Learning From A London Site Blowup

How does an E-tailer handle a tidal wave of customers created by a big marketing stunt? In the case of U.K. group-buying site Groupola, the answer appears to be: It doesn't.

Last Friday (July 2), Groupola's Web site ground to a halt after the company offered 200 SIM-less iPhone 4s to its members for 99 pounds ($150 U.S., 80 percent less than Apple's price) and got 5 million login attempts as soon as the phones became available.

That didn't have to happen--and it shouldn't have. The whole point of a marketing stunt is to shake things up. It's supposed to bring in a lot of customers all at once. But those customers will just go away irritated if they show up and an E-tailer can't handle them. In the brick-and-mortar world, the answer is obvious: staff up with temporary salespeople. In E-Commerce, the answer turns out to be pretty much the same: rent some temporary help in the form of cloud computing.

Apparently that didn't occur to Groupola. The company promoted the loss-leader iPhones as a special deal to drum up new members for its newsletter, which E-mails a special discount offer every day on some product or service. If enough members buy into the offer, they get the deal; otherwise, the offer is withdrawn. That was never a danger with the iPhone 4, which is in short supply in the U.K., even for mobile carriers.

Minutes after the offer went live, Groupola's Web site all but collapsed from the load. Some would-be customers couldn't get to the iPhone page. Others managed to key in their payment card information before they got kicked off the system. Groupola later insisted in a press release that "Between 9 and 9:30 a.m. this morning more than 5 million people have attempted to log on to the site, far more than was anticipated." The fiasco sparked accusations--in blogs, news stories and on Groupola's Facebook page--that the offer was an E-mail harvesting scam and that the too-good-to-be-true offer always had been just that.

That's probably not the reaction Groupola wanted. It's not the reaction any E-tailer's Marketing department is looking for, either. Marketing departments love big-bang events, the online equivalent of a brick-and-mortar sale that has people lined up around the block or down the mall. But E-tailers' IT departments hate these events, and with good reason. That huge bulge in demand is an impossible capacity-management problem. If you've got enough hardware to handle it, you have way too much hardware on a day-to-day basis. If you don't have enough hardware, you'll lose the battle of the bulge.

Does this sound like a problem tailor-made for cloud computing? Let's see: an occasional large surge in demand? Check. A well-defined service that demand is for? Check. A very predictable timeframe for the surge? Check.

So if Marketing drops a sudden, potentially catastrophic Big Event into IT's lap, cloud computing could be a relatively clean way of farming out that piece of it. And only that piece of it; no retailer wants to re-create its whole Web infrastructure up in the cloud. The cost would be too high, unless there are plans to move there permanently.

But getting some temporary help for the big sale? That makes at least as much sense online as it does in the real world. Running a brick-and-mortar Big Sale without extra staff would be a recipe for a riot. Doing the same thing online is a do-it-yourself denial-of-service attack--except this time the "attackers" are customers who want to buy.

In Groupola's case, the company might have used a link in its newsletter that went straight to a cloud-based mini-Web site just for the iPhone offer. That approach still would have required working out how to handle logins for the lucky first 200 visitors who could snap up the available phones. After that, the job of the cloud-based site would just have been to inform the crowds that the phones had sold out.

For more conventional retailers facing a big online marketing event, using the cloud might require something more complicated--say, redirecting all the traffic for the online store to the cloud, where most customers would presumably choose to click through to the high-traffic promotion, which would be built up in the cloud. The rest (a relatively small number of customers) would click for the regular online store.

That way, the non-cloud servers won't be crushed by the crowd and the cloud provider will take the brunt of the pounding. After all, that's what it's getting paid for.