Site Speed and Conversion Rate: Checking the Famous Stats

Site speed does affect conversion rate, but far less dramatically than the most quoted numbers suggest. "Every second costs 7% of conversions" comes from a 2008 survey of companies. "53% of visitors leave after 3 seconds" is 2016 Google data about publishers' mobile sites. "Amazon loses 1% of sales per 100 ms" is a line from a 2006 slide with no published data behind it. Controlled experiments, where one group of users got a deliberately slower or faster site, paint a more modest picture: effects of a few percent, most visible where pages take several seconds to load. Below I trace each popular number back to its source, show the studies worth trusting, and estimate how many extra leads a small site can realistically expect.

Site speed and conversion rate: why some numbers deserve trust and others don't

Nearly every number about speed and sales comes from one of three methods.

• Survey. Companies or users say what they think, or how they believe they would behave. That tells you about opinions, not behavior.
• Correlation. Conversion of fast and slow visits, pages or sites is compared. A slow visit differs in more than speed: different device, network, often country and page. Any of those differences can explain the gap in conversion.
• Controlled experiment. Users are randomly split into groups that differ only in speed. Then the difference in sales can be attributed to speed.

Where the famous site speed statistics come from

A summary of the seven most quoted numbers. Details on each follow.

"Every second of load time costs 7% of conversions"

This number has two unrelated sources, and articles mix them up.

The first is the Aberdeen Group report "The Performance of Web Applications: Customers Are Won or Lost in One Second", December 2008. According to its press release, it was a survey of more than 160 organizations: each extra second of delay could reduce page views by 11%, conversions by 7% and customer satisfaction by 16%. So the 7% is what companies reported in a questionnaire, not measured shopper behavior. The same release says business performance starts to suffer at 5.1 seconds of delay – a snapshot of the 2008 web.

The second is Akamai's online retail performance report from April 2017. The data was collected by SOASTA, which Akamai acquired that same month: a month of anonymized data from large online retailers, about 10 billion visits. The press release says a 100 ms delay can hurt conversion by 7%. According to MediaPost's write-up of the report, the 7% applies to smartphones, with tablets at roughly 4%. The method compares conversion at different load times, so it is a correlation.

The "7%" thus moved from "per second, survey" to "per 100 ms, correlation", and both versions circulate without saying which is which.

"53% of mobile users abandon a site that takes longer than 3 seconds to load"

The original is a September 2016 post on Google's DoubleClick blog (now Google Ad Manager), "The need for mobile speed", written for ad-funded publishers. The footnote describes aggregated Google Analytics data from roughly 3,700 mobile sites that opted into benchmark sharing, March 2016. The wording is cautious: visits are likely to be abandoned. The post doesn't explain how abandonment was defined or why the threshold is 3 seconds. For context, the same post says the average mobile page then took 19 seconds to load over 3G.

In spring 2017 Akamai repeated the number in its own press release, and today you can find it attributed to Akamai. That's how a statistic loses its source: in 2026 it gets applied to desktop sites, 4G and 5G, e-commerce and B2B, while what was measured was publishers' mobile sites in 2016.

It usually comes paired with "the probability of bounce increases 32% as page load time goes from 1 to 3 seconds" – Google and SOASTA, 2017. The Think with Google data page with this number exists, but the page that described the study now redirects to a general section. Third-party write-ups say the numbers came from a neural-network model; I couldn't verify that against the original.

"Amazon loses 1% of sales for every 100 ms of latency"

The number comes from slides of a talk Greg Linden of Amazon gave at a Stanford data mining course in December 2006; Kohavi's 2013 paper cites it too. By the descriptions, it was an experiment: pages were deliberately slowed in 100 ms steps. But sample size, duration and statistical significance were never published. In December 2025 performance consultant Martin Michálek called it a historical meme: the data is almost 20 years old, it describes the Amazon of 2006, and applying it to a blog or a B2B site is wrong.

"Walmart: +2% conversions for every second of improvement"

The source is a February 2012 deck by Walmart.com engineers, "Page Performance & Site Conversion". The WPO Stats catalog summarizes it as up to 2% conversions per second and up to 1% revenue per 100 ms. Retellings tend to drop "up to", although it means a maximum, not an average.

In the slides I could access, the method is real user monitoring (RUM) and a comparison of load times for buyers versus non-buyers. That is a correlation: a buyer might see a fast site because of a good phone and network, rather than buy because the site is fast. The deck also says that at baseline half of visits had response times over 4 seconds, so the starting point was slow.

"A 0.1 second improvement lifts conversion by 8.4%"

This is "Milliseconds Make Millions" (2020), commissioned by Google and carried out by the agency 55 and Deloitte. It covered 37 brand sites in Europe and the US and more than 30 million sessions. Mobile load times were tracked hour by hour for 30 days at the end of 2019 and matched against analytics metrics; no redesigns happened during the study. Results: retail +8.4% conversion and +9.2% spend, travel +10.1% conversion.

The condition behind the headline is often dropped: the effect was observed when every page of the journey improved by 0.1 seconds on four metrics at once, including server response time and First Meaningful Paint (which Google has since deprecated). Methodologically it is the strongest number in this section, but it is still observational: speed varied by itself, nobody was randomized into groups. Hours when a site is slow because of load or an ad campaign may also bring a different audience. And the sponsor was Google, which promotes speed.

"A site that loads in 1 second converts 3x better than one that loads in 5"

This is Portent's study, updated in 2022: 20 sites (6 e-commerce, 14 B2B lead generation), about 100 million page views over 30 days. For B2B, pages loading in 1 second converted 3x better than pages loading in 5 seconds; for e-commerce, 2.5x.

It compares different pages on different sites. A fast page is often a light landing page for a specific query, a slow one a heavy catalog, and their conversion differs for reasons unrelated to speed. The article doesn't say which load metric was used – readers ask in the comments – and its worked example uses a 2.93% conversion rate for 4 seconds instead of the 0.67% shown in its own chart. The correlation is strong, but it isn't an estimate of what speeding up will earn you.

What controlled speed experiments actually show

Studies where user groups differed only in speed are few, and all come from large companies. Their results are an order of magnitude more modest than the popular numbers.

• Google, 2009. Jake Brutlag described slowing search results by 100–400 ms. Searches per user dropped 0.2–0.6% on average over 4–6 weeks. The effect grew: with a 400 ms delay, searches fell 0.44% in the first three weeks and 0.76% in the next three. After the delay was removed, those users still searched 0.21% less for five more weeks. People remember that a site is slow, and it doesn't wear off immediately.
• Bing, 2013. Kohavi's team slowed 10% of users by 100 ms and another 10% by 250 ms for two weeks. Conclusion: every 100 ms is worth 0.6% of revenue. The conditions were special: Bing's server already responded in under a second for 95% of requests, and millions of users took part.
• Etsy. The same paper mentions that Etsy found no effect from a 200 ms delay; Kohavi's team argues the test most likely lacked statistical power – on less traffic, a small effect simply isn't visible.
• Financial Times, 2016. For about four weeks, subscribers saw the site with a 1, 2 or 3 second delay. With a 1-second delay they read 4.9% fewer articles in the first 7 days; with 3 seconds, 7.2% fewer. On mobile, 1 and 2 second delays showed no noticeable effect in the first week.
• Talabat. Michálek relays a recent case Kohavi published: time to interactive went from 2.1 to 1 second, revenue rose 0.3%. Small, but verifiable.

Vodafone: the one well-documented speed A/B test

An A/B test of a landing page receiving paid traffic: half of visitors got the optimized version, half the original, about 34,000 visits a day each. The versions looked and worked the same; only server-side rendering, images and loading order changed. LCP (Largest Contentful Paint, the time to render the main element) for real users improved by 31%.

Sales rose 8%, the lead-to-visit rate 15%, the cart-to-visit rate 11%. Absolute values are redacted, and test duration and significance aren't given. One more detail: DOMContentLoaded actually got 15% worse in the faster version, so which speed metric you improve matters.

Separately, Chrome's 2020 analysis of millions of page loads found that when a site meets the Core Web Vitals thresholds, users are 24% less likely to abandon a page before it finishes loading (22% for news sites, 24% for shopping sites). That's about leaving before load, not about purchases – but those visits certainly don't make money.

The overall picture: in controlled conditions, changing speed moves target metrics by anywhere from a fraction of a percent up to +8% sales and +15% leads in Vodafone's case. Large effects appear where the page was slow (Vodafone's LCP was over 8 seconds). On a fast site the next 100 ms are worth fractions of a percent, which is big money only at Bing's scale.

How site speed affects small business sales

It depends on how slow you are now

All of these effects were measured relative to each site's existing speed. Moving "0.6% per 100 ms" from Bing, which responds in under a second, to a site whose first screen appears after 6 seconds tells you nothing about that site's payoff. Vodafone is closer to a typical small business site: a slow mobile landing page for ads, heavy images, client-side rendering. There, shaving a few seconds yielded +8% sales.

What it means in leads: a calculation

My illustrative calculation. A site gets 300 visits a day, about 9,000 a month. Conversion to a lead is 2%, about 180 leads a month. The chart shows how many leads you gain if speeding up lifts conversion within the range seen in experiments. Move the slider on the last bar to model your own scenario.

Then it's the owner's arithmetic: multiply the extra leads by what one lead from ads currently costs you and compare with the price of the speed work. For a slow ad landing page, where a Vodafone-like scenario is realistic, the math usually favors speeding up. For a site that already loads fast, expect the low end, which may not pay off.

Why a small site can't prove this with an A/B test

I calculated how long an A/B test has to run to reliably detect a conversion lift at a 2% baseline. A 5% lift needs about 315,000 visitors per group: at 300 visits a day, that's almost six years.

That matches the rule from Kohavi's paper: detecting an effect 10 times smaller takes 100 times more users. Speed effects are exactly the kind that need large samples, which is why Google, Bing and Vodafone prove them, not a shop with 300 visits a day. Etsy's story shows the opposite trap: an underpowered test "found no effect", which is easy to mistake for proof that speed doesn't matter.

How to estimate the impact of speed on your own site

• Look at speed for real users, not the Lighthouse score. Vodafone measured LCP with PerformanceObserver in visitors' browsers and sent it to analytics. A lab run on a powerful machine shows what can be improved, not what customers experience.
• Start where slow pages meet money. Mobile traffic, ad landing pages, the page with the form. Google's 2016 data and the Vodafone case are both about mobile and paid visits.
• Fix seconds, not milliseconds. If the first screen appears after 5–8 seconds, expect an effect. If it appears after 1.5 seconds, the next 200 ms are unlikely to pay off for a small site.
• If you have the traffic, test like Vodafone. Two versions that look identical, differing only in optimization, with traffic split randomly and simultaneously.
• If traffic is low, compare before and after carefully. One traffic source, matching days of the week, no parallel changes to ads or the site. Treat the result as an estimate, not proof.
• Don't trust cases where speed changed together with design. Speed's share of such a result is unknown.

Summary

• "−7% per second", "53% after 3 seconds", "Amazon 1% per 100 ms" and "Walmart 2% per second" are a survey, correlations and old experiments without published data. They can't forecast your site's revenue in 2026.
• Controlled experiments (Google, Bing, Financial Times, Vodafone) confirm that speed affects behavior and money, but effects are fractions of a percent to a few percent; the largest among those found is +8% sales at Vodafone.
• Slow mobile pages for paid traffic gain the most: at Vodafone, LCP going from 8.3 to 5.7 seconds brought +8% sales.
• A small site can't prove a speed effect with an A/B test: a 5% conversion lift needs about 315,000 visitors per group.
• Treat speed work as an investment: expected extra leads times their value, compared with the cost of the work.