In Theory, E-Commerce Sites Are Way Too Slow. But Do Customers Care?

Speed-tuning for retail Web sites may have finally hit a wall. A report released Wednesday (Jan. 25) says Nike, JCPenney, JCrew and Amazon had the fastest retail sites in 2011. But the survey also notes that the most popular and profitable sites are actually slower to load than the average site, because they contain so much content, and that content delivery networks don't actually speed up load times.

In theory, load times of 3 seconds or more should cost retailers half their customers. If that's true, E-tailers should be going out of business. Maybe it's time to dump those theories.

The author of that report is a Web site performance vendor called Strangeloop Networks, so it's hard to blame the company for pushing performance. But according to the study of 2,000 E-Commerce sites (which includes in its top 10, so this is a pretty loose definition of E-Commerce), only three retail Web sites have sub-3-second load times. Strangeloop says research dating back 40 years shows that two seconds is the optimal load time, and 57 percent of customers will abandon a site after 3 seconds.

Forty years of research? That's back to mainframe days and, yes, two seconds is a long time to wait for the next mainframe screen to appear on your terminal. But there's nothing magical about that two-second number. Consumers don't throw out their PCs because Microsoft Word takes longer than 3 seconds to load or a giant spreadsheet takes more than 3 seconds to finish calculating.

And they clearly haven't abandoned E-Commerce sites in droves, despite the fact that 1,997 of the 2,000 sites in this study take more than 3 seconds to load. In fact, typical load times for major retail sites run between 3 and 6 seconds, with the average load time for retail sites hovering around 10 seconds. The theory may dictate a two-second holy grail. Reality disagrees.

That's the biggest trouble with trying to drive E-Commerce site performance by the numbers: Faster isn't always better. What's better is whatever sells more merchandise.

This is probably why homepages are getting bigger.This is probably why homepages are getting bigger, even though smaller pages load faster. The average number of objects (images, buttons and, particularly, scripts) on a site rose 14 percent in 2011 to 98, from 86 objects in 2010, according to the Strangeloop report. Some of that is just clutter, but some comes from third-parties providing functions such as recommendation engines. Still more of it is for analytics and customer tracking.

If customers were actually abandoning your site in droves, you'd have good reason to go slicing away at the CRM, analytics and recommendations just to speed things up—after all, you can't sell anything to a customer who has walked away. But there's no evidence from retailers that customers actually jump ship as quickly as the theorists say they will—as long as customers believe there's something worth waiting for.

Maybe the most surprising item in the Strangeloop report is that although more retail sites are using content delivery networks, that isn't translating into quicker Web sites in front of users. According to the report, CDN use by retailers has grown by 17 percent. But sites using CDNs load on average in 10.058 seconds, while non-CDN sites have an average load time of 9.72 seconds.

Strangeloop figures the difference appears because the retailers using CDNs have bigger sites, or maybe they figure that with the improved speed of a CDN they can afford to have bigger sites. But that misses the point. A 10-second load time should be deadly, with or without a CDN. Clearly, it's not.

None of that means looking for ways to speed up Web sites is a bad idea. Making sure customers aren't waiting, say, half a minute for your site to load is certainly worth the trouble, especially if it's due to one of those third-party services. And mobile-commerce sites still vary wildly in how long they take to load on a smartphone, which suggests there's lots of benefit to be gained from some relatively easy tuning.

But fine-tuning to shave off a fraction of a second? Ridiculous. At this point, trotting out the same idealized numbers from a bygone mainframe era is pointless. The real wait that customers will accept today isn't 3 seconds—it's 10 seconds. Faster than that is nice. But until customers actually do start leaving after just a few seconds, E-Commerce sites have bigger problems to worry about.