Median page load times have risen 9% since 2011, a report says.
The median page load time for e-commerce sites viewed on Internet Explorer 9 is 6.50 seconds, 9% slower than last year’s median of 5.94 seconds, according to data from web optimization company Strangeloop Networks Inc. For the second view of a page, which is often faster since most web browsers, including Internet Explorer 9, temporarily save some files, median load time was 2.16 seconds, 15% slower than 1.83 seconds in 2011. For retailers, this means business: studies show that page load times affect site views, customer satisfaction and conversions.
Strangeloop’s report, which has been conducted annually since 2010, measures web performance among the top 2,000 e-commerce web sites as ranked by web analytics vendor Alexa Internet Inc., a subsidiary of Amazon.com Inc., on a simulator that mimics the browsing experience of users around the world. The browsers in the latest report include Internet Explorer versions 7-10, Google Inc.’s Chrome 20 and Firefox 13; Strangeloop does not have comparative data for last year's load times on the browsers besides Internet Explorer 9, it says.
Behind the slower load times are increasingly content-rich retail web sites. Pages load slower when they have more content, including pictures, videos and ads, because each of those items requires a separate call to a server which takes 20 to 50 milliseconds. The report found that page sizes, on average, increased 5% this year, with the median page requiring 77 server round trips, up from 73 last year.
Those effects are compounded by server calls made by third-parties that add features such as a Facebook Like button, or post their own content, such as a product recommendation from a vendor on a retailer’s site.
This means retailers don’t always have full control of their site performance, says Joshua Bixby, Strangeloop CEO. He advises they make service-level agreements, or SVAs, with vendors they employ to guarantee a minimum load time for any services their pages use. “Don’t just throw up your hands and say, ‘My recommendation engine is slow,’” he says. “Really push those vendors.”
Retailers should also regularly monitor how each page element loads and make sure that there isn’t one culprit—for example, a Twitter widget—that brings down the whole page when it crashes, he says.
The report also found that the median page load time for sites viewed on Internet Explorer 10 is 6.392 seconds, versus 6.906 seconds for Chrome 20 and 6.395 seconds for Firefox 13. However, that difference doesn’t signify much; Bixby stresses that today most web browsers are equal in terms of customer experience. “If you use a modern web browser, which tool you select is not a big deal,” he says. Strangeloop only looked at the performance of various browsers because the media is interested, he says.
Regardless of the browser, retailers have to be concerned about slower load times, which can produce profound results, he says. In February, Wal-Mart Stores Inc., No. 4 in the Internet Retailer Top 500 Guide, reported a sharp decline in conversion rates, without revealing exact numbers, when its page load times increased from one to four seconds. Every one second improvement, however, increased conversions by up to 2%, with every 100 milliseconds of improvement increasing incremental revenue by up to 1%, the retailer says. Other large e-commerce sites have reported similar findings.
“This data tell us that when it comes to performance, many retailers are still leaving money on the table,” says Bixby. “Site owners should be conducting routine tests across a variety of browsers. They should be measuring performance using a variety of metrics, and they need to ensure third-party scripts have been optimized. Implementing these basic performance tenets will drastically improve revenue.”
Within five years, Bixby expects that all web pages will load in 100 milliseconds, which he says is plenty fast for consumers. Although this report shows the opposite, he says a 9% decrease in page load times is just “a blip on the radar and I have confidence in the ingenuity of man- and womankind.”
Beginning with this report, Strangeloop will begin to benchmark web performance quarterly instead of annually, he says.