How The Washington Post cut its page load time by 85 percent

Two years ago, The Washington Post’s site was in rough shape. Years of tacked-on features had made it bloated and sluggish, forcing desktop readers to wait upwards of eight seconds for a page to load.

The page was “almost in the E.R. when it came to performance,” said Gregory Franczyk, chief architect at The Washington Post. His one word description of its state: “bad.”

So starting a few years ago, the Post made a concerted effort to turn that around by looking at every element on its pages, eliminating redundancies and ditching features that readers didn’t use, such as some of its recirculation widgets. Its page navigation, a major culprit that took up half of  the code weight of the page, was slimmed down on The Post’s article pages, losing its features.

The Post also adopted Google’s image format, WebP, which Google says shaves images’ size by a third without compromising their quality. The Post credits a lot of its performance gains to Pagebuilder, its custom page template software that compresses and streamlines the site’s design code.

The result: The Post has reduced its “perceived completeness time” — which it defines as the time it takes a page to appear complete to readers — to 1.7 seconds — an 85 percent performance increase compared to the previous iteration of the page. Unlike “load time,” which details how long it takes for every element on a page to load, perceived load time measures what a reader actually sees, making it a more useful metric, according to Franczyk.

The Post was far from alone in this. Many publishers have crammed their pages with network ad tags and third-party plug-ins to maximize traffic and revenue, slowing load times. This approach to performance is jarring compared to tech companies and even commerce sites, which are constantly optimizing their sites to boost user engagement. Facebook, for example, used the poor performance of publisher pages to pitch its Instant Articles feature, which it says can load publisher articles faster.

“In commerce, the one-second load time is the one that everyone wants to hit. In media, it doesn’t seem as if there was ever that much emphasis on performance,” said Franczyk.

Other publishers are also zeroing in on page performance as a potential Achilles heel — and it’s not just legacy media. In May, Vox Media declared “performance bankruptcy” for its sites, saying its load times lagged competitors. Dan Chilton, Vox Media’s ‎front-end engineering director, said that the company’s performance issues were a product of its developers being more focused on shipping code than on optimizing what Vox had already produced.

There’s a lot that publishers don’t have complete control over, however. This increased interest in page performance is driven in part by the sharp rise in mobile adoption. Pages that load quickly on desktop computers, which tend to have speedy, stable Internet connections, rarely fare as well on mobile devices, whose connection speed varies depending on their tower connections. Likewise, publishers can only do so much to reduce the resource demands of the ads on their page, which are often the most significant sources of page slowdown. That’s in part what’s given birth to the rise of ad blocking software, which can reduce data transfers by half to three quarters, according to a recent article on The Monday Note.

“We have very little control over ads that load late or slowly, but we wanted to make the core use experience as solid as possible. That’s what we have control over,” said Franczyk.