Site performance best practices

  • Updated
  • Optimizely Web Experimentation
  • Optimizely Personalization

Optimizely Web Experimentation and Optimizely Personalization work to ensure that the snippet delivers the best possible experience to visitors to your site, but a few best practices can improve optimal site performance. Follow the guidelines below if you are concerned about performance issues or page flickering.

See whitepaper on optimizing the performance of client-side experimentation to learn more.

Use proper snippet placement

Preview mode uses a different snippet than the one in production on the live site and is often slower. For this reason, Optimizely does not recommend using the preview mode when investigating the performance impact of Optimizely Web Experimentation or Optimizely Personalization.

Place the snippet as the first script tag in the head of the page after all charset declarations, meta tags, and CSS inclusions. If you use jQuery within variation code and jQuery is already included natively on the page, place the snippet directly after the jQuery.

Include the snippet in the server response

Include the snippet directly in the response from the server. Do not deliver the snippet with any tag managers or inject it with client-side scripting.

Self-host the snippet

Self-host the snippet to improve load times. For more information on self-hosting the snippet, see CDNs and Optimizely Web Experimentation.

Deploy the snippet synchronously

Installing the snippet synchronously helps prevent page flickering. Asynchronous loading eliminates any delay in page load times but greatly increases the chances of flashing. Learn more about synchronous and asynchronous snippet loading, including the strengths and drawbacks of both load types.

Use preconnect and preload tags

Add preconnect and preload tags to cdn.optimizely.com at the top of the head for faster synchronous loading. Optimizely recommends using preconnect to open a connection to logx.optimizely.com, the event endpoint, beforehand.

Replace //cdn.optimizely.com/js/12345678.js in the example below with your snippet.

<!DOCTYPE html>
<html lang="en">
    <head>
        <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
        <link rel="preload" href="//cdn.optimizely.com/js/12345678.js" as="script">
        <link rel="preconnect" href="//logx.optimizely.com">
        <!-- Add other meta information here -->
        <!-- Add stylesheets here -->
        <script src="//cdn.optimizely.com/js/12345678.js"></script>
        <!-- Add scripts and other content here -->
    </head>
    <body>

Follow jQuery recommendations (if your site uses jQuery)

The snippet does not require jQuery to run, so you can remove jQuery from the snippet and your site. If you write experiment code in jQuery, you must make jQuery library 1.11.3 (minimum) available to the snippet.

Optimizely recommends including jQuery natively on the page, synchronously, above the snippet. Optimizely can execute immediately using this library with no page flickering.

Minify any jQuery and load it from a fast, trusted source. This load time has little or no impact on page load times and experiment execution. You may continue to use jQuery if you prefer, but it is not required.

Use the holdEvents and sendEvents APIs

Use Optimizely Web Experimentation’s holdEvents and sendEvents APIs along with hold or send event APIs to hold events upon initial page load and send them after page load is complete in consolidated batches. This consistently reduces page load time.

If you choose not to use the holdEvents and sendEvents API, you should use preconnect.

Increase the Cache Expiration (TTL)

Increase the snippet Cache Expiration (TTL) to a time that makes sense for the pace of experimentation, such as 10 or 20 minutes. Longer cache expiration times improve site performance. The default setting is 2 minutes.

Minimize the number of pages and events in your projects

By default, the basic snippet includes all pages and events you define in a project. Also, all pages you define for inclusion in a custom snippet are in the snippet as delivered on the page, regardless of their attachment to a running experiment or not.

To keep the overall snippet size small, avoid creating pages where you do not expect to run experiments, and archive any unused pages. Try to limit the number of pages and events in your projects to the minimum required to execute active experiments. You can also exclude unused pages from the snippet via automatic page trimming.

The snippet excludes unused audiences and experiments, so archiving unused audiences and experiments does not affect snippet size. Also, running A/B tests as standalone experiment objects or underneath Personalization campaigns does not reduce snippet size or improve performance. All experiments and campaigns are represented with the same JSON data model in the snippet.

Each experiment is limited to a maximum of 1,048,572 bytes.

Use custom snippets 

Create a custom snippet to reduce the size of your snippet and improve site performance. 

Use custom code and extensions to deliver substantial changes within Optimizely Web Experimentation

Increasing the snippet size diminishes site performance. One main cause of increased snippet size is unminified JavaScript, CSS, or jQuery code, which you might add to your site if you are testing substantial site changes like a page redesign.

Currently, you have three options for delivering substantial site changes within Optimizely Web Experimentation. The best options in terms of minimizing snippet size are to:

  • Use Visual Editor to make changes instead of extensive JavaScript, CSS, or jQuery code.

  • Minify large blocks of custom variation code or shared code.

  • Use extensive unminified JavaScript, CSS, or jQuery code (this option increases snippet size the most).

To deliver substantial site changes within Optimizely Web Experimentation or Optimizely Personalization, you should use a combination of custom code and extensions to deploy different blocks of code synchronously (for above-the-fold changes) and asynchronously (for below-the-fold changes):

  • Use an extension to define the above-the-fold changes as synchronous.

  • Use custom variation code to define the remaining page changes as asynchronous.

You can load each instance of an extension called into an experiment synchronously or asynchronously. Create as many extensions as you need, use multiple extensions in one variation, or use the same extension multiple times in one variation.

See the options in Use show/hide elements to deliver substantial changes outside of Optimizely Web Experimentation section for more effective snippet size reduction and performance improvements.

Use show/hide elements to deliver substantial changes outside of Optimizely Web Experimentation

To deliver substantial site changes like page redesigns outside of Optimizely Web Experimentation, like redirect experiments, you should embed all images and HTML blocks in the native HTML page and use show/hide to create the versions of the page you are testing.

You can also use an Optimizely Web Experimentation redirect test or host large code blocks as minified code in an external public file and call them into the experiment.

Each option is more effective than the options for using custom code and extensions to deliver substantial changes within Optimizely Web Experimentation for successfully reducing snippet size and increasing performance.

Use analytics to target your testing

Use an analytics tool to identify traffic that represents your visitors so you can optimize your site for the majority of people who visit. For example, if you find that most of your traffic is from mobile devices, you can target your experiments for mobile users. You can also target your testing for 4G LTE and faster speeds.

Use time to first paint rather than time to interactive

If you want to measure bounce rate, use the time to first paint (TTFP) metric rather than time to interactive (TTI). TTFP can be measured, whereas TTI is less measurable because many factors can influence when a visitor actually interacts with a page after it is ready.