Research presented today at the NRF Big Show in New York highlights 2016 holiday findings from popular retailers.
More complex sites present new performance challenges
When customers began complaining about difficulty in completing purchases on Panasonic’s web site last year, Jeremy Dalnes, vice president of e-business for the consumer electronics manufacturer, took a close look at the site’s performance data.
He found that the second page in the checkout flow was taking far longer to load than other pages. The several second delay was causing customers to abandon transactions.
Further investigation revealed that the delay stemmed from an image that had been left behind when Panasonic distributed its web site content onto a global delivery network of an outside vendor in a move to speed up site performance by locating content nearer to site visitors. The site was looking for the image-and eventually finding it on Panasonic’s server-but not before several seconds had elapsed.
Knowing the tech details
Fixing the problem dropped the time it took to load that checkout page by 80%, solving the problem, Dalnes says. The lesson Dalnes took away was that e-commerce managers need to understand enough of the technical details of web site performance to discuss problems intelligently with information technology staffers.
Many e-commerce pros are having those discussions because performance issues are not uncommon-82% of web site operators said they experienced performance issues on their public-facing sites in the previous year in a survey last fall by JupiterResearch.
What’s more, maintaining web site performance is becoming more complex as more content is hosted away from the retailer’s own server, says John Lovett, an analyst who specializes in web site performance at JupiterResearch.
One source of the complexity is the growing use by online retailers of vendors that host software on their own servers, a model known as software-as-a-service. These include some leading vendors that serve up customer reviews for retail sites. When a site visitor clicks on the customer review on the retailer’s site the information must be pulled from the vendor’s site and assembled on the fly for the user to view.
Deconstruction = complexity
Images might be hosted by a separate content management vendor on its servers. And a recommendation engine hosted by another vendor may present other products that a visitor might like based on her behavior. All of that content must be put together and presented to the visitor as if it came from the retailer.
“This deconstruction increases the complexity of performance management because multiple origin servers are involved, content calls are required from numerous sources, and all these elements must be dynamically assembled in fractions of a second,” Lovett wrote in a recent report.
Add to that retail sites’ growing use of new technologies, such as video and zoom, and the thousands of combinations of browsers, software and hardware that consumers use, and the problem of ensuring that a page renders as intended becomes extremely complex.
It’s natural that online retailers would want to incorporate into their sites all the latest features that consumers are seeing on other web sites, says Imad Mouline, chief technology officer for Gomez Inc., a performance-management company. “End users’ expectations are changing,” Mouline says. “And more retailers are wanting to make their sites as rich as possible.”
It’s your site
But before implementing complex elements like graphics and video, Mouline encourages e-retailers to test those elements-and not at the office where technicians are likely to have fast computers and Internet connections. Test them in the type of environments where consumers will use them, he says, and, to ensure that technicians will be seeing the site the way visitors would, test them across all the different combinations of operating systems and web browsers like Internet Explorer, Firefox and Safari. A page that loads quickly with a particular operating system and browser may cause a session to time out because of slow load time with a different combination of technologies.
Retailers also must closely monitor any site features produced by outside vendors to make sure their work doesn’t cause conflicts with content produced in house. Routine tests are important, because retailers may not always know when an outside provider has updated its content. “It’s still your site and your brand,” Mouline says. “Users aren’t going to blame a third-party vendor.”
The importance of web pages loading quickly increased last month when Google began penalizing search ads from companies whose landing pages load slowly. Landing pages that take more than 3 seconds above the average for a given region to load will receive a lower rating, Google says. That could mean the ad would appear lower on a search results page, or not at all.
Third-party content on a landing page can slow down load time, as a delay from any provider prevents the page from being fully loaded in Google’s view, Mouline says. Examples of third-party content include tracking tags from analytics vendors and ads from ad networks. “You have this third-party content coming from a variety of places being assembled in real time at the end user’s browser,” he says. “All of that impacts page load time.”
Other delays can be caused by pages that refresh the graphics periodically-such as rotating ads every 15 seconds. Search engines often include the time it takes to refresh the ads as part of the load time.
Ads also will be penalized for redirecting the user to a page that appears briefly before sending the user to the ultimate landing page. In some cases those intermediate pages, which have their own URLs, are used to register the click on the ad, Mouline says. “Minimizing redirects and refreshes is good,” he says.
Gomez is offering a free site that can test how quickly landing pages load when an ad is clicked on by users in New York, Los Angeles, Chicago, London and Beijing. Advertisers who sign up for Gomez’s paid web-monitoring service can test their pages from 35,000 locations and customize the testing, for instance testing only during the day for advertisers that do not serve up ads at night.
Design for performance