Changing the way feature rollouts work
I would like to discuss the beta or new feature rollout system and how, in my opinion, it should be changed.
As of now, the way Discord chooses who and who not to distribute new features to in a rollout (such as the new profile customization beta as well as previously the reply feature, etc.) is completely random.
Obviously, this makes sense. One of the most accurate ways to assess whether a feature will work on all platforms is to randomly select users on all platforms.
However, I don't think it is THE most accurate way. It would likely be more efficient and useful to rollout new features in a way that accounts for certain factors.
For example, X number of users receive Y new feature in Z country. Discord can further break this down to account for a variety of factors. 1000 users from each continent, 333 of them mobile, 333 of them desktop, and 333 of them web browser.
In basic terms, I'm proposing that discord uses a stratified random sample for feature rollout, as opposed to a simple random sample.
I know that there are a variety of factors to consider in a stratified sample here, but I'd argue that this is essential to feature rollout. This way, if there is an issue with the new feature, it can be more closely linked to a specific factor instead of having the developers hunt down the issue from scratch.
That's just a thought to consider, and I'm sure Discord already has given it some thought.
I would also like to mention one other new feature rollout method; would it not be logical to first rollout to discord canary users, then beta, then main platform?
I understand the need to see how Y feature works on the various platforms, but wouldn't it still make sense to have a progression system based on the platform.
For example, at first 1000 canary users get the feature, then 1000 beta users, then 1000 main Discord users.
I mean, they have such a great lab rat pool of users in Canary and I think that they aren't getting the most out of it.
That's all,
-Tristen
-
I would also like to add that this idea is just a basic outline. I'm pretty curious as to what people think about using a Stratified sampling method, and if I have any facts wrong about how Discord currently rolls out features, let me know c:
1
Please sign in to leave a comment.
Comments
1 comment