The sites evaluated were Amazon.com, Barnes & Noble, Best Buy, Buy.com, Circuit City, Costco, Dell, Crate and Barrel, eBay, Eddie Bauer, Gap, JCPenney, L.L. Bean, Macy’s, Nordstrom, Office Depot, Overstock.com, Staples, Target, Wal-Mart, Williams-Sonoma and Yahoo Shopping.
The results didn’t show any one e-commerce site doing well in all areas, but the company did release the companies that fared the best in selected categories.
For example, Office Depot and Staples delivered the best response times, with Office Depot on top for average T1 response time (with Staples coming in second) and Staples on top for average analog dial-up response time (with Office Depot coming in second and Costco coming in third).
For site reliability, Eddie Bauer came out on top, with Barnes & Noble coming in second and Staples third. For overall service level ranking, the top five were Williams-Sonoma, Wal-Mart, Eddie Bauer, Staples, Circuit City and Best Buy.
What struck Ben Rushlo—one of the report’s authors who serves as Keynote’s senior competitive researcher—most intriguing was the huge difference between some of the larger sites and how poorly some of the weakest performers fared.
During the one month of testing (roughly 480 hours of Web performance evaluated), the unidentified site with the weakest reliability was down for about 10 hours. Compare those numbers with that category’s top performer—Eddie Bauer—“which had not a single error for the entire month that we measured them,” Rushlo said.
The evaluation considered any improper response (such as clicked link not delivering the correct page or delivering a "page not found" message) an error, Rushlo said.
Ten or more errors constituted a failure. If more than 30 percent of the clicks generate an error during one hour, that would be considered a site outage or the site being down.
The industry average for successful clicks is about 98.6, Rushlo said, adding that the top 11 of the e-commerce sites Keynote evaluated averaged 99.5 percent, with Eddie Bauer at 100 percent, Wal-Mart at 99.6 percent and Williams-Sonoma at 99.3 percent.
Rushlo said the report’s results strongly suggested that many of these sites were not actively watching their own performance or many of the glitches would have been repaired.
“Some of the worst sites had such glaring issues that it was clear that no one was captaining the ship,” he said. “There needs to be an internal team that shows the site inside and out and is constantly testing and monitoring.”
When he confronts managers of sites that fared poorly, “it’s very typical to have someone say, ‘We haven’t measured that,’” Rushlo said.
“When we work with a customer, quite frequently they’ll have a half-person dedicated” to tracking Web performance.
Rushlo said the final segment of an e-commerce site is often where they found the greatest problems.