Website Performance Analyzed for 1,500 Agencies

That’s a scary number (more here, from Trinity). Like a third of movie goers turning away from a theatre when they notice a long queue at the ticket box. I once worked for an advertising agency that produced amazing creative work. We had all sorts of awards at reception, including a couple of Cannes Lions. We also had a VERY slow website and wonder whether we lost business because of the slow website performance.

Fast forward 10 years and I’m wondering the same. So I went back to their website but this time armed with GTMetrix and had a look again. I’ll share what I found in a bit.

First, here’s a summary of what I discovered when I looked at many more agency websites:

The back story

A pandemic strikes and you find yourself stuck at home with too much time on your hand. Routines are disrupted and escaping to the gym is a no-go. What do you do, then? It’s obvious: check and compare the hosting performance of 1,486 websites of creative agencies in the US, UK, and Australia.

TTFB? Check. Page Speed Score? Check. DOM Content Loaded Time? Check.

If you know what this means you’re likely a developer and curious about the data. You’ll want to check GTMetrix reports of the outliers in the list and figure out, in the waterfall view, what process is particularly slow or surprisingly fast. A TTFB of 21ms? Woah!

(Some devs, instead, will head over to social media to remind folk why GTMetrix isn’t the best tool for this kind of analysis and that we should have used Google SpeedInsight instead.)

Creative types – agency owners, designers, marketers, and SEO guns – will have switched off at the first acronym mentioned. It’s complicated stuff and technical detail for a right-brained person is soul-destroying.

But page load speed is so critical to the overall experience that technicalities must be considered earlier on in the design process. A ton has been written about it already so I won’t repeat the same stats and recommendations — just have a look at this article on page speed by Cloudflare. Instead, I’ll try to frame it differently.

Here’s an example and, I suspect, a failed experiment. A month or so ago Kmart Australia, seeing their customers shift from in-store to online, decided to route website visitors to a virtual waiting room to manage capacity. Shoppers were asked to wait a couple of minutes before being allowed to browse their website.

I’m not sure how long this lasted but people were left perplexed. There wasn’t so much of an outcry on the social but a general WTF type of sentiment.

The good news is that most developers building and managing ‘mission critical’ websites make the right decisions in regards to infrastructure and capacity management. The queue paradigm is certainly a creative solution but doesn’t quite work from a business perspective.

How we checked the website performance of 1,486 agency websites

No, this wasn’t a manual process — I didn’t have that much time on my hands. Fortunately, I work with some clever people who can write scripts that pull data, work magic and spit out numbers on a spreadsheet.

Here’s what we did:

• Used Clutch to find 1,500 creative agencies, so excluding the more technical categories
• Focused on the markets where we had servers at that time – the US, UK, and Australia
• Ran scripts to automate the testing process, using GTMetrix to record key performance metrics and marrying this to Clutch data
• Spot-checked our list manually to remove suspicious entries, like the logo design agency with over 10,000+ staff located in a village in the UK with a population of 3,455.
• We only tested homepages, which gives us a limited view but of the more visited page of a website.

To get more representative performance metrics we took three ‘readings’ on different days and kept the ones with the best Fully Loaded Time result.

Here are the other ones we looked at more closely, including their definition on GTMetrix:

Please note: a website performance is affected by a ton of external factors unrelated to web hosting, the choice of CMS, theme, or how pages are coded. This was never going to be a scientific paper. Also, we only checked homepages which are the most visited and tend to be heavier than the rest.

With a good grasp of what the KPIs are and mean we are now ready to look at the data.

And the winner is…

Lyft SEO! With a minuscule Fully Loaded Time score of 360ms – that’s a third of a second!

But as you can see from the screenshot below – all 13KBs of it – they don’t quite have a website. We excluded Lyft and a number of other outliers in our analysis, as mentioned earlier.

What kind of website performance should agencies be aiming for?

Let’s set some benchmarks before looking at results: the scores agencies should be aiming for. For this, we’ve asked Vito Peleg of WP FeedBack to share some of his wisdom.

Vito has made a name for himself in the WordPress community for launching a very successful plugin — WP FeedBack PRO — A platform that makes it easy for you to manage your WordPress agency, by allowing you to capture client feedback on live websites and putting it all into a central dashboard.

He has also recently hosted the WP FeedBack Virtual Summit, a 7-day event that attracted several thousand attendees and that we were proud to sponsor.

Before all this though, he ran his own agency in London, so he is very much across the creative process and all the technical stuff behind the scenes.

Results: the 10,000ft view

Results are pretty consistent across every region, with the exception of Australia that returned a particularly high TTFB average. This may be due to Australian agencies hosting their websites abroad chasing American or European business. Or just opting for the most discounted offer they find online. Hard to say.

TTFB

TTFB is slow across the board. American websites are the quickest off the mark but an average of 617ms is sluggish.

Let’s see what’s slowing down one of the websites on our list with the highest TTFB, which also happened to have a Fully Loaded Time close to 14 seconds.

This, incidentally, is what the waterfall view in GTMetrix looks like. A Gant chart for all the files associated with a web page. There are multiple things going wrong here, starting with that huge 1.15s delay on a 301 redirect.

This screenshot will look scary to most but grab the attention of speed optimization pros as wherever you see a delay there is an opportunity to speed it up and cut the total page load time down.

First Contentful Paint

This is when, in most cases, the browser has rendered enough information for the viewer to consume. GTMetrix explains this very well in their First Contentful Paint Explained article, and was kind enough to let us use the image below shows where this step fits in the ‘Page Load Journey’.

Illustration: GTMetrix

While the hosting environment has a significant influence over TTFB, subsequent steps are more about how a page is engineered: HTML, scripts, the sequence they are added in and calls to external resources such as Google Fonts and analytics scripts. These are all components a developer has control over.

It’s interesting to see in our data that First Contentful Paint is a respectable 1.5s for all regions.

Here’s the catch: key content may have loaded but not the rest of the page’s assets. The more this is noticeable the more it becomes annoying, particularly if you’re navigating multiple pages.

Onload Time

Onload Time is where you want to get to ASAP. There may be some processes still working but by the time you hit this milestone but you will have delivered the full experience.

We’re seeing an average delay of 0.8s between the previous First Contentful Paint step, which isn’t bad and not far off Vito’s recommendation.

Looking at the numbers, it’s interesting to see that this value (apart from a few outliers) is pretty consistent across the data set. The same can’t be said of TTFB and I’m wondering whether this may be due to agencies using the usual page builders, themes, plugins, etc… so building websites that are technically similar?

I wasn’t going back to check how 1,500 websites were built so did this little bit of analysis instead, comparing the 0.8s average — Online Time minus First Contentful Paint — with homepages of six of the most popular page builders.

Congrats to Divi and Themify for taking gold with 0.6s, with Elementor only 0.1s behind. Beaver Builder and Visual Composer may just have had a bad GTMetrix day. Bearing in mind that this was a quick test with a very small data set, so as unscientific as one can get.

With our batch of agency websites performing well for this measure can we speculate that their poor performance is due to being slow off the mark with their TTFB?

Total Page Size

With an average of 5.1MB agency homepages is quite overweight. This is more than twice the 2MB HTTP archive report for desktops for the January 2020 – May 2020 period.

The assumption here is that creative agencies are designing homepages with a stronger visual punch, loading them with images, videos, and Javascript-powered effects.

While it’s good practice to keep a page lean there is nothing stopping them from building heavier ones but packing a 6.4 L V8 engine under it.

Crowd Faction: a good looking dragster

Crowd Faction is an Australian-based marketing agency offering a range of digital and print services and are Convesio customers too. We know them well.

Had they been part of the study they would have contributed to pushing the numbers down:

There were no compromises to achieve this either: their homepage looks great, provides useful info to the reader, AND is under 1MB.

We asked Martin Cartwright, owner of the agency, what tools he used to build it:

Digging deeper into the data

We found a much more interesting story in cohort data looking at Hourly Rate, Minimum Project Cost, and Employee Count metrics, that are publicly available on Clutch.

Hourly Rate

Let’s plot the average TTFB values against the hourly rate agencies shared on Clutch (bearing in mind that about a third of them did not).

The trend is inequivocabile: TTFB reduces as the rate goes up.

We see a similar trend looking at Fully Loaded Time:

Websites of agencies charging $200/hour or more are 3 seconds faster than those charging less than $25.

So you get what you pay for, right? Assuming agencies with a slow website build slow ones for their customers too.

Minimum Project Cost

A similar relationship between cost and website performance can be seen in this cohort too.

TTFB drops from 863ms for agencies offering projects from $1,000 to 699ms for those that don’t get out of bed for less than 50K.

The drop in Page Load Time is much more pronounced, though:

Could this be that agencies working on bigger budget projects dedicate more time and resources to speed optimization?

Employee Count

Does the size of the company impact website performance too? Let’s have a look.

It’s interesting to see that freelancers are building ‘lighter’ homepages for themselves, and if TTFB is a measure of the quality of hosting the performance narrative continues.

Big agencies have heavier and slower homepages and I can tell you why too. When the CEO decides to re-design the website a bunch of Art Directors, Account Managers, and Strategists gather in the boardroom to fight for a piece of homepage real estate. Art directors tend to win and their Cannes-nominated TVCs become the homepage’s background.

Ten years ago I worked for such an agency and produced this type of work, which I will tell you about next.

Grey: Famously Slow since 2010

Grey is a top-tier advertising agency that has been around since 2017 and the creative force behind some of the best ads you’ve seen.

For the record, grey.com is not one of the agency websites we looked at as part of this study. They are not Convesio customers either and it’s not a WordPress website either.

Back in my days in the Melbourne office some of the more interesting projects included building 3D tomatoes flying around the homepage of a pasta sauce brand and the Everybody Hurts campaign to raise awareness about speeding. The latter went viral thanks to its integration with Facebook; this was at a time when the API economy was just being born and Facebook was still good.

The creative team produced consistently good work and I felt that our website never did it justice. Rather than showcase our flying tomatoes it featured three lame dancing robots. In Flash.

(Please note: this is a reconstruction from memory as I couldn’t find any screenshots nor Archive.org pages.)

I was quite vocal about my dislike of our homepage and no doubt pissed Management off by calling the robots pretentious, unoriginal, and stuck in 2004 at a conference.

Fast forward to 2020 and I wonder what grey.com looks like now. Let’s take a peek.

Much better! The lame dancing robots have been fired and they’re showcasing some great content.

But does GTMetrix agree? Let’s see…

Oh, dear. The numbers are so outrageously bad that there must be something obviously wrong. Given the homepage is a series of videos in full-screen mode (which also means that the Art Directors won the board meeting room battle again) we’ll check for these first in GTMetrix.

This is what calls to Vimeo look like in GTMetrix’s waterfall view.

The thing is… I didn’t stare at a blank screen for 30 seconds waiting for the page to load up. The Onload Time was 8.9s – still very high – but the IRL experience wasn’t so bad. Not for me anyway, as I’m sitting comfortably home with a cable connection to a metropolitan hub. I suspect that if I had been living somewhere remote, I would still be waiting for the homepage to load.

Grey’s audience – big businesses and fellow creatives – are sitting behind a fast internet connection too. So there is less of a need to save bandwidth and optimize for speed.