-impact case studies, and the observational studies from Deloitte, Akamai, and Portent agree with the experiments in direction. That agreement across methods is the strongest reason to treat speed as a revenue input rather than an engineering vanity metric.

Yes, though modestly. Google's page experience documentation states that Core Web Vitals 'are used by our ranking systems', and cautions in the same breath that good scores do not guarantee good positions. Relevance still dominates. Speed behaves like a tiebreaker between comparable pages, and like table stakes on competitive queries.

The financial case never depended on rankings. A slow page loses visitors you already paid to acquire: ad clicks, email clicks, and referrals bounce before rankings enter the picture. And since only 48% of sites pass Core Web Vitals on mobile per the Web Almanac 2025, passing is itself a competitive position. In practice, speed work sits inside technical SEO, because the audit that surfaces slow pages tends to surface the indexing and structure problems that cost rankings on their own.

Yes, because websites gained weight faster than devices gained power. The Web Almanac's 2025 performance chapter reports that Total Blocking Time on mobile, a lab measure of how long scripts freeze the page during load, rose 58% year over year. Teams optimize the headline metrics while shipping more JavaScript underneath them.

Averages also hide the audience. Mobile devices generate roughly 60% of global website traffic (59.6% as of September 2025, per Statista's StatCounter series), much of it on mid-range hardware over variable cellular connections. Your office Wi-Fi and current-year laptop are the edge case. A quick check on your own phone proves little either: you are testing cached assets on a strong signal with newer hardware than your median visitor owns. Field data exists to settle exactly this argument.

Start with images, because the largest element on most pages is a picture and Largest Contentful Paint usually measures it. Per the Web Almanac 2025, 57% of LCP hero images are still JPEG and 26% are PNG, against 11% WebP and 0.7% AVIF. AVIF, an image format derived from the AV1 video codec, produces files 40-50% smaller than an equivalent-quality JPEG in independent tests. Netflix's engineering team found it beat both JPEG and WebP on compressed size at equal quality, with the largest render-time gains on slow connections. All major browsers support it. Most of the web has not claimed the cheapest speed win available.

The second shift is where pages get built. Static site generation (SSG) builds pages into finished HTML ahead of time instead of assembling each one on a server per request. A content delivery network (CDN), a set of servers that stores copies of your site around the world, then serves that HTML from a location near each visitor. Together they remove most server wait time before a byte of your page reaches the browser. Sites on modern frameworks (Next.js, Astro, Hugo, Eleventy) tend to be fast by default rather than fast by heroic optimization; it is the architecture we default to in our own web development work.

The practical consequence: a rebuild on a modern stack often delivers the improvements in the table above as a side effect of the architecture. The reverse holds too. No amount of tuning rescues a page that ships several megabytes of JavaScript ahead of its content, which is how mobile Total Blocking Time rose 58% in a year while hardware kept improving.

The returns are front-loaded. Portent's 2022 analysis of more than 100 million page views across 20 sites found B2B pages loading in 1 second associated with 3 times the conversion rate of pages loading in 5 seconds, and 5 times the rate of pages loading in 10. The same data shows the curve flattening sharply after the first few seconds: the early gains, moving a slow page to a merely quick one, are where the money is, while shaving fractions of a second off an already-fast page rarely covers the engineering time it costs.

That curve implies a spending order:

1. Images first. Convert hero and product images to AVIF or WebP, size them to their displayed dimensions, and lazy-load everything below the first screen. Heroes still ship as JPEG or PNG on 83% of pages per the Web Almanac 2025, so the format change alone puts you ahead of most competitors.
2. JavaScript second. Remove or defer third-party scripts, tag managers, and chat widgets. Script weight is what drove mobile Total Blocking Time up 58% year over year.
3. Delivery third. Cache pages and serve them through a CDN so distant and repeat visitors skip the server round trip entirely.
4. Micro-optimizations last. Font preloading and critical-CSS inlining are real improvements that belong at the end of the list, after the first three have moved the field data.

Measure the result the way Google does. PageSpeed Insights shows the 28-day field record for any URL with enough traffic; check it before a change and again a month after. Our free SEO report runs the same field and lab checks in one pass and orders the fixes by impact. Either route, the passing standard is identical: 75% of real visits meeting all three Core Web Vitals thresholds.