![](https://www.techmonitor.ai/wp-content/uploads/sites/29/2016/08/UploadsNewsArticle4956185main.jpg)
Google has released results from a research project that could feed into its Accelerated Mobile Pages (AMP), an innovation which it hopes will revolutionise advertising on mobile devices.
Partnering with SOASTA, an analytics company, engineers Daniel An and Pat Meenan from Google described in a blog called Think With Google how they used machine learning to develop models that would predict conversions and bounce rates for given websites.
Built on real-world data from e-commerce sites, the models correlated the impact of factors on a page, such as the number and size of images, to work out which factors would make shoppers buy from a site or abandon it.
The conversion model had a prediction accuracy of 93%, and the bounce model was even more accurate, at 96%.
The main finding was that complexity harmed conversion rates.
The number of elements on a page was found to be the greatest predictor of conversions. More images and elements led to fewer conversions.
The number of images was the second greatest predictor of conversions. The engineers wrote in the article that “graphic elements such as favicons, logos, and product images can easily comprise up to two-thirds (in other words, hundreds of kilobytes) of a page's total weight.”
Sessions that converted users had 38 percent fewer images than sessions that didn't convert.
DOM ready time, the amount of time for the page’s HTML code to be received and parsed by the browser, was the greatest predictor of bounce rate. Bounced sessions had DOM ready times that were 55% slower than non-bounced sessions.
In addition, the number of seconds it took for all of the elements on the full page to load had the biggest impact on whether a user would stay on a mobile site. The median mobile website load times for bounced sessions were about 2.5 seconds slower than nonbounced sessions.
Online advertisers and vendors have long grappled with the problem of lower conversion rates on mobile devices compared to desktops.
Part of what feeds into this need is the different experience of the mobile device. Smartphone owners rarely if ever use their mobile devices to just do one thing, tending to flick between apps rather than stay in one place. People often use their mobile devices while doing other things, activating the idle device when a notification comes up or to find the answer to a query. Mobile devices can be kept on the person and easily accessed at will. This makes users less tolerant of slow-loading pages.
Based on the research, Google had several recommendations. The first was to set budgets for performance and limit page elements accordingly.
Google also argued to avoid the use of JavaScript to improve the DOM ready time of sites and optimise the fonts and structure of web pages.
The news came as Google launched AMP for ads (A4A), a technology which separates ad requests from ad rendering, allowing for quicker ad rendering without taxing the CPU or memory. CPU usage will also be limited to on-screen ads to save battery life.
Google will hope that the research and the ad-specific launch will create greater demand for AMP overall, which aims to allow webpages with rich content like video, animations and graphics to work alongside smart ads and load instantaneously.
The study produced an open-source code available on GitHub that can be applied to any website. This will allow them to test website performance.