The One Speed Experiment That Actually Proves Causation

You have probably heard some version of this claim: “Speed affects conversions. Faster sites make more money.” Maybe you nodded along. Maybe you filed it under “vendor marketing” and moved on.

If you filed it under vendor marketing, you were more right than wrong. Most speed-revenue claims are correlation data, two things moving together in a large dataset, with no experimental control over what else changed. That is useful directional information. It is not proof.

There is one public experiment in the speed-revenue literature that is different. It controls for speed as the only variable and measures the sales outcome. That experiment is what this article is about.

The Evidence Class Problem

Before getting to the experiment, it is worth naming the distinction that most speed content collapses.

Correlation is when two variables move together in a dataset. Shopify’s 2026 analysis of all actively-selling Shopify stores over a 28-day window found that every 100ms slower LCP: Largest Contentful Paint, the time for your page’s main content to appear on screen, correlates with approximately 3.5% lower conversion rate. At the extreme end, stores with 2.5-second LCP show roughly 30% lower conversion than stores with 1.5-second LCP.

That is a meaningful, large-sample finding from a credible publisher. It tells you the direction with confidence. What it cannot tell you is whether speed caused the conversion difference, or whether slower stores also have weaker brands, thinner ad budgets, worse product assortment, or less rigorous checkout flows. All of those confounders exist in a real-world dataset where you did not control what changed and what did not.

Controlled experiment is different. You take the same traffic source, split it randomly into two groups, change exactly one thing for one group, and measure the difference. If you did it right, and the split was genuinely random, the sample was genuinely large, and you changed exactly one thing, the outcome difference can be attributed to the thing you changed.

That is a much harder bar to clear. Almost nothing in the speed-revenue literature clears it.

The Vodafone Experiment

Vodafone ran a controlled A/B test on a single landing page. Google published the case study on web.dev in 2021.

The setup: two groups of visitors, split from the same traffic source, arriving at the same landing page with the same offer. One group received the faster version of the page, the test variant. One group received the original version, the control. The only difference between the two versions was page speed, specifically a 31% improvement in LCP.

LCP is what the visitor experiences as “the page loaded”, the moment the main content appears. A 31% improvement means the test variant loaded its main content significantly faster than the control. Everything else, the offer, the layout, the copy, the calls to action, remained unchanged.

The result: the group that received the faster page converted at a measurably higher rate. The specific outcomes Vodafone and Google documented: +8% sales, +15% lead-to-visit rate, +11% cart-to-visit rate.

Vodafone’s landing page was processing approximately 100,000 clicks and 34,000 visits per day. This is not a boutique test with 500 sessions that could easily be noise. The sample size is large enough that the directional result is worth taking seriously.

This is what causation actually looks like in an experiment: one variable changed, one outcome measured, no confounders introduced. You cannot say “it could have been the product” or “it could have been the ad creative” or “Vodafone has strong brand recognition.” Those factors were held constant. Speed was the variable.

What the Deloitte Study Was: And Was Not

Before the Vodafone test became the clearest public evidence, the Deloitte / Google “Milliseconds Make Millions” study, published in 2020, using data collected from 37 European and American brand sites in late 2019, was the most-cited rigorous study in the speed-revenue space. I am citing it here with its date label: this is 2020 data, using end-of-2019 measurements. It is historical.

What it found: a 0.1-second improvement in mobile site speed was associated with an 8.4% increase in conversions and a 9.2% increase in average consumer spend across retail sites. In the luxury segment specifically, a 0.1-second improvement was associated with a 40.1% increase in progression from product detail page to “add to basket.”

This was a rigorous multi-site study with controlled monitoring periods. It is not the same as an A/B test. Deloitte monitored sites that were implementing speed improvements and compared before-and-after outcomes. There was no random assignment of visitors to fast and slow variants, which means other things may have changed alongside speed during the monitoring window. It is a strong observational study. It is not controlled-experimental evidence.

The distinction matters. Deloitte’s result is consistent with Vodafone’s, both show speed improvement associated with better conversion outcomes. But Vodafone’s design rules out confounders in a way Deloitte’s cannot.

What This Means for a Tampa Bay SMB

Here is where an honest extrapolation is required, and where you should hear the caveat alongside the number.

Vodafone’s A/B test was run on a single landing page processing 100,000 clicks per day. The statistical precision of an A/B test depends on traffic volume: larger traffic samples produce smaller margins of error around the result. At 100,000 clicks per day, Vodafone’s +15% lead-to-visit improvement is a very stable finding. At 500 monthly landing page visits, which is a realistic SMB volume, running the same test yourself would require months of traffic to achieve statistical significance. You would not be able to replicate the Vodafone experiment at SMB scale with confidence.

That does not mean the mechanism is irrelevant to you. The mechanism, that a faster LCP produces fewer premature exits before the visitor reaches the conversion point, operates the same way on a 500-visit page as it does on a 100,000-visit page. Physics does not change with company size. What changes is your ability to measure the effect with certainty at your traffic volume.

So here is a directional illustration, labeled explicitly as an extrapolation from a different-context experiment to a different-scale business:

A Tampa Bay B2B technology services firm has 500 monthly landing page visitors and a 3% lead conversion rate. At baseline, that is 15 leads per month.

If a 31% LCP improvement tracked directionally with Vodafone’s +15% lead-to-visit result, an extrapolation from a single enterprise-scale experiment to an SMB landing page, that would imply approximately 2 additional leads per month.

That extrapolation carries real uncertainty. Vodafone’s result applied to a specific industry, a specific offer, a specific traffic source. Your results would depend on your content, your visitor intent, your competitive alternatives. The directional sign, faster page, more leads, is what the evidence supports. The specific magnitude is not something you can read directly from Vodafone’s numbers.

Two additional leads per month may be significant for a B2B services firm with meaningful deal values, or it may be modest. The point is not the exact number. It is that the mechanism is real, the causal evidence exists for one experiment, and the question worth asking is what a version of that test on your landing page would reveal.

”Vodafone Is a Telecom Giant. This Doesn’t Apply to My Firm.”

The A/B test was on a single landing page, not Vodafone’s entire infrastructure, not their global brand presence, not their enterprise sales process. The thing that was tested was a web page. Your business also has web pages.

The mechanism under test was LCP: whether the page’s main content appeared fast enough that visitors stayed and converted. That mechanism does not require enterprise scale to operate. A visitor landing on your services page who waits three seconds for the hero content to appear makes the same neurological decision as a visitor to Vodafone’s landing page, they are assessing whether the page is loading or broken.

The specific conversion lift Vodafone observed is not transferable to your account as a prediction. Statistical significance at that traffic volume cannot be claimed at your traffic volume. But the fact that controlled speed improvement caused measurable conversion improvement, not just correlated with it, is the most useful thing this evidence base offers.

Why the Distinction Matters

If you are evaluating a paid engagement to improve your site’s speed, the quality of the evidence you base that decision on matters.

Correlation data from Shopify’s 2026 analysis tells you direction: faster stores convert better. It does not tell you whether fixing your speed will produce that effect, because it cannot rule out that faster stores are simply better-run businesses in every dimension.

The Vodafone A/B test tells you that fixing LCP on a single landing page, while holding everything else constant, caused a measurable sales improvement in one controlled experiment. It is the one piece of evidence in this literature where the word “caused” is warranted.

That is why I reference it specifically rather than citing the many correlation studies that are more recent and more numerous. Correlation data is useful for understanding trends. One causal study is worth more for evaluating a specific investment decision.

What Your Landing Page Actually Looks Like

The free assessment at iServU starts with your actual page speed measurements, not Lighthouse lab scores but real CrUX field data showing what your visitors actually experienced. Then it maps those measurements against the evidence classes: where you have clear causal evidence that speed matters, and where you have directional signals that are useful but not proof.

Gabriel Lopez-Seco runs the assessment directly. No account manager, no handoffs to a junior analyst.

Schedule your free assessment at iServU.

Sources

1. Google web.dev / Vodafone (2021). Vodafone case study, controlled A/B test; 31% LCP improvement; +8% sales, +15% lead-to-visit rate, +11% cart-to-visit rate; ~100K clicks and 34K visits/day. https://web.dev/case-studies/vodafone
2. Shopify (2026). Store Speed and Conversion, correlational analysis, 28-day Jan–Feb 2026, all actively-selling Shopify stores; −3.5% conversion per 100ms slower LCP; −30% conversion for 2.5s vs 1.5s LCP. https://www.shopify.com/enterprise/blog/store-speed-conversion
3. Deloitte / Google (2020, HISTORICAL, end-of-2019 data). Milliseconds Make Millions: 37 European/American brand sites; 30M sessions; +8.4% conversions and +9.2% consumer spend per 0.1s improvement; +40.1% PDP-to-basket progression in luxury. https://web.dev/case-studies/milliseconds-make-millions