23 Jul 2013
Today at Radware, we’ve just released our State of the Union: Ecommerce Page Speed & Web Performance [Summer 2013]. If you’re not familiar with this report, it’s a quarterly benchmark of the page composition and performance of leading ecommerce sites. The purpose of these reports is to learn how pages are changing over time and what impact, if any, these changes have on per-page performance. Each report has generated eye-opening findings, and this one is no exception.
Why should these findings matter to you?
When I talk to site owners and developers, I routinely hear the same pair of assumptions:
Assumption #1: Pages are getting faster.
As our browsers, devices, and networks continue to evolve, it stands to reason that load times must be improving, right?
Assumption #2: Even if load times aren’t improving, we shouldn’t worry about it.
These days, pages are smarter than ever about serving key content fast. From an end-user perspective, that’s what really matters.
If you hold one or both of these assumptions — or if your job function requires that you collaborate with someone who holds these assumptions — then this report should matter to you.
First, our test methodology…
As with previous reports, we used WebPagetest (a tool that simulates the real user experience across a variety of browsers and connection types) to measure the load times of the home pages of leading ecommerce sites via a DSL connection. Each URL was tested nine times (the maximum number that WebPagetest’s batch-testing function will allow), and the median result for each URL was used in our analysis.
We have, however, made a couple of significant changes from the methodology we used in previous reports:
- Focus on Chrome instead of Internet Explorer. In the past, we focused our analysis on the results for IE; however, Chrome is now the dominant browser in the United States, and the goal of these reports is to report on the performance experience of the majority of internet users.
- Focus on the top 500 Alexa-ranked sites, not the top 2,000. In the past, we looked at the top 2,000 sites. We’ve tightened our focus on the top 500 sites as these sites are more likely to have a performance strategy and to follow performance best practices. They are also more likely to face the performance challenges inherent in complex, dynamic sites. We wanted to see how these ostensible leaders address these challenges.
In analyzing our newest test results alongside our historical results, we extracted the Chrome tests for the top 500 sites in our previous benchmark tests, going as far back as Spring 2012.
Our tests generated a handful of interesting findings, but today I want to focus on these three:
Finding #1: Median load time for an Alexa 500 retail website is 7.72 seconds.
According to our data, leading retail websites are getting slower. The median page took 7.72 seconds to fully load for a first-time visitor. This represents a 13.7% slowdown since Spring 2012, when the median page took 6.79 seconds to load.
In recent months, load time has gathered detractors about its meaningfulness as a user experience metric. This is what led us to measure and pinpoint a new metric — time to interact — as we’ll discuss next…
Finding #2: Median time to interact for an Alexa 100 retail site is 4.9 seconds.
When we’ve released these reports in the past, we’ve met with some healthy skepticism about the relevance of load time as a performance metric. This skepticism has ranged from polite:
Are you measuring entire page load? Page load times could increase while tricks with rendering time may actually reduce the psychological experience of load time. I guess what I’m saying is could entire page load time increase but user perceived load time decrease? Could entire page load time be less relevant these days?
I call fear-mongering. A lot of tactics these days do actually increase load time BUT decrease start rendering time. Lazy and deferred loading make it so that the page as a whole takes longer, but really, what matters is the time until the user gets the first glimpse that the page is loading.
Regardless of how they’re phrased, these are valid points, which we set out to address in our latest round of testing.
In addition to measuring full page load, we also analyzed the filmstrip view (a cool WebPagetest feature that lets you view page load in a timestamped series of screen captures) of the top 100 sites to pinpoint each page’s time to interact (TTI), AKA the moment that the page’s featured content (with most ecommerce sites, this is some type of banner or carousel) loads and becomes interactive (meaning the featured content is clickable, usually via a call-to-action button).
We found that the median home page took 4.9 seconds to become interactive. If you’re a regular reader of this blog, you don’t need to be reminded that this falls well short of ideal. Consider the following findings from separate studies:
- 57% of visitors will abandon a page that takes 3 seconds or more to load. (source)
- A site that loads in 3 seconds experiences 22% fewer page views, a 50% higher bounce rate, and a 22% fewer conversions than a site that loads in 1 second, while a site that loads in 5 seconds experiences 35% fewer page views, a 105% higher bounce rate, and 38% fewer conversions. (source)
Given these findings, here’s a filmstrip view of the median time to interact (captions were added later by me):
This filmstrip view is not an anomaly: it’s typical of a significant portion of the top 100 sites. We repeatedly noted that primary content — and in many cases, as here, all content — was delayed well past the 3-second wait-time threshold.
I have to admit that this finding came as a surprise to me. Given that these are the top retail sites in the US, and therefore ostensibly more likely to have a strong focus on performance, I had predicted that time to interact would hover around or just above the 3-second mark — not great, but approaching acceptable.
Finding #3: Almost 1 out of 10 sites had a time to interact of 8+ seconds
This, to me, is even more compelling than the median TTI findings. We found that only 8% of the top 100 sites had a sub-2-second time to interact, while 9% had a TTI of 8 or more seconds.
Again, it’s useful to consider these findings in light of separate usability research from Jakob Nielsen, which found that a user who has to endure an 8-second download delay spends only 1% of their total viewing time looking at the featured space on a page. In contrast, a user who receives instantaneous page rendering spends 20% of their viewing time within the feature area of a page.
Takeaways: Do you know your site’s time to interact?
I encourage you to download the report to learn what our other findings were, find out which sites delivered the best user experience, and get tips on how to address common performance issues.
I also encourage you to use WebPagetest or any other tool that measures the end-user experience and test not just your site’s load time, but also its time to interact.
See a high-res version of the infographics that accompany this report.