There is a massive amount of data that shows that faster web pages lead to better returns for the web site owner; returns measured in the revenue generated, the customer satisfaction achieved or number of click-throughs achieved. The specific return depends on what the web site is designed to accomplish.
But it turns out that this relationship is actually somewhat counter-intuitive. And that may explain why not enough web site owners are placing the appropriate focus on speed.
As a browser of web sites I tend to think that once a web site reaches a certain speed it becomes good enough. And that improving the speed further doesn’t make any difference. But when we look at real behaviors we find that the opposite is true. Above a certain speed very little improvement in the return is achieved. But below a certain speed an exponential return is achieved. Look at the graphic below.
This data is from a presentation given by Walmart. It shows a clear inflection point around 3 seconds. Above 3 seconds not much change is made to conversion rate (the number of people who stay to buy something). But below that there is a dramatic improvement.
When that analysis is extended to many web sites and many consumer interactions we find exactly the same relationship. And that also explains why 2 seconds has become the target for page load times. 2 seconds is the number, in 2012, which people should strive to achieve. I say 2012 because actually that number has changed over time. Five years ago the inflection point would have been around 11 seconds.
But hang on a minute. If I have a 12Mbps cable broadband in my house and my neighbor has a 1.5Mbps DSL broadband connection how can we possibly both experience 2 seconds? To put that in perspective if everything upstream of the last-mile connection is identical, and both of our home environments are the same, the page load on the DSL connection will be about 7 times slower. And if another neighbor was on a 56 Kbps modem it would be about 42 times slower.
To try and answer that lets first look at what people are using.
But this isn’t very helpful. All this really says is that there are three physical paths into people’s houses; copper, fiber and cable (coax/fiber). Within each of those the speeds still vary dramatically. DSL ranges from less than 1Mbps up to 40Mbps depending on the type of DSL and the distance the copper extends from the house to the first piece of active equipment.
So if you are a web site owner what should you target? How should you think about the metrics that determine if you have a good website when it comes to speed. If the industry keeps telling you 2 seconds is the target, but you know that the distribution of broadband speeds varies dramatically what should you do?
I think you need to have two target measures.
First you need to be aware of the distribution of access technologies that your customers are using. But that isn’t the only thing that your customers choose that affect the speed at which they receive your web pages. Their choices include; broadband technology, geographic location, time of day, DNS service, browser, operating system, computer, WiFi device. So I would suggest using a recommendation from a leader in this space – Walmart. Measure the maximum load time that a certain percentile of your users falls inside. Walmart measures the 95th percentile and has a target of less than 20 seconds. This encourages them to think about their entire customer base.
An example of this sort of distribution is shown below and comes from a deeper analysis on this subject by Ilya Grigorik.
Secondly, choose the set of variables from those above, that you expect to deliver a 2 second page load time. To make that easy lets say a 10Mbps technology (that could actually be DSL, Cable or Fibre) in the same country as your web site with a user running IE8 on Windows 7 over WiFi. This enables the web site design team to be focussed on squeezing out the last millisecond in order to get to 2 seconds.
Then you need to measure and analyse the data as often as possible. This is a very dynamic environment. Not only does your web site change but also all of those choices made by your customers have an impact on the distribution of page load times.
This may all seem horribly complex but it is manageable if you recognise the variability, have a few simple target measures, collect and analyse the data regularly and then above all focus on speed.