On the first day his post drove ~60,000 visitors to the Branded3 site. That’s around 20-30x our usual daily traffic, just to one page.
Our Google Analytics graph for April/May looks like this:
In isolation it looks as if we usually have quite low traffic.
The annotation under the graph says:
Stephen Kenwright breaks Branded3 server: The antisocial network: Path texts my entire phonebook at 6am
(ProTip: You can add annotations by double-clicking any date – this is useful for notable changes to your website such as a large change in traffic or a website redesign)
While our awesome development team were scrambling to fix the servers that had broken with the Path post, it got me thinking.
Be prepared for your server to die
If any website has a sudden surge in traffic, it can kill your server. Imagine a pipe with a slow trickle of water and suddenly a flood comes through. The pipe will probably burst. This is known as the “Slashdot Effect” and usually happens with popular posts on small websites.
Douglas Radburn helped keep the Branded3 website alive when Stephen’s post suddenly got a lot of attention – this is his advice about what to do in this case:
Ensuring your platform is optimised is an important step. Always make sure you’re using the most up to date version of your blogging platform and plugins. Caching options are readily available and easy to configure through plugins, even without specialist knowledge. Set one up now when your traffic is low, and you’ll ride the storm. Cloud hosting allows near-instant scalability in high-traffic situations so that you can quickly react, and depending on your platform, this is done seamlessly.
Having one really popular blog post messes with your metrics
Having a huge surge of visitors that behave very differently to the norm will definitely affect your metrics.
Here are some metrics that changed dramatically from March (pre Path post), April (Path post live) and May (Path post still getting high attention)
|Conversion Rate||0.11%||0.05%||0.03%||Conversion rates plummet|
|Bounce Rate||80.78%||88.61%||86.85%||Higher Bounce Rate|
|Visit Duration||59s||37s||50s||Visit Duration decreased|
N.B.: Bounce rates for websites with blogs are nearly always higher than bounce rates for websites without blogs. Many people who want to read a blog post come to the website to read a blog post and leave. This is a bounce.
Imagine giving this report to a client – it’s as if their website is performing badly. Users are becoming less engaged and less inclined to buy something.
If you’re looking at Analytics for a site that’s not yours, you might not know immediately what’s happened to make the report look terrible.
Why are my metrics dead?
Before doing anything – if you ever see this little icon in the top right, click it and drag the slider to “Higher Precision” to make your data is as accurate as possible.
Look for spikes or dips in traffic and find out what’s causing them. For the Branded3 website it was obvious there was a spike from 30/4/13-1/5/13.
If it’s a dip and you know you had a problem with your website (such as it went down that day), you should exclude those dates from your analysis. Export the data from Analytics and manipulate it in Excel. If you didn’t have an issue with your website, and there was a drop in traffic for another reason (such as losing rank in Google) you should still include these dates in your analysis.
If there’s a traffic spike, narrow the date range to the days that were most affected by the traffic spike. You can do so by changing the dates in the top right:
Look around to see what may be causing the spike. The places I’d hit would be:
- Landing Page Report
- Referrals Report
- Keyword Report
Look at the top term – is it new or increased? If so, it’s probably that. If not, have a look at how many rows are in the table.
If there are a lot compared to the previous days, it means you suddenly have a lot more landing pages/sites linking to you/keywords you rank for than before.
Should I filter my reports?
Before you make this decision, you have to decide what you’re reporting in. In this case I’m interested in my conversion rate.
Conversion rate is the percentage of visitors that go on to undertake a specific action on your website (buying a product or signing up to a newsletter).
Ask yourself: is this what I want to see? Maybe a certain demographic of visitors will never convert and you should focus your efforts on the behaviours of the visitors that might. For example, if you know that only UK visitors can order products, then focus on their behaviours and increasing the conversion rate for those visitors.
Your decision to filter the data for conversion rate should be based on the questions “are these visitors likely to become my customers?”. In the Branded3 case – anyone who searched for the term “Path” or landed on the Path blog post was not very likely to become a customer – we don’t offer any Path services; we just wrote a popular article about Path.
You can use an Advanced Segment to get rid of irrelevant data, and export the relevant data to get a better idea of what your conversion rate really is. Here’s a video on how to use advanced segments:
It’s up to you whether a specific demographic of people would be potential customers or not. In our case, it was incredibly unlikely – they were just interested in reading a story about Path.
Advanced Segments or Filters?
- Can be applied at any time
- Don’t permanently affect the data
- Can be applied historically
- Best to use for excluding an anomaly in data
Using an advanced segment to exclude traffic for the Path post we get these stats:
|Conversion Rate||0.11%||0.11%||0.07%||Conversion rate dropped|
|Bounce Rate||80.78%||80.89%||80.35%||Bounce rate stable|
|Visit Duration||59s||59s||63s||Visit Duration increased|
This looks a little more normal! We might want to have a look into the drop in conversion rate, but at least it’s not as alarming as before.
- Are permanent
- Permanently affect all data going forward
- Can not be applied to historic data
- Best to be used in a completly seperate profile, to analyse a specific demographic (for example, only UK visitors)
In the past, we had one page that drove 2000-3000 visits a day because it ranked really well in Google for a high traffic term that it just wasn’t relevant for. We eventually re-directed the page. If we hadn’t, I would have recommended a filter be applied to our Analytics account to exclude that page – those visitors were not going to convert, so why should I analyse their actions?
- If your website dies, do something about this ASAP
- Figure out what you want from your reports
- Decide who your customers are and how to best analyse their behaviour
So yes, content can kill your site, but it’s not something that can’t be fixed.