As we do every quarter at Radware, we’re releasing a new “state of the union” report – an in-depth snapshot of web performance of the world’s top ecommerce sites.
In our research, we’ve found some recurring themes that prove problematic for user retention for these e-retail heavyweights:
- Pages keep getting bigger in their total size
- The number of resource requests is increasing dramatically
- Performance takes a hit due to page complexity and large, unoptimized images.
Why do these trends threaten user retention?
Because multiple studies have shown that users will abandon (or “bounce” from) a slow-loading website after just three seconds. It’s critical that sites shoot for three seconds or less for their Time to Interact (TTI), which is a measurement of the amount of time it takes for a web page to load its main image(s) and central interactive content.
According to estimates from Statista, ecommerce sales for 2015 will be in the neighborhood of 1.7 trillion U.S. dollars, and that total is expected to grow to 1.92 trillion in 2016. Those are big numbers, certainly. But how much bigger would those numbers be if all of the ecommerce sites had a TTI of three seconds or less?
That’s a rhetorical question, but we can examine our key findings from our latest report and find out what’s impacting the user experience with websites in their load times. If the issue is big enough for the general media, as well as general consumer tech news sites to discuss (read about that here), then it’s certainly more relevant than ever.
Our Purpose and Process for the State of the Union Report
As we’ve done since 2010, we measured and analyzed the performance of the top online retailers, as ranked by Alexa.com, looking at key web page metrics from load time to TTI, page size/composition and the adoption of performance best practices. We used widely-available tools, including WebPagetest.org and HTTPArchive.org, to obtain real-world snapshots of these pages’ metrics (as well as for web pages in general) and to evaluate their change from quarter-to-quarter and year-to-year. Our findings are detailed in our quarterly “state of the union” reports, some of which you can find here.
Finding 1: The median page’s Time to Interact is 5.5 seconds, and fully loads in just over 15 seconds.
WebPagetest.org features some great tools, one of which is the useful filmstrip view, enabling a look at the page’s loading in sequential images captured in as little as .1 second increments.
As you can see in the above screen capture, for that particular website, nothing displayed before the 5.5-second mark. This is representative of the median page’s Time to Interact, which has actually slid from the 5.2 seconds we observed in our testing for the recent Spring 2015 State of the Union. Indeed, many – but not all, importantly – were more sluggish in their load times from just earlier this year.
Again, the ideal target is a TTI of three seconds, and the danger of users bouncing goes up the longer the page takes to render its central content after those three seconds have elapsed.
Finding 2: The Median Page is 1905 KB in Size and Contains 169.4 Resources
While page size is part of the problem, page complexity is arguably an even greater performance challenge. In one year, the median number of resources for a top 100 ecommerce page has grown by 69.4% — from 100 resources in Summer 2014 to a whopping current median of 169.4. Each of these page resources represents an individual server call.
|Summer 2014||Summer 2015||Change %|
|1677 KB||1905 KB||+ 13.6%|
|100 resources||169.4 resources||+ 69.4%|
All of these requests rob pages of precious seconds of load time since 80% of the time required to display a typical web page is spent loading the resources needed to render the page and performing client-side processing. Only 20% of the time required to display a typical web page is actually consumed by loading the page’s HTML.
With requests at a median of almost 170, multiplied by 65-145 milliseconds (or more) for desktop browsers per each round trip, and you’ve got a significant tally piling up.
Finding 3: Most Sites Fail to Take Advantage of Core Image Optimization Techniques
Images typically compose between 50% to 60% of a page’s total weight. Yet 48% of the top 100 retail sites fail to compress images, which would have lightened their sites’ overall payloads. Fewer bytes mean reduced bandwidth and faster pages.
Only a scant 3% received an ‘A’ grade for image compression.
Consider the impact these featured images had in this extreme example from our findings:
Those were real-world numbers, and didn’t change much when re-tested a few weeks later, so it wasn’t a fluke. These images were killing the TTI and overall load time, and could have benefitted hugely from an image optimization automation solution.
There’s Always Hope – if You Optimize
Every site – from ones that are already responsive all the way down to the ones that aren’t – can continue to fine-tune their performance by employing the best tools and for front-end optimization. The performance gains are there, as we illustrate by examining the gains IKEA and Sierra Trading Post made in the time between our testing for the Spring 2015 report and our current one, jumping in their rankings from 93 and 77, respectively, to the 19 and 4 positions as ranked by their TTI’s.
How did they do that?
As we detail in our new report, they got a handle on their feature images, which had them at 10.5 seconds (IKEA) and 7.6 seconds (Sierra Trading Post) for their TTI’s the last time around, only to make a comeback this time with TTI’s of 3.6 seconds and 2.3 seconds, respectively.
Those are serious performance gains, and a great example of how to move the needle for your site’s performance.
Kent is an entrepreneur, software architect, and technology innovator. Before taking his former role of VP Acceleration at Radware, Kent was CTO at Strangeloop Networks, where he was instrumental in authoring all of Strangeloop’s issued and pending patents. Prior to helping create Strangeloop, he served as CTO at IronPoint Technology. Kent also founded Eclipse Software, a Microsoft Certified Solution Provider, which he sold to Discovery Software in 2001. In more than 25 years of professional development experience, Kent has served as architect and lead developer for successful production solutions with The Active Network, ADP, Lucent, Microsoft, and NCS. ”Port View”, an application Kent architected for the Port of Vancouver, was honoured as ”Best Administrative System” at the 1996 Windows World Open Competition.