The biggest and most powerful lever businesses can use to increase online revenue is website speed. When pages load faster, visitors spend more time on your site, view more pages, add more to their baskets, abandon fewer baskets, and ultimately spend more.
And that’s not all: when website speed improves, so does brand perception, SEO success, and customer retention. If an online business has a slow website, it is certainly wasting most of what it is spending to acquire traffic, as well as the time it is investing in Conversion Rate Optimisation projects (which are inevitably undermined by something as fundamental as poor website performance). If a business already has a fast website, going faster will kick its revenues into overdrive.
Knowledge of the super-strong link between website speed and business performance is beginning to percolate online and even in businesses which are not traditionally focused on performance. However, the vast majority of ecommerce and marketing professionals do not view website performance as part of their discipline - or as something which should be at least partly their responsibility.
This is destined to change. If you are in ecommerce or marketing you will soon be expected to have a deep understanding of website performance - and a mastery of strategies for improving performance - just as you were expected to master A/B testing when it burst into the mainstream.
Now is the time to get ahead of the curve. In the following article you will learn how to interpret five website performance metrics which are already affecting your bottom line. Our hope is that, having read this, you will be able to take the next step, and collect and analyse these metrics on behalf of your business. Now we are getting ahead of ourselves, though; the performance hero’s journey starts here...
1. Perceptual speed index
The Perceptual Speed Index (PSI) metric was developed at Google, and is best thought of as a measurement of how fast your ‘above-the-fold content’ renders in a web browser. Lower is better - the lower a page’s PSI score, the faster the content on that page renders.
PSI will vary between customers, geographies, devices, browsers and even across different requests from the same customer, so be wary of taking a single measurement as a true reflection of your PSI. As a minimum, segmenting PSI by device and by geography will help you to understand the impact that mobile vs. desktop and physical distance from your content is having on your render speed.
This term originates from the print newspaper industry, where it was used to refer to the most important stories and pictures shown above the point where the printed paper was folded prior to distribution. In web design terms it refers to the most important content, usually including a primary call-to-action, which is immediately visible to a visitor when a page they have requested has loaded and they haven’t yet scrolled away.
As with most performance metrics, you should be wary of tracking the average when it comes to your PSI; averages hide much of the vital information about your site performance. Instead, replace the average with some later percentile metrics such as 95th and 99th percentile measurements.
Percentiles tell us the value a certain percentage of observations are equal to or less than. For example, if we have 95th percentile PSI of 2500, then 95% of our page impressions had a PSI of no worse than 2500. Correspondingly, we know that 5% of our page impressions had PSI worse than 2500. Critically though, we don’t know how much worse, which is why later percentiles like the 99th are also important.
These percentiles capture more accurately the experience seen by your audience, and when we look at some session metrics later on we’ll see why these percentiles are so valuable.
2. Time to first interactive
While PSI is a great measurement of content rendering speed, many sites provide useful interactivity before the content is fully rendered. Tracking the time until your pages first become interactive is a useful way of seeing how long you’re keeping your customers waiting before they can do anything useful. Remember: you want your customers to be able to take calls to action as soon as they want to act.
Measurement of time to first interactive is still on the experimental side, but the tools capable of measuring it, like WebPageTest, are typically accurate enough to be useful. As with PSI, avoid tracking the average time to first interactive, and instead track the later percentiles.
3. Page views per customer
You’re probably already tracking session length - i.e. the total time a customer spends on your site during each visit. However, it’s also incredibly valuable to track the number of page views made by each customer in a session. Page view counts are not only a great way of measuring engagement, they also provide the ideal basis for interpreting percentile metrics for PSI and time to first interactive. Here’s why:
The basic intuition here is that the more pages a customer views, the more likely they are to see a bad PSI (or time to first interactive). And how do we know / quantify a bad PSI score or time to interactive result? Well: we start by looking at those 95th and 99th percentiles I mentioned earlier.
This is all a bit abstract so let’s consider an example. We’ve been tracking performance data for our site for a month and we’ve determined that the 95th percentile for our time to first interactive metric is 4800 ms; in other words, 95% of our page loads are interactive within 4.8 seconds.
We’ve also determined that the average user views 9 pages. So what does this mean for an average user session? Well, we know that each individual page request has a 5% chance of taking longer than 4.8 seconds to be interactive, but what is the chance that any one of those requests will be that slow? A whopping 37%! Over one third of our customers will experience a bad page load at least once during their session. A high-quality customer experience has consistently fast pages, which is why later percentiles matter so much when tracking performance.
There is a fundamental tension between page load speed and session length. We want our customers to visit more pages, see more content and view more products. But each new page view increases the chance that the customer will experience a potentially session-killing load time. This is why it’s critical to focus on improving the later percentiles for the metrics we’ve been discussing when optimising your site’s performance.
4. Requests per page
Every web page is made up of multiple distinct resources; images, scripts, styles and HTML all go into making a good web page. The more resources we include in each page the slower that page gets, and the more chance we have of at least one resource having a slow load time.
It’s all too easy for the number of resources on a page to grow little-by-little, week-on-week - eventually adding to a massive increase in load time metrics. Tracking the resource count over time is an easy way to prevent this insidious growth and catch potential performance problems early.
For truly sophisticated resource tracking, segment the tracking based on resource type and on where it’s being served from. You can reduce the impact of resource bloat by serving more resources from a high-performance CDN like Cloudflare or Akamai rather than from your own servers. Segmented tracking can help you keep an eye on how many resources you’re serving and where they are coming from.
5. Page weight budgets
Page weight is the other side of the ‘resource count’ coin. You might only have a few resources on your pages, but if they take a long time to download then they are affecting your page load times just as much.
Establishing a budget for how many resources and the collective size (in bytes) of those resources is a handy way to set performance guidelines for your team, and to track how your pages are stacking up against that budget.
A standard budget assigns, per page, a maximum count and byte weight, for each type of resource: images, styles, fonts and scripts. We prefer to segment scripts into first-party and third-party scripts, because of the impact that third-party tracker scripts can have on page performance. It’s not uncommon these days to see trackers inflate page load times more than any other elements.
- Your PSI score is a measure of how quickly your above-the-fold content loads. Lower is better.
- Don’t make assumptions based on performance metric averages - look at the later - i.e. 95th and 99th percentiles
- To get a true measure of user experience you need to gather both your page load times and pages viewed per customer - because the more pages people view the higher probability they’ll see a slow load time
- Track the resources on your pages, segment them by type, and stop resource-bloat in its tracks
- Establish page weight budgets and stick to them!
How is your online business doing in terms of our five key metrics?
If you’d like some help working out where you are currently or with understanding any of the terms or ideas referenced above, please don’t hesitate to get in touch using the form on our homepage.
Subscribe to We're SKIPJAQ
Get the latest posts delivered right to your inbox