Why ‘Fast’ Websites Lose Money: The Counterintuitive Science of E-Commerce Speed

Stop Chasing Speed Scores. Start Chasing the Second Your Customer Gives Up

Everyone quotes the same handful of speed statistics. Here is what the research actually supports, what the best retailers did differently, and the mindset that turns milliseconds into revenue.

Here is the lie doing the rounds in e-commerce: shave 100 milliseconds off your site and earn 1% more sales.
Clean. Simple. Perfect for a slide deck. And almost nobody repeating it can show you the experiment behind it.

Speed matters. It matters a lot. But the number you borrowed from somebody else’s store is not your number. The retailers who win here have stopped worshipping a score and started obsessing over one moment: the second a customer decides this isn’t worth the hassle.

Let’s fix your thinking first. Then your tactics.

The magic number doesn’t exist

The famous claim that every 100 milliseconds costs Amazon 1% of sales gets quoted endlessly. The sample and method behind it have never been fully disclosed, so treat it as a story, not a forecast.

The best-known real dataset is “Milliseconds Make Millions”, a Deloitte study commissioned by Google. It covered 37 brands across Europe and the US, with more than 30 million mobile sessions analysed over four weeks.
A 0.1-second improvement in mobile speed was associated with:

8.4% more retail conversions
9.2% higher average order value
10.1% more travel conversions
9.1% more retail progression from product page to add-to-basket

Impressive. Now the bit most people skip. Those speed gains happened naturally; the study was observational, and Google paid for it. It tells you speed and sales travel together. It does not promise your store the same lift.

Farfetch shows exactly why that distinction matters. Their own analysis linked every extra 100 milliseconds of Largest Contentful Paint to roughly a 1.3% drop in conversion. Then they went and fixed product-page image loading, cutting load time by more than 600 milliseconds, and ran an A/B test.
The lift? Between 1% and 5%. A genuine win, and a more modest one than the correlation suggested.

Same company. Same metric. Two different answers.

The contrarian bit: a relationship you observe is not an effect you can buy. Stop asking “what does speed do for conversion?” and start asking “what does it do for mine?” Only your own test can tell you.

What the winners actually did

Vodafone built two landing pages that looked and worked identically, split traffic 50:50, and made one of them faster. The optimised version had a 31% better Largest Contentful Paint. The result: 8% more sales, a 15% higher lead-to-visit rate and an 11% higher cart-to-visit rate. One conventional technical timing actually moved the wrong way while sales went up.
Lesson: they judged success by what customers experienced, not by whichever metric was easiest to screenshot.

Rakuten 24, the online retail arm of Japan’s Rakuten Group, ran an A/B test focused on Core Web Vitals and reported a 33.13% higher conversion rate and 53.37% higher revenue per visitor.

The public write-up doesn’t share sample sizes or confidence intervals, so read it as proof of what’s possible, not a number to put in front of your board.

See the pattern?

They held everything else steady.
They split traffic and compared.
They measured money, not just milliseconds.

That’s the whole secret. It isn’t glamorous. It works.

Speed isn’t one thing. It’s three.

Most teams say “speed” and mean “page load”. Your customer experiences three different problems:

Loading: how fast the meaningful content appears. Google’s guide for a good experience is a Largest Contentful Paint of 2.5 seconds or less.

Responding: how fast the site reacts to a tap, filter or button. The target is an Interaction to Next Paint of 200 milliseconds or less.

Stability: whether things jump around while people try to use them. The target is a Cumulative Layout Shift of 0.1 or less.

Those thresholds are judged on real visitors at the 75th percentile, not on one tidy lab test. And Nielsen Norman Group’s long-standing guidance is that around 0.1 seconds feels instant, while around one second is enough to break a person’s flow of thought.
Picture a shopper on their phone. The product page loads quickly. They tap their size, and nothing happens for a second. They tap again and get the wrong one. The page was fast. The experience was broken.

That’s the gap between a speed score and a customer.

Start where the money is

Checkout first.

A peer-reviewed study in Operations Research (Gallino, Karacaoglu and Moreno, 2023) found that slowdowns hurt online retail sales, and that customers are more sensitive to delays at checkout than at other stages. Not every page deserves equal love. The moments when someone is trying to hand you money deserve the most.

Checkout design matters well beyond speed, too. Baymard Institute’s usability research estimates the average large e-commerce site could lift conversion by around 35% through better checkout design.

Then product discovery.

Baymard’s 2025 benchmark assessed more than 21,000 product-list and filtering parameters across 170-plus leading sites. It rated product-list UX as poor to mediocre on 58% of desktop sites and 78% of mobile sites. That’s a usability benchmark, not a speed measurement, but it shows how much friction sits between “I’m browsing” and “I’m buying”. A filter that responds slowly, or returns the wrong products, is a leak either way.

Mobile needs its own plan

Don’t average your devices together and call it a strategy.

The 2025 HTTP Archive Web Almanack found that 48% of mobile websites met all three Core Web Vitals benchmarks, against 56% on desktop. The responsiveness gap is starker: 77% of mobile sites achieved a good Interaction to Next Paint score, compared with 97% on desktop. That data covers the wider web, not just online shops, but the message is clear. Smaller screens, touch controls and patchy connections change the game, so measure mobile on its own terms.

Fast and confusing still loses

Here’s another belief worth challenging: that speed fixes everything.

In a 2010 study (Schmutz and colleagues, International Journal of Human-Computer Studies), an online experiment with 199 participants compared product-listing formats. A list layout was linked to lower cognitive workload and more economically favourable choices than a grid. No speed intervention was involved. The point is that how you present options changes what people choose.

Make the page quick. Then make it easy to choose.

And where a wait genuinely can’t be removed? Research published in the Journal of Consumer Research (Ding and Kyung) found that moderate-speed loading animations made a fixed wait feel shorter than static, very slow or very fast alternatives. In one experiment with a 30-second wait, 67.21% of people carried on to complete a voluntary study after a moderate-speed animation, versus 32.20% after a static one. Important caveats: those were long waits and a survey, not a purchase. This is not a licence to slow your site down. Remove real waiting first, then make the unavoidable bits feel honest.

“Yeah, but…”

“Nobody’s attention span is longer than eight seconds anyway.”

The goldfish line is everywhere, and it isn’t a sound basis for decisions. There is no reliable universal attention-span number that predicts how long a shopper will tolerate your page. Eye-tracking research from Nielsen Norman Group shows people scan according to their goal and how the page is organised. Fix the task, not the goldfish.

“My speed score is green, so we’re done.”

A lab score is a rehearsal. Core Web Vitals are judged on real visitors, on real devices, on real connections. Vodafone’s test showed that a technical metric can slip while the customer experience and sales improve. Measure what shoppers feel.

“We shipped a speed fix, and sales went up. Speed works!”

Maybe. But an analysis of 1,184 observations across six e-commerce sites (Fatta and Lu, 2018) found that discounts and free shipping boosted conversion, with the effects moderated by seasonality and page-load speed. If your release landed during a promotion, you can’t separate the two without a control group.

“We don’t have the traffic to A/B test.”

Fair challenge. To detect a lift from 2.0% to 2.2% conversion (a 10% relative improvement) at the usual 5% significance level and 80% power, you need roughly 81,000 visitors per variant. At a 5.0% baseline, aiming for 5.5%, it’s roughly 31,000 per variant. If you can’t get there, test a bigger, bolder change, track a higher-frequency step such as confirmed add-to-basket alongside purchases, and lean on usability testing. Just never treat an underpowered flat result as proof that speed doesn’t matter.

“Faster is always better, so I’ll chase every millisecond.”

Returns aren’t a straight line. The retail-delay research found customers’ sensitivity differed by journey stage and did not rise in a simple linear way, and effects vary by device, category and starting point. Fix your worst bottleneck first, then re-measure.

Your playbook

Instrument the journey by device: search, filters, product page, add-to-basket, checkout start, purchase.
Use real-user data: track Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift at the 75th percentile.
Fix checkout first: basket updates, form validation, shipping selection and payment steps.
Fix mobile discovery next: search, filters, variant selection and product images. Prioritise the main product image and defer what isn’t essential.
Keep controls steady: don’t let size selectors and buy buttons shift while people reach for them.
Test it properly: randomise, pick one primary outcome in advance, add guardrails such as revenue per visitor and error rates, and cover full weekly cycles.
Report honestly: show absolute and relative changes, and give your uncertainty.

The real philosophy

Conversion optimisation isn’t about tricking people into buying. It’s about removing every reason they have to leave.

Speed is one of those reasons. Confusing layouts are another. So are clunky checkouts, wobbly buttons and filters that think about their answer for too long. A customer doesn’t care which team owns the problem. They only know whether the experience respected their time.

So here’s your homework for this week:

Open your analytics and pick one journey: mobile checkout is a brilliant place to start.
Find the slowest step a real customer hits.
Fix that one thing.
Test it against a control.
Let your own numbers, not someone else’s, tell you what it was worth.

Stop collecting borrowed statistics. Start collecting your own proof.

The sources synthesise findings from a diverse range of evidence, spanning rigorous academic theory to real-world commercial applications. The key areas of synthesis include:

Chernev, A., Böckenholt, U., & Goodman, J. (2015). Choice overload: A conceptual review and meta-analysis. Journal of Consumer Psychology, 25(2), 333–358.
Ding, Y., & Kyung, E. J. (2026). Optimising Animation Speed: Convex Effects on Perceived Waiting Time and Digital Customer Experience. Journal of Consumer Research, 53(1), 136–161.
Galletta, D. F., Henry, R., McCoy, S., & Polak, P. (2004). When the Wait Isn’t Worth It: An Empirical Study of Web Site Delay Effects. Journal of the Association for Information Systems, 5(5), 222–242.
Gallino, S., Karacaoglu, N., & Moreno, A. (2023). Need for Speed: The Impact of In-Process Delays on Customer Behaviour in Online Retail. Operations Research, 71(3), 876–894.
Luceri, B., Latusi, S., & Martinelli, E. (2022). Drivers of mobile shopping adoption, satisfaction, and continuation: A meta-analysis. Journal of Retailing, 98(4), 512–528.
Nichifor, E., Lixăndroiu, R., Chițu, I., et al. (2021). How Does Mobile Page Speed Shape In-Between Touchpoints in the Customer Journey? A Research Regarding the Most Trusted Retailers in Romania. Journal of Theoretical and Applied Electronic Commerce Research, 16(5), 1421–1437.
Schmutz, P., Roth, S. P., Seckler, M., & Opwis, K. (2010). Designing product listing pages: Effects on sales and users’ cognitive workload. International Journal of Human-Computer Studies, 68(7), 423–431.
Stadnik, W., & Nowak, Z. (2017). The Impact of Web Pages’ Load Time on the Conversion Rate of an E-Commerce Platform. In International Conference on Information Systems Architecture and Technology (pp. 318–328). Springer. doi.org/10.1007/978-3-319-67220-5_31
Wang, Q., Ma, D., Chen, H., Ye, X., & Xu, Q. (2020). Effects of background complexity on consumer visual processing: An eye-tracking study. Journal of Business Research, 111, 270–280.
Wang, Q., Yang, S., Liu, M., Cao, Z., & Ma, Q. (2014). An eye-tracking study of website complexity from a cognitive load perspective. Decision Support Systems, 62, 1–10.
Baymard Institute. (2025–2026). E-Commerce UX Benchmark & Checkout Usability Research. Baymard Institute.
Deloitte Digital & Fifty-Five. (2020). Milliseconds Make Millions: The Impact of Mobile Speed on E-Commerce Conversions. Commissioned by Google.
Google / Farfetch. (2022). Luxury Retailer Farfetch Sees Higher Conversion Rates for Better Core Web Vitals. Google Performance Case Studies.
Google / Vodafone. (2021). Vodafone Performance Case Study: How Core Web Vitals Optimisation Drove an 8% Lift in Sales. Google Performance Case Studies.
HTTP Archive. (2025). The Web Almanack: Web Performance & Core Web Vitals Report. HTTP Archive.
Littledata. (2026). E-Commerce Benchmark Report: Add-to-Cart and Purchase Conversion Metrics across Shopify Merchants. Littledata.
Nielsen Norman Group. (2020). The Need for Speed, 23 Years Later. NN/g Usability Research.
Unbounce. (2024). Conversion Benchmark Report: Analysing Conversion Performance across Industries and Devices. Unbounce.