Why ClickFrenzy Failed – A Consultants Perspective

The not so funny thing about ClickFrenzy was every one said it was going to fail. The funny thing no one has actually spelled out is how they could have avoided it.

Everyone is saying they failed to plan which is “planning to fail”, but what is it that they failed to plan for, and how could they have avoided it?

Every year around Thanksgiving in the Unites States, they have the Black Friday sales. They are famous for the massive numbers of people that turn out and rush the doors the moment they open to grab themselves a bargain.

About the same time, in the Unites States, they also have a follow up online sale called Cyber Monday that capitalises on shoppers who either didn’t get to the stores or couldn’t find the products or discounts they wanted at the earlier Black Friday sales. Given the Unites States is a nation of over 300 million people, the Cyber Monday online only sales is exposed to an extremely large audience.

The challenge however is not so much how to design a website that can host the entire population, but rather estimate what percentage of that population is going to jump onto the site all at once. This is the obvious failure ClickFrenzy made – failing to provide sufficient technology infrastructure to cope with the volume of concurrent connections.

Now, from what I understand, ClickFrenzy was built using cloud based servers  - and I’m betting they were using some sort of virtualisation suite that would allow additional servers to be dynamically started up as demand would increase – and if they weren’t, I’d add that to the list of failures to learn from.

Additionally, let’s assume that at the time they went live, the organisers were pretty locked-in on who the retailers would be and the specials that would be made available so they should have been able to load the entire product database into memory or cache in order to load the pages faster.

At a technical level, the capacity & performance of the Click Frenzy AND associated retailers websites is another factor that contribute to the overall success or failure. Each Retailer might typically have their website designed for regular e-commerce volumes of traffic that might allow for, lets say for example, an average 100 – 1000 concurrent users, however for events like this, the sites need to be able to handle the maximum peak or 99th percentile volume of users, however this is not always practical or financially viable.

An example chart provided here indicates how a website may experience on average about 5000 page views, but on its busiest day, it experiences over 50000 page views. What do you design for on a day to day basis? the average? the maximum you have experienced in the past? or a project maximum in the future that may exceed any historical maximum? The example provided here even displays a trend where for the last week, the average has increased to 7500 indicating an ongoing increase in traffic to the site.

So the solution lies with knowing customer behaviour. How long would you be willing to stand in line to get a discount. In an online sense, how late would you be willing to stay up in order to log on for the same discount?

This is one of the key points that tie the above suggestions together.

If you make the site live at 7pm when the entire population is awake and most likely at home within reach of their PC, you are open to an extremely huge market – easily in the millions of potential customers – all likely to jump onto the site at exactly 7pm. In the US, 1 million concurrent users might be catered for and certainly financially worthwhile, however in Australia, I’m sure there are some costings that suggest you do not build for that many users.

The way around this, or at least my answer, is to go live at midnight.

That easily reduces the online population considerably and provides a much more affordable platform from which to launch. Then with the aid of the virtualisation solutions mentioned earlier, as the population awakes the following morning, additional servers can easily and quickly be turned on(or at least quickly enough) to cope with the growing volume of online users throughout the  following day.

This effectively takes the peak and spreads it out over the course of 24 hours and avoids the peak becoming the opening minutes.

It is interesting to be standing on the outside sometimes looking in when you’re not the person responsible for the site going down. Rather than standing back and laughing “Ha Ha!” in Simpsons’ Nelson Muntz mocking style, as a performance consultant, I wince. This is the exact sort of scenario I consult on to prevent from happening. Every failure, be it your own or one like that experienced by ClickFrenzy, should be taken as an opportunity to grow, learn and evolve. The organisers of ClickFrenzy have taken on a very brave challenge and I am sure they will be back next year much wiser for their experience, and much better prepared.

Oh, and for the record, I made several purchases as part of the ClickFrenzy sale and will likely do so next time ClickFrenzy comes around.