Are Legacy Browsers Worth Supporting?
|Table of contents|
The article introduces a method to determine which legacy browsers are worth supporting.
- 2019-10-22 Article is published.
In June 2019 verkkokauppa.com – the largest Finnish online retailer – was visited or tried to be visited by 34 different browser vendors, by 1518 browser versions, by 20 operating systems and by 4271 screen dimensions. That's a lot to support especially when considering the number of possible combinations.
A fundamental problem in web development since the early days has been drawing the line which browsers and devices to support and which not. Ultimately the question is how to keep the website serving as many users as possible while making use of the helpful features of the modern browsers which the vast majority of users use anyway.
When tackling the problem two questions are asked: if non-standard code (aka browser hacks) are justified, and how small ROI (Return of Investment) is justified in order to support a broader set of browsers. In other words, how much time is worth investing to support more browsers.
Perhaps we should also take a closer look if the browsers are used for anything serious like generating revenue or if some browsers for some reason are to be used more like for fun and therefore shouldn't be focused on.
Let's look at one approach how to determinate which browsers are worth supporting.
First we want to gather all data available to make as good decisions as possible. Our tools and sources are:
- Google Analytics 360 (GA) to know how the users behave.
- Browserslist (BL) to see what browsers the potential customers use.
- Can I use (CIU) to see what technologies different browsers support.
- CrossBrowserTesting (CBT) to test how the website behaves in different browsers.
- BrowserHacks (BH) to find technical solutions to specific browser errors.
While the actual data is business critical and classified a few interesting highlights can be shared from June 2019. Revenue weighted data can't be shared.
Operating systems by visits
OS Share -------------------------- Android 38.79 % Windows 29.56 % iOS 24.70 % Macintosh 5.66 % Linux 0.98 % Chrome OS 0.20 % Windows Phone 0.06 % Playstation 4 0.01 % Tizen 0.01 % Xbox 0.00 % Source: GA
Other operating systems include: Xbox, Blackberry, OS/2, FreeBSD, Nintendo WiiU, NetBSD, OpenBSD, Nintendo 3DS, Playstation Vita and SunOS.
Screen dimensions by visits
Dimensions Share -------------------------- 360 x 640 13.01 % 1920 x 1080 11.09 % 375 x 667 9.36 % 768 x 1024 4.74 % 1366 x 768 3.84 % 1536 x 864 3.48 % 375 x 812 3.17 % 360 x 740 3.06 % 360 x 720 2.64 % 320 x 568 2.63 % Source: GA
Browser vendors by visits
Browser Share ------------------------------ Chrome 53.02 % Safari 24.62 % Firefox 7.47 % Samsung Internet 3.93 % Android Webview 3.60 % Edge 2.38 % Internet Explorer 2.21 % Safari (in-app) 1.62 % Opera 1.00 % YaBrowser 0.09 % Source: GA
Browser versions by visits
Browser Version Share ------------------------------------------ Chrome 74 27.93 % Chrome 75 20.86 % Safari 12 20.05 % Firefox 67 5.39 % Samsung Internet 9 2.94 % Chrome 73 2.18 % Internet Explorer 11 1.99 % Android Webview 74 1.82 % Safari (in-app) (n/a) 1.46 % Edge 17 1.21 % Safari 11 1.21 % Chrome 72 1.16 % Android Webview 75 1.04 % Source: GA
According to this data 9.15 % of Verkkokauppa.com users use mobile in-app browsers (Samsung Internet, Android Webview and Safari in-app) which is surprisingly high number .
Browser versions used by at least 1 % of Finnish internet users
Browser Version Share ------------------------------------------ Chrome 74 36.3 % Chrome 73 16.5 % Safari 12 16.7 % Firefox 66 4.9 % Chrome 71 2.4 % Internet Explorer 11 1.8 % Chrome 72 1.6 % Samsung Internet 8 1.5 % Chrome 57 1.4 % Edge 17 1.4 % Safari 11 1.2 % Chrome 70 1.0 % Samsung Internet 9 1.0 % Source: BL
On average Verkkokauppa.com users seem to use more modern browsers than Finnish internet users in general.
Looks like the in-app browsers are gaining popularity. They should be investigated in a more detailed manner.
But the question of this study, what browsers to support? Should we focus on browsers that are used most today or browsers that could be used?
The easy answer is that let's support all browsers. The better browser support, the more revenue, right? Unfortunately in business we can't waste resources to support browsers that do not generate revenue. But first we must know the reason why a browser isn't generating revenue. Either users just don't want to use that particular browser or the website isn't working on that browser. If the latter is the case, it should be fixed – or not.
To make the decision we can't look at the current usage data because it doesn't reveal the browsers which don't work with the website. Luckily we have market data from BL to tell the browser shares of Finnish internet users.
To use BL data we must decide first how many different browsers we have the resources to test often enough to be able to keep the promise of supporting it. Let's say a realistic set could include 10-20 browsers but how to select those browsers to have as broad coverage as possible?
After making a few iterations it looks like that by supporting 11 browser versions we can attain over 93 % coverage if we assume that when a website works in a specific version (like Chrome 49) it works in all newer versions too (Chrome 57, 59 etc).
Here are the commands and data to find out the numbers:
$ ./node_modules/browserslist/cli.js "> 0.1% in FI" and_chr 74 and_ff 66 android 4.4.3-4.4.4 android 4.2-4.3 chrome 73 chrome 72 chrome 71 chrome 70 chrome 64 chrome 63 chrome 59 chrome 57 chrome 49 edge 18 edge 17 firefox 66 firefox 65 firefox 60 firefox 52 ie 11 ios_saf 12.2 ios_saf 12.0-12.1 ios_saf 11.3-11.4 ios_saf 11.0-11.2 ios_saf 10.3 ios_saf 10.0-10.2 ios_saf 9.3 ios_saf 8 op_mini all opera 58 safari 12.1 safari 12 safari 11.1 safari 10.1 samsung 9.2 samsung 8.2 samsung 7.2-7.4 samsung 4
$ ./node_modules/browserslist/cli.js --coverage=FI "> 0.1%" These browsers account for 93.66% of all users in the FI
Then we combine the versions to the lowest one to minimize the number of tests required:
Browser Version --------------------------- Android Chrome 74 Android Firefox 66 Android WebView 4.2 Chrome 49 Edge 17 Firefox 52 Internet Explorer 11 iOS Safari 8 Opera 58 Safari 10.1 Samsung Internet 4
After combining the versions we have a set of 11 browser versions to cover 93.66 % of Finnish internet users (2019-07-02). The coverage is actually a bit higher because there are also some newer but less popular versions of for example Chrome which were left out of data because of the 0.1 % threshold.
One takeaway of this study is that surprisingly many users use mobile in-app browsers. Those browsers are difficult to test and support but they could probably generate more revenue if they were looked more carefully.
The introduced approach to select browsers worth supporting is a so-called quantitative model based on hard data. But in real life the result should be adjusted by the knowledge of experienced developers who may have a good grasp what are the estimated efforts of supporting different browsers. For example supporting Samsung Internet 4 might require too much effort compared to its share.
To sum up: use data but use it wisely.
If the browsers could be tested automatically, the set could be much broader and the coverage could get closer to 100 %. CBT has an API but it lacks some browsers.
Another takeaway here is that the GA data could be monitored to see if generated revenue by some browser version suddenly drops to zero in one day. The reason could be a regression bug that has affected the specific browser to make it not work on that website anymore. Depending on the amount of the revenue loss, the bug is fixed.
The last point is a little bit out of scope of this study but the anti-pattern from the '00 that tells users the website is best viewed in browser X, had a good point that no-one is going to switch their browsers because of some website tells so, even if it's Verkkokauppa.com. But on the other hand if no-one tells the user about it, how can he know his browser is out-of-date? It's like a closed door in a physical store and not telling when it will be open. We should always offer relevant information for the user but let herself to decide what to do.