Fast Fashion… How Missguided revolutionised their approach to site performance Mark Leach, Missguided Andy Davies, Eggplant DeltaV Conf, May 2018

Mark Leach Missguided @AndyDavies Eggplant

In the beginning…

Conversions What if we improved our sites speed

April 2017 Android Median iOS Median

Base of good practices already in place Using HTTP/2 and CDN Compression enabled for text resources (& minification) Images optimised using a specialised service Responsive images Bundling assets

The early fixes

< link

rel

" preload "

href

" https://static.missguided.co.uk/skin/ frontend/mgresponsive/default/fonts/missguided/missguided-bold.woff2 "

as

" font "

type

" font/woff2 "

crossorigin

< link

rel

" preload "

href

" https://static.missguided.co.uk/skin/ frontend/mgresponsive/default/fonts/missguided/missguided- light.woff2 "

as

" font "

type

" font/woff2 "

crossorigin

Pre-loading fonts speeds up rendering

Browser downloads fonts before render tree is built

faster rendering

Homepage images received but page hasn’t started rendering yet What’s delaying rendering?

Optimizely was very chatty

June 2017 Android Median iOS Median

Summer 2017 Proving the value of performance

⭐

⭐ Customers love reviews But some review services can have a large impact on performance

Median Page Load Time (s) 0 2 4 6 8 10 12 14 Week 0 Week 1 Week 2 Week 3 Week 5 Week 5 Android iOS So what happens when you remove it?

Median Page Load Time (s) 0 2 4 6 8 10 12 14 Week 0 Week 1 Week 2 Week 3 Week 5 Week 5 Android iOS Removed for Android visitors only So what happens when you remove it?

0% 25% 50% 75% 100% 125% 150% 175% 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Week 0 Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Page Load Time (avg across session) Cumulative Revenue vs Week 0 Effect on Android revenue

0% 25% 50% 75% 100% 125% 150% 175% 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Week 0 Week 9 Page Load Time (avg across session) Cumulative Revenue vs Week 0 Android revenue increased week-on-week

56%

Nothing stays still in retail!

-4s

£ +26% (Baselined against other platforms

August 2017 Android Median iOS Median

https://www.flickr.com/photos/rs_butner/7505567608

Winter 2017 “Can we make it to three seconds?”

Assembled a performance team https://www.flickr.com/photos/soldiersmediacenter/7112630285

Working? Useful? Usable? Focused on reducing time to useful and usable Especially rendering delays

https://www.flickr.com/photos/poetatum/3457696479 No magic go faster button

https://www.flickr.com/photos/colleenmorgan/2721589361 “Frontend Performance Archaeology” Katie Sylor-Miller

Visualising JavaScript and CSS bundles can be a great conversation starter

Visualising JavaScript and CSS bundles can be a great conversation starter “How much of this is used?”

Visualising JavaScript and CSS bundles can be a great conversation starter “How much of this is used?” “Suspect we can remove this”

Visualising JavaScript and CSS bundles can be a great conversation starter Shrunk download size of core script bundle by 42% (uncompressed size by 55%)

Performance Improvements were mostly dull, boring stuff Removing unused / un-needed code Removing duplicate styles Replacing social media libraries with ordinary links Moving 3rd party libraries onto Missguided domain Adding Resource Hints - preconnect, dns-prefetch Defering loading until later where possible

Cleaned up Optimizely Switched to Optimizely to use jQuery on the page rather than bundled version Removed A/A tests (were being used as Hotfixes) Stripped out duplicate plugins, experiments for other environments, and expired ones

https://www.flickr.com/photos/derekbruff/9759290413 There were a few bumps along the way

<picture> <source type = "image/webp"

srcset

"foo.webp"

<source type = "image/jp2"

srcset

"foo.jp2"

<img src

"foo.jpg"

</picture> picture element enables format selection

<picture> <source type = "image/webp"

srcset

"foo.webp"

<source type = "image/jp2"

srcset

"foo.jp2"

<img src

"foo.jpg"

</picture> picture element enables format selection Browsers that support webp

<picture> <source type = "image/webp"

srcset

"foo.webp"

<source type = "image/jp2"

srcset

"foo.jp2"

<img src

"foo.jpg"

</picture> picture element enables format selection Browsers that support jpeg2000

<picture> <source type = "image/webp"

srcset

"foo.webp"

<source type = "image/jp2"

srcset

"foo.jp2"

<img src

"foo.jpg"

</picture> picture element enables format selection Everything else

https://bugs.webkit.org/show_bug.cgi?id=179231 It’s a Safari pre-loader bug! The type attribute in <source> tags are ignored by the preloader. A common pattern is to use the

<source> tag for content negotiation selection to specify webp for chrome and jp2 for safari. For example: <picture> <source type = "image/webp"

srcset

"foo.webp"

<source type = "image/jp2"

srcset

"foo.jp2"

<img src

"foo.jpg"

</picture> However, the HTMLPreloadScanner only considers the media query when selecting (or not) the appropriate <source> element. As a result the preloader greedily selects the foo.webp and then later requests the correct foo.jpg. It should also evaluate the `type` attribute.

Optimizely snippet got larger (temporarily) While switching out Optimizely’s jQuery some extensions needed to be duplicated This made the bundle larger and slower until the migration was completed

So where are we now?

April 2017 Android Median iOS Median

June 2017 Android Median iOS Median

August 2017 Android Median iOS Median

April 2018 – HUGE improvement over 12 months

April 2018 – HUGE improvement over 12 months Android Median iOS Median

What did we learn?

Linking revenue increases to performance improvements is still hard… Many other factors to account for including: Pricing Promotions New brands Trends Seasonality etc.

Optimizely needs careful management Easy to blow-up the script size Optimizely are working on features to help They also capture performance data and will share

Perseverance Pays

Where next?

Add performance monitoring to build process

Further Optimizely improvements Move experiments into React f ront-end and use Optimizely for feature flags

Did we make it to 3 seconds?

Over May bank holiday weekend median was just above 3.5s

Thank You! https://www.slideshare.net/andydavies