There is a lot of work involved in making a website fast and keeping it fast. Certainly more than this document can cover, but the key trait we see in teams that create fast web properties is this: a Performance Culture.
An important step to ensuring that everyone is on the same page is to create circles of interaction:
Here are some real world examples of Performance engineering in action:
An Edgemesh customer was able to directly correlate bounce rate with Time to Interactive (TTI). As TTI increased the bounce rate increased dramatically. Once the engineering and marketing teams were aligned to improve TTI, they agreed to remove the marketing scripts on site to the required amount. As a result, more than 30% of the 3rd party resource calls were removed from the page and global bounce rate decreased by more than 40%.
An engineering team was intensely focused on minimizing the First Paint Time through more advanced (and expensive) server side optimizations. When a new engineering member joined the team, she integrated tooling that mapped performance metrics to key marketing budget metrics (bounce rate * CPC). As a result the team found that first paint had no meaningful impact on these metrics but instead they were driven by the DOM Content Loaded metric. This metric was blocked on a 3rd party JavaScript dependency (on servers they could not tune). The team moved the 3rd party script onto their origin and made meaningful cost savings with the first quarter (enough to buy even better hosting)
An agency design team rebuilt an e-commerce site for a client. They optimized it with Google Lighthouse. When the site was deployed, Real User Metrics showed the site was actually 60% slower than the previous version despite showing better lab scores. The field metrics showed that a surprisingly large proportion of their e-commerce customers (who dominate page interactions as recurring visitors) were loading the site over '3G' and similar connections on lower end devices. The lab based optimizations (like prerender) were actually causing bad performance for real customers. The agency incorporated the real user metrics feedback, removed the prerender calls and added a Service Worker style client side optimizer to accelerate the entire experience. As a result, cart abandonment rates dropped 14% over the next 60 days.
