Target.com Crash Not A Site Failure But A Failure To Forecast
Much has been made over the last two days of the Target.com crash and availability issues following its Missoni promotion that launched on Tuesday. Coming on the heels of the site’s relaunch on IBM Websphere Commerce after many years on the Amazon enterprise solution led many to speculate that there was a serious mistake made in the building of the new site. But this is not a failure of the new site, this is a failure to forecast.
When Target's marketers planned the media blitz and direct marketing around this Missoni promotion, launching the collection with great fanfare — and at the same time online and in-store— they likely planned on a web traffic response and resulting sales of 2-3x normal. That would be a terrific outcome, no doubt justifying the expense and effort of the multichannel promotion. I am sure no one forecasted a 10x normal response, and peak traffic 3x Black Friday traffic peaks. And beyond just the click-through, the consumer behavior on the site after click-through was unusual. This was not a product they were promoting, but a collection, and consumers were placing many, many items in their cart, well beyond the typical 1-2 items per cart a B2C site like Target's would expect.
Managing this kind of load on the databases was not something the system was configured to handle. And while addressing these issues is not all that hard, doing it under fire is a challenge. As the site came back up and traffic remained high, Akamai’s shopper prioritization solution helped manage the traffic and allow in only a portion of the customers hitting the site to ensure that search indexes and carts could be handled, the free shipping offer handled, and that a positive customer experience could be ensured for those in the site. With only a small portion of the traffic allowed in at any given time, many customers thought the site was down, when in fact it was being managed well in an unanticipated situation. In fact this scenario is very similar to the master-of-the-scaling-dark-arts Amazon.com t faced with its Lady Gaga promotion earlier in the year.
A few things can be learned from this:
· Sites management is a function of planning and testing to that plan. Get the plan wrong by a high degree and you are either missing your numbers, or possibly burying your site. If you are a marketer planning on making a huge splash with direct marketing and media be proactive to help your site management teams (internal and partners) anticipate that and prepare and test the infrastructure and software for the load. In this case it may have been hard to anticipate the response, and in that case it may be unavoidable.
· Test into marketing not only to tune the message, but understand the response. A best practice when launching a very attractive or high-profile direct marketing initiative is to release a small portion of the email file and ad buy and then gauge the response. You would also probably want to test multiple versions while you’re at it to tune the offer. If you see a wildly successful response, you continue to throttle the release of the promotion to ensure your sites are able to handle the load.
· This is not a function of cloud versus non-cloud. Examining the contributing factors here tells us this is not a matter of a cloud-based solution or not. This was a matter of not forecasting a huge marketing win and testing to a scenario like this — huge load with unusual basket sizes with a free shipping promotion on top of it. If this were a cloud-based solution it is very likely the same issue would have presented itself, and perhaps may have affected other sites in a multitenant environment.
Hard to say if a situation like this one will present itself again in the near future — for you, for Target, or for Amazon — but we would be remiss not to look at the contributing factors and put process and plans in place to address them.