How BMW creates instant online experiences with Speed Kit's predictive preloads

Bayerische Motoren Werke AG (BMW) is one of the largest premium car manufacturers globally. Offering speed, agility, and a premium experience with the help of innovation, which is at the very core of BMW’s identity.

Providing a superb user experience on the web across 118 countries is paramount for a consistent brand experience. This is a challenge Speed Kit already helps with through its intelligent caching solution.

Aiming for a frictionless car shopping experience, BMW and Speed Kit worked to further reduce already optimized loading times by delivering instant page navigations through an even more innovative technology: predictive preloads.

Explained: Pre-caching and pre-rendering

Speed Kit's predictive preloading technology consists of two parts:

• Pre-caching: Pages that the user is likely to navigate to next are already requested and fetched into the Speed Kit cache, ready to be rendered quickly when the user actually requests them. This reduces page load time by minimizing network delays.
• Pre-rendering: Pre-cached pages are rendered in a hidden browser tab so that when the user actually requests the page, it can be displayed immediately. This further reduces page load time by minimizing render time.

Risks of pre-caching and pre-rendering

While pre-caching and pre-rendering pages before the user navigates can significantly enhance loading times, it introduces several challenges:

• Predicting navigations: The key to the approach is to predict the user navigations correctly and early, while making sure network and processing overhead do not effectively slow down the experience, instead of improving it. Basic techniques, like pre-caching on link hover, often fail to meet these requirements and can actually make the process inefficient or even counterproductive.‍
• Increased infrastructure load: Pre-caching pages for anticipated user navigations increases the infrastructure load on a website. To cover a wide range of potential navigations, HTML traffic usually increases by 5-20 times. This significant increase can be challenging to manage for complex websites, such as e-commerce platforms, and may become prohibitively expensive.‍
• Undesirable session side effects: Each pre-cached page has the potential to alter the data stored in the server regarding the user’s session, even if the user never actually navigates to the page. This can disrupt personalization, and in some cases, may even log the user out unexpectedly.‍
• Skewed analytics tracking: When pages are pre-rendered in the background, user tracking may be triggered even if the user does not navigate to that page. Pre-caching alone can also affect server-side tracking, distorting analytics data. This not only increases tracking costs but also compromises data quality, which must be avoided to maintain accuracy.

With these challenges in mind, pre-caching and pre-rendering must be tackled in a novel way to ensure that fast loading times do not come with some very unpleasant side effects.

Speed Kit's predictive preloads

Speed Kit’s predictive preloads feature effectively addresses the challenges mentioned earlier, enabling BMW to achieve faster load times with instant navigation.

Before reviewing the results, it's important to understand how Speed Kit ensures a smooth and efficient implementation of pre-caching and pre-rendering through its predictive preloads:

• Prediction: Speed Kit utilizes a combination of machine learning models, real user monitoring data, and user behavior insights to make the most accurate predictions.
• Server load: By delivering all pre-rendered pages from its edge infrastructure, Speed Kit ensures that there is no impact on the origin infrastructure.
• Client load: HTML file sizes are reduced by up to 90% through dictionary compression, and JavaScript execution for pre-rendered pages is optimized to minimize client-side load.
• Side effects: To prevent unwanted side effects and conserve CPU resources, JavaScript is only executed when the pre-rendered page becomes visible to the user.

An A/B test was conducted to measure the performance impact of predictive preloads, building on the already highly effective Speed Kit acceleration. The results demonstrate the high accuracy of the machine learning model and its significant impact on performance.

The approach successfully predicted and pre-cached 88% of the site’s navigations. This means that for the vast majority of navigations, no network request was needed to load the page. Of the pre-cached navigations, 79% were pre-rendered in the background, effectively eliminating the render overhead on the user's device.
‍

The histogram of load times (measured via Largest Contentful Paint) highlights the impressive performance improvements achieved through pre-rendering. Approximately 60% of navigations were displayed to users instantly (below 300 ms).
‍

Conclusion

Predictive preloads is a huge leap forward towards our goal of achieving a web without loading times. The implementation of predictive preloads marks a significant milestone in BMW's journey toward a frictionless web experience. By leveraging machine learning to predict user navigation and pre-render pages, the solution not only drastically reduces load times but also ensures that the user’s experience is immediate and seamless.

The results from the A/B test clearly highlight the effectiveness of the approach, with 88% of navigations being pre-cached and 79% of those pre-rendered. This leads to a substantial reduction in page load time, with 60% of pages loading in under 300 ms. These improvements contribute directly to the goal of providing an ultra-responsive, premium web experience that aligns with BMW’s brand promise of agility and innovation.

Speed Kit’s predictive preloads not only overcome the challenges associated with traditional pre-caching and pre-rendering methods but also ensure that the solution is scalable and sustainable, without burdening BMW’s infrastructure or affecting analytics accuracy.

This case study demonstrates how leveraging machine learning and intelligent caching can transform web performance and user experience, setting a new standard for how modern websites should perform.