Conversion Science · established evidence
Milliseconds Make Millions: What a 30-Million-Session Study Says a Tenth of a Second Is Worth
The relationship between site speed and conversion is one of the most studied questions in digital retail, and the largest public dataset on it is unambiguous about direction. In 2020 Google commissioned the agency 55 and Deloitte Digital to analyze more than 30 million real user sessions across 37 retail, travel, and lead-generation brand sites. A 0.1 second improvement in mobile load speed was associated with an 8.4 percent increase in retail conversions and a 9.2 percent lift in average order value; travel conversions rose 10.1 percent. One caveat the headline usually drops: the study was funded by the company that sells speed as a ranking signal, and it measured association, not proven cause. Even discounted for that, the size and consistency of the effect make site speed one of the few conversion levers with real evidence behind it, and one most sites are still leaving on the table.
What the study actually measured
The report, titled "Milliseconds Make Millions," was published in 2020. Google commissioned it; the media agency 55 and Deloitte Digital carried out the analysis. That provenance matters and is returned to below, but the scale of the underlying data is what makes the study worth reading closely in the first place.
The analysis drew on more than 30 million real user sessions across 37 European and American brand sites, spanning retail, travel, and lead-generation businesses. Rather than a lab benchmark run on a single test page, it used field behavior from actual visitors on live commercial sites. The independent variable was mobile load speed, and the unit of change the report headlined was small on purpose: a single tenth of a second.
The dependent variables were the outcomes a business actually cares about, measured per vertical: conversion rate, average order value, page views, and bounce rate. The finding was not that fast sites happen to be run by good companies; it was that within the dataset, incremental speed improvements tracked incremental commercial gains at a consistent slope.
The per-vertical numbers, without rounding up
The report broke its results out by business type rather than reporting one blended figure, which is the correct way to read it. A tenth of a second does not do the same work for a luxury retailer, an airline, and a lead-capture form, and the study did not pretend it did.
- Retail: a 0.1 second improvement in mobile load speed was associated with an 8.4 percent increase in conversions and a 9.2 percent increase in average order value.
- Travel: the same 0.1 second improvement was associated with a 10.1 percent increase in conversions.
- Lead generation: bounce rate improved by 8.3 percent, the mechanism by which faster pages keep more visitors in the funnel long enough to convert.
Why average order value moves too
The 9.2 percent order-value figure is the more interesting of the two retail numbers, because it points at something beyond "more people finished checkout." A faster experience appears to change whether a session converts and what a converting session is worth. The plausible reading is that friction is heaviest on the deeper, higher-consideration parts of a purchase, the extra item added, the upgrade considered, the larger basket, and that removing lag preserves the patience those decisions require. The study reports the association; it does not isolate the psychological mechanism, so this remains an interpretation rather than a proven pathway.
Why a tenth of a second is enough to matter
A tenth of a second sounds too small to change a buying decision, and on any single visit it is. The effect is not that one shopper consciously abandons a cart because a page took 100 extra milliseconds. It is that speed acts as a tax levied on every session, and the losses compound across the whole population of visitors.
On mobile, where connections are variable and attention is thin, each increment of delay raises the probability that a given visitor leaves before the page becomes useful. The study's 8.3 percent bounce-rate improvement for lead-generation sites is the visible edge of this: faster pages simply retain a larger fraction of the people who arrive. Multiply a small per-session retention gain across millions of sessions and a modest percentage becomes a material revenue line, which is precisely the arithmetic the report's title is pointing at.
This is also why speed is a structural lever rather than a persuasion tactic. It does not try to talk a hesitant visitor into converting. It removes a cost that was quietly suppressing conversion for everyone, including the visitors who were already inclined to buy.
How to read a study the seller paid for
The most important thing to say about this research is the thing its headlines omit: Google commissioned it, and Google has a commercial interest in businesses treating speed as a first-order priority, because speed is one of the signals its own systems reward. That does not make the numbers wrong. It does mean they should be read as corroborating evidence rather than as fully independent proof.
Two further limits belong in any citation of this study. First, the study reports association, not causation. It observed that faster sessions and better commercial outcomes moved together across a large dataset; it was not a randomized controlled experiment that held everything else constant and varied only speed. Confounding is possible: sites that invest in speed often invest in everything else too. Second, the effect sizes are population averages across 37 large brands in specific verticals. They are a strong directional signal, not a coefficient you can multiply against your own traffic to forecast a guaranteed return.
Read with those three caveats in place, funded, correlational, and averaged, the study still clears a high bar. Thirty million sessions is a large sample, the per-vertical breakdown is disciplined, and the direction of the effect agrees with a wider body of independent web-performance evidence. The correct posture is neither to quote the 8.4 percent as a promise nor to dismiss the finding because of who paid for it.
The corroborating case studies, and their limits
The multi-brand association study does not stand alone. Google's web.dev Core Web Vitals case-study series documents named before-and-after results from individual companies that optimized specific speed metrics.
- Rakuten 24 reported a 33.13 percent increase in conversion rate and a 53.37 percent increase in revenue per visitor after improving Largest Contentful Paint.
- Vodafone Italy reported an 8 percent increase in sales after improving Largest Contentful Paint by 31 percent.
- redBus reported a 7 percent increase in sales after improving Interaction to Next Paint, the metric that measures how quickly a page responds to a tap.
Why these prove less than they appear to
Each of these is a single company observing its own before-and-after, not a controlled experiment across a population. Other things changed on those sites in the same period, and companies that publish these case studies are self-selected toward success stories. Treat them as existence proofs, evidence that the mechanism is real and can be large, rather than as elasticities you can expect to reproduce. That is why this layer of evidence is tiered emerging while the 30-million-session study is tiered established.
Mapping the finding to the standard your site is graded on today
The 2020 study measured mobile load speed at a moment before the current field-data standard was fully formalized. Today the speed that search engines and buyers experience is measured through Core Web Vitals, graded at the 75th percentile of real users. Three thresholds define a passing experience: Largest Contentful Paint at or under 2.5 seconds, Interaction to Next Paint at or under 200 milliseconds, and Cumulative Layout Shift at or under 0.1.
The translation from the study to this standard is not one-to-one, and it should not be sold as though it were. What carries across is the principle the data established: on real mobile sessions, incremental speed tracks incremental commercial outcome, and the increments that matter are small. A page can look finished, score well in a lab tool, and still fail the field-data check that actually reflects what visitors feel. The way to act on "milliseconds make millions" is to measure your own page against these thresholds and fix what is failing, not to quote the percentage.
Turning the evidence into a number for your business
The wrong way to use this study is to multiply your revenue by 8.4 percent and call it a forecast. The study does not support that use. The right way is to use the evidence to size the opportunity and then measure your own baseline before promising anything.
Start with the direction, which the evidence supports firmly: if your site is slow on mobile and failing its Core Web Vitals, speed is very likely suppressing conversions and order value right now, silently, across every visitor. Then replace the population averages with your own numbers. Measure where your real pages sit against the 75th-percentile thresholds, estimate the revenue tied to the sessions you are losing to delay, and scope the fix against that, not against a headline percentage borrowed from 37 other brands.
That sequence, direction from the evidence, magnitude from your own field data, is the difference between an ROI framing you can defend and a growth-hacking promise you cannot. Speed is one of the few conversion levers where the underlying research is genuinely strong. It deserves to be acted on with the same discipline the good research was conducted with.
The evidence
Key findings, with their sources
-
A 0.1 second improvement in mobile load speed was associated with an 8.4% increase in retail conversions.
established Google / 55 / Deloitte Digital, "Milliseconds Make Millions", 2020 (>30 million sessions across 37 brand sites).
-
The same 0.1 second improvement was associated with a 9.2% increase in average order value for retail.
established Google / 55 / Deloitte Digital, "Milliseconds Make Millions", 2020.
-
For travel sites, a 0.1 second speed improvement was associated with a 10.1% increase in conversions.
established Google / 55 / Deloitte Digital, "Milliseconds Make Millions", 2020.
-
For lead-generation sites, bounce rate improved by 8.3% with the speed improvement.
established Google / 55 / Deloitte Digital, "Milliseconds Make Millions", 2020.
-
The study analyzed more than 30 million real user sessions across 37 European and American retail, travel, and lead-generation brand sites.
established Google / 55 / Deloitte Digital, "Milliseconds Make Millions", 2020 (methodology).
-
In a single-company case study, Rakuten 24 reported a 33.13% conversion-rate increase and a 53.37% increase in revenue per visitor after improving Largest Contentful Paint.
emerging Google web.dev Core Web Vitals case-study series (N=1 before/after, not a controlled experiment).
Calibration
What is proven, what is promising, what is unproven
| Evidence tier | Tactics | What the evidence says |
|---|---|---|
| Established | Direction and rough magnitude of the speed-to-conversion relationship on mobile; per-vertical association figures. | "Milliseconds Make Millions" (Google / 55 / Deloitte Digital, 2020): 30M+ sessions, 37 brands. Large-N field data. Google-commissioned, so corroborating rather than fully independent, and correlational rather than causal. |
| Emerging | Specific large lift figures from named companies (Rakuten 24, Vodafone Italy, redBus) tied to individual Core Web Vitals improvements. | Google web.dev case-study series. Real and publicly documented, but each is a single before/after with no control, so an existence proof, not a generalizable elasticity. |
| Inferential | Reading a specific percentage as a causal effect you can forecast against your own revenue. | Not supported. The study observed association across a population; multiplying a headline lift against your traffic is a projection the data does not license. |
Reference
Glossary
- Largest Contentful Paint (LCP)
- The time it takes for the largest visible element of a page to render. A core measure of how quickly a page feels loaded; the passing threshold is 2.5 seconds or less at the 75th percentile of real users.
- Interaction to Next Paint (INP)
- A measure of how quickly a page visibly responds after a user taps or clicks. The passing threshold is 200 milliseconds or less; it is the most commonly failed Core Web Vital in 2026.
- Cumulative Layout Shift (CLS)
- A measure of how much a page unexpectedly jumps as it loads. The passing threshold is 0.1 or less; high CLS causes mis-taps and lost trust.
- Core Web Vitals
- Google's set of field-data metrics (LCP, INP, CLS) that grade the real loading, responsiveness, and visual stability a page delivers to actual visitors, measured at the 75th percentile.
- Average order value (AOV)
- The average amount spent per completed order. Speed appears to influence whether a session converts and how much a converting session is worth.
- Bounce rate
- The share of visitors who leave without meaningful interaction. Faster pages retain a larger fraction of arriving visitors, which is the visible mechanism behind speed-driven conversion gains.
Straight answers
Frequently asked questions
How much is 0.1 second of site speed really worth?
In the largest public study, a 0.1 second improvement in mobile load speed was associated with an 8.4% increase in retail conversions and a 9.2% increase in average order value, with travel conversions up 10.1%. Those are population averages across 37 brands and 30 million sessions, not a return you can guarantee for a specific site. Take the direction as well-evidenced, then measure your own baseline before projecting any number.
Should I trust a study that Google paid for?
Read it as corroborating evidence, not as fully independent proof. Google commissioned the report and benefits when businesses prioritize speed, and the study measured association rather than proven cause. Even so, the sample is large, the per-vertical breakdown is disciplined, and the direction of the effect agrees with independent web-performance research. The right posture is to neither quote the percentage as a promise nor dismiss the finding because of who funded it.
Does this apply to a small local business, not just big retail brands?
The direction almost certainly does; the exact percentages do not transfer. The study measured large retail and travel brands, so its numbers are specific to that population. For a local service business the same mechanism applies, a slow mobile page loses visitors before they book or call, but the way to size it is to measure your own pages against Core Web Vitals and the sessions you are losing, rather than borrowing a retail figure.
Is faster always better for conversion?
Faster is better up to the point where the experience feels instant, and the evidence is strongest for sites that are currently slow on mobile. If your pages already pass Core Web Vitals comfortably, further speed work yields diminishing returns and other levers matter more. The largest gains in the research come from moving a failing or borderline page onto the right side of the thresholds, not from shaving milliseconds off an already-fast site.
What speed targets should my site hit now?
The current field-data standard is Core Web Vitals, graded at the 75th percentile of real users: Largest Contentful Paint at or under 2.5 seconds, Interaction to Next Paint at or under 200 milliseconds, and Cumulative Layout Shift at or under 0.1. A lab speed score can flatter a page that real visitors still experience as slow, so measure against field data, not just a one-off lab test.
Provenance
Sources
- Google / 55 / Deloitte Digital, "Milliseconds Make Millions", 2020 (established; large-N industry study, Google-commissioned so corroborating rather than fully independent, and correlational not causal)thinkwithgoogle.com
- Google web.dev, Core Web Vitals case-study series (Rakuten 24; Vodafone Italy; redBus) (emerging; single-company before/after case studies, not controlled experiments)web.dev
- Google web.dev, Core Web Vitals thresholds (LCP, INP, CLS) at the 75th percentile (established; the current field-data standard)web.dev
Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.