Search
Speed is a revenue number
Site performance gets filed under engineering housekeeping. The measured effect on conversion says it belongs on the finance side of the conversation.
7 min read
Page speed is usually discussed as a technical score to be improved for its own sake, somewhere between correct heading levels and valid markup. The measurements do not support treating it that way.
~7%
Conversion lost to a single second of additional load time
33%
Conversion increase Rakuten measured after passing all three Core Web Vitals
42%
Share of mobile sites that pass all three
The evidence is unusually clean
Performance is one of the few areas of web work where the causal evidence is strong, because it can be A/B tested directly. Vodafone Italy ran two pages that were visually and functionally identical except that one loaded with a 31% better largest contentful paint. The faster page produced 8% more sales. Nothing else differed.
The pattern repeats at finer granularity. Retail measurements put roughly 8% more conversions on every 0.1 second of improvement. Product pages served with a two-second largest contentful paint convert dramatically better than the same pages at four to five seconds.
Two pages, identical in every respect a visitor could describe. One loaded faster and sold 8% more.
Why so many sites are slow
Fewer than half of mobile sites pass all three Core Web Vitals, and the causes are rarely mysterious.
- Images shipped at the wrong size
- The single most common cause. A photograph exported at full camera resolution and scaled down in the browser costs the visitor the whole download.
- Third-party scripts nobody audits
- Analytics, chat widgets, heat mapping, three abandoned tracking pixels. Each was added for a reason, none were removed when the reason expired.
- Everything loading at once
- Work that could wait until after the page is usable competes with the work that makes it usable. Deferring what is not needed immediately is often the largest single win available.
- Layout that moves while loading
- Content jumping as images and fonts arrive is measured, penalised, and — more to the point — makes people mis-tap and leave.
Where to look first
- 01
Measure the field, not the lab
A test from a fast connection on a good laptop flatters everyone. Real user data from actual devices is the number that correlates with revenue.
- 02
Fix the largest element on the page
Largest contentful paint is usually one hero image or one heading. Making that one thing arrive sooner moves the metric more than a dozen small optimisations.
- 03
Remove the scripts nobody owns
List every external script, name what it does and who reads its output, and remove the ones with no answer. This is usually the fastest win and it costs nothing.
- 04
Defer whatever is not needed for first paint
Anything decorative or below the fold can wait until the main thread is free. The visitor sees no difference; the measurement does.
Performance is part of every build rather than a phase after it.
How Build worksFollow-up
Questions this raises.
- What counts as a good largest contentful paint?
- Google's threshold for a good LCP is 2.5 seconds, measured at the 75th percentile of real visits. The gap worth closing is between slow and fast rather than between fast and perfect — the last few points of a score routinely cost more than they return.
- Does page speed actually affect Google rankings?
- It is a ranking signal, but a modest one — relevance beats speed. The stronger argument is commercial rather than positional: the measured effect of speed on conversion is far larger than its effect on where you rank.
- Why does my score change every time I test it?
- Lab tests simulate one device on one connection and vary run to run. Field data — what actual visitors experienced — is the number that correlates with revenue, and it is the one worth tracking. A single lab score is a diagnostic, not a measurement.
- What usually makes a site slow?
- Images shipped far larger than they display, third-party scripts nobody has audited in years, everything loading at once instead of deferring what is not needed, and layout that shifts as fonts and images arrive. The first two account for most of it.