A tenth of a second is shorter than a blink. Online, it can be long enough to change what people do.
In 2020, Deloitte studied mobile traffic across retail, travel, luxury and lead-generation websites in Europe and the US. The researchers looked specifically at what happened when sites became just 0.1 seconds faster. Among retailers, that improvement was associated with an 8.4% increase in conversions and a 9.2% rise in average order value. Travel sites saw conversions increase by 10.1%.
The report had an appropriate title: Milliseconds Make Millions.
It sounds exaggerated until you consider the scale of modern websites. For a business generating £100 million online, even a one per cent difference in sales represents £1 million. Small changes in customer behaviour do not stay small when they are repeated across millions of visits.
The surprisingly expensive tenth of a second
Deloitte’s figures should not be treated as a universal formula. Making any website 100 milliseconds faster will not automatically increase its revenue by 8.4%. The study observed a relationship across multiple businesses rather than establishing a fixed rule.
But individual companies have found remarkably similar patterns.
When eBay made its search pages faster, it reported that every 100-millisecond improvement in loading time was associated with a 0.5% increase in the number of items added to baskets.
Luxury fashion retailer Farfetch approached the question by comparing its own performance data with conversion rates. Once its Largest Contentful Paint, a measure of how quickly the main content becomes visible, moved beyond 2.5 seconds, conversion began to decline. Its analysis found an average 1.3% reduction in conversion for every additional 100 milliseconds of LCP.
Vodafone went a step further with an A/B test. Visitors were split between otherwise similar landing pages, one of which had been optimised for better web performance. A 31% improvement in LCP was accompanied by an 8% increase in sales.
Milliseconds, in other words, are not merely an engineering concern.
Why do tiny delays matter?
People are unlikely to consciously identify a delay of 100 milliseconds and decide to leave because of it. The effect is more subtle.
A website is rarely a single interaction. A shopper opens a category, applies a filter, views a product, changes a size, adds it to a basket and moves through checkout. Each action creates another opportunity for delay.
The effect becomes more noticeable on highly interactive services, where a single session can involve dozens of requests. An online casino, for example, may combine a game lobby, slots from multiple game providers, live-dealer video streams, account verification and payment flows within the same session. Each relies on pages, APIs and third-party services responding quickly. A small delay in one interaction may barely register; repeated throughout the journey, those delays begin to shape how responsive the entire platform feels.
And waiting is only one part of web performance. A page can appear quickly but respond slowly when a button is pressed. Images can load late and push other content around. Menus can hesitate. A payment button can leave the user wondering whether the first click registered.
Speed, from the user's perspective, is really a collection of small moments.
Faster is not the same as stripped down
There is an obvious temptation to solve the problem by removing anything that takes time to load. But the fastest possible website would not necessarily be the most useful one.
Images sell products. Search tools help users navigate large catalogues. Video can explain something that text cannot. Security checks and payment systems have jobs to do.
The challenge is therefore not to make a site empty. It is to decide what deserves the user’s time and what does not.
That can mean compressing images, loading non-essential content later, reducing unnecessary code or making sure the most important part of a page appears first. Much of the work is invisible when it succeeds.
Which is fitting, because 100 milliseconds is almost invisible too.
The difference is that, at sufficient scale, its value may be anything but.
