EPTThis was Twitch’s Kixelated on a recent afternoon, moments after the lingering crowd in SingSing’s stream page were all pleasantly startled – then quickly disappointed – by the unexpected restart of audio from the neglected tab.
Some miscreant, doubtlessly
![[image loading]](http://i.imgur.com/9kdvtsO.jpg)
While the Following page (linked at the top of the front page and in the expanding left side bar seen everywhere else) will show hosted channels by streamers you follow at any time, the primary intent of the host feature seems to be to hand-off a stream’s audience at the end of a broadcast. This serves multiple purposes:
- Instead of bringing viewers to a post-stream decision point to look for another broadcast or just passively stop watching Twitch, the audience is immediately watching something else. The passive option is to keep consuming.
- Twitch is provided with more behavioral and associative data to try to leverage into something actionable by themselves or broadcasters. They have already written a piece on the effects of hosting, and while some of the most important data is a bit unclear – post-host viewer behavior – it is a start.
- Finally, hosting helps drive discovery.
There are a lot of challenges if you are driving strategy at Twitch, and most of them are interrelated. How do we leverage the data we currently collect? What key metrics are we unable to assess and how do we collect for them? How do we stabilize newer partners? How do we deal with the inevitable occasion of site outages during major broadcasts? How do we and our partners further monetize? How do we improve user engagement? How do we improve our core technology: ingesting thousands of streams and spitting them out to millions of viewers? How do we grow?
Twitch is at the nexus of a huge and complex ecosystem. Everything from the whims of individual broadcasters to big strategic moves by video game developers affects them. The success or failure of tournaments and professional players, macroeconomics, data infrastructure, even global politics: all of these things have an impact. And they still have to worry about whether DansGame is anti-aliased in Theater Mode (it’s not). The breadth of their interest is reflected on their jobs page. The depth is probably best understood by talking to their engineers. But I’m going to talk about some of it here.
Autoplay killed the video star
The core issue driving the problem Kixelated was pondering is the fact that all Twitch video players are on auto-play. There’s no technology more core to Twitch than their video players, and engineers have been working on a new, HTML5-driven player for a long time. No doubt many new features will be implemented within the video frame once this is in place, but in the meantime the binary nature of broadcasts is problematic.
Until a video frame is paused, it basically has two states: on and off. Broadcasts just “end” currently. A user unplugging their PC or being strafed by A-10 fire is indistinguishable from them signing-off verbally and ending the broadcast. As such, little information is provided to either Twitch or the end user regarding the end of the broadcast. Corner a Twitch data scientist (as creepily as possible) and ask them if they’d like to be able to easily distinguish among the following types of stream terminations:
- User sent an “end broadcast” signal to Twitch
- Unexplained interruption of ingest
- Regional internet outage
- User sent a “restart broadcast” signal to Twitch
- User sent an “end broadcast into host channel” signal to Twitch
- Twitch went boo-boo
![[image loading]](http://i.imgur.com/wK83cJR.png)
Or there's always the status quo.Of far greater value in totality than the refined data this would offer are the numerous user experience implications in distinguishing among these. But that ability doesn’t exist yet, so in the meantime, every terminated stream acts the same and every freshly booted video player acts as though there is no broadcast until it connects and blurts at you as soon as it is able.
The latter discourages some types of desirable user behavior. The one already covered is the act of simply leaving a tab open when you aren’t watching a broadcast. Having a tab randomly start to play video is one of the biggest red cards in UX on the internet. But it also discourages the use of some of Twitch’s discovery band-aids.
Discovery is more than a fallen cable channel
One of the greatest challenges by vendors and content providers on the internet is connecting people to what they want. Think of every super-specific category you’ve ever seen as a Netflix user (Visually striking critically acclaimed gritty foreign films? *CLICK*), Amazon’s ubiquitous purchase recommendations, and Twitter’s recommended follows. The case for Amazon is quite obvious: they want you to buy more stuff, so they show you more stuff they think you might want to buy. You pay a flat rate for streaming Netflix and nothing for Twitter, while both services pay more the more you use them. But Netflix wants you to find them indispensable, and Twitter wants you staying in the network and contributing, so they both work very hard at the same problem.
![[image loading]](http://i.imgur.com/uvJLtim.png)
I have two Twitter accounts I use on a daily basis.
The recommendations I get are specific to how I use the accounts.
![[image loading]](http://i.imgur.com/iruYFkH.jpg)
I haven’t bought a mousepad in a very long time but I do need one: I keep running
my mouse off the pad during heated Dota battles. How did they know!?
Twitch’s motivations for good discovery are an amalgam of these three companies. Like Netflix they have monthly subscribers to the network, but there is a significant variable cost associated with increased use. Like Amazon, a small percentage of successful recommendations will lead to a purchase – individual channel subscriptions. Like Twitter, their content is user-generated, and successfully connecting users to the content they want to see keeps people creating content. Unlike any of these services, Twitch’s discovery system is quite primitive.
They have visibly been trying to work this problem. There is a reason there are several featured streams on the front page. There is a reason they have their Partner Spotlight program. There is a reason that, if you expand your sidebar while at a channel there are a handful of suggested channels associated with the one you’re watching. Broadcaster teams have been around for as long as I’ve been a serious Twitch user.
But all of these band-aids currently pale in comparison to the site shouting at you almost everywhere you go: here’s what the most people are watching. I’ve been to Sodapoppin’s stream. It’s not my cup of tea. Telling me he’s one of the top six streamers right now does nothing for me. I can’t bring myself to enjoy League of Legends even when a streamer I like is playing it. Regardless of what’s hot right now I’d rather stab myself in the eyeballs than watch nearly anyone playing nearly any MMO – I can just close my eyes and remember my five years being the GMMT of a WoW raiding guild if I want to bore myself to death.
Showing people what’s popular definitely has its place. That place is not overwhelming primacy. Twitch seems to know this on some level; the front page leads with hand-selected channels after all. But after that it basically falls apart. Only one page shows channels hosted by streamers you follow. Only one area of the UI – usually hidden – provides you with recommendations, and these are only those channels popularly associated with the one you’re watching. They’re not tailored for you at all. Only CS:GO has sub-sorting mechanisms. Everywhere else, everything is sorted by popularity.
Again, Twitch knows this is a problem. How big a problem they see it as is a mystery. As is how they intend to work the problem.
Bro, do you even ideate solutions?
There tend to be two major constituencies regarding problems like this at tech companies.
![[image loading]](http://i.imgur.com/Sk8LhiG.jpg)
Leveraging the dynamics of social networking will obviate crude solutioneering, as digital human
interaction organically drives discovery, from the most mainstream content consumption to the longest
of long-tail use-case scenarios.
Social is indeed part of the solution set to such problems but many companies have tried and expensively failed to force social into their platforms where it doesn’t naturally belong. Twitch may successfully evolve into a social platform, which has its advantages. It can also do a lot to improve its interactions with the vital external social networks. However, the former is a long road if it is going to be successful, and the latter will only ever drive a small percentage of discovery.
![[image loading]](http://i.imgur.com/Vilpk1g.jpg)
Our sick algorithm will tell us everything we need to know about what people want to watch.
More data = more accurate. Step aside, social. This is the age of Big Data.
Of course algorithmic recommendations are a great tool. They’re also greatly enhanced by social networks thanks to the additional layer of associations they provide. Twitch already has made some progress on this but as a daily user I only discovered the obscure location of recommendations in the last week, literally while scouring the website for any features I hadn’t seen.
While both of these lines of effort have merit, the primary solution is quite simple, and synergizes with both social propagation and algorithmic discovery.
![[image loading]](http://i.imgur.com/wFWirxc.jpg)
What if – and I’m just spitballing here – we empowered users to
find for themselves the content they are interested in?
It’s not sexy. It’s not social. It’s not algorithmic. But it’s another thing Amazon gets brilliantly right: the most important tool to give users is the ability to drill down for themselves. This isn’t something easily accomplished by Netflix as you can only get so specific about shows before you run out of meaningful options. Twitter users are even harder to categorize but Twitter is a platform built on user-driven propagation and provides loads of data that can be used to create recommendation algorithms.
Twitch broadcasters can be hard to categorize. That’s where social and algorithmic discovery engines have their role. Broadcasts, on the other hand, have a lot of useful details. Details Twitch isn’t using. It’s time to change that.
(Note: Please see the fantastic Welcome to BusinessTown Tumblr for more of the above three images)
Everybody wants to be a category
Currently, the normal process for finding a broadcast is to go to a game’s directory. Alternatively you can tune in by name to a streamer, one you follow, or one linked on social media or elsewhere. You might use a group page but based on the neglected-looking interface I think that is quite uncommon. I’ve read about language selectors but have never seen one in the wild. CS:GO is the one game I’m aware of where there is sub-sorting within a game page: categorically by map, and by viewers or rank (which is then sub-sorted by viewers). PS4 and Xbone streamers can be viewed by platform.
And that’s it. And to this I say: metadata? More like mehtadata.
![[image loading]](http://i.imgur.com/4sUfRjm.jpg)
Let’s talk about how it should be. The first order of business is being able to browse broadcast types categorically. Here’s an example of how you could break this down:
- Competitive: Twitch and competitive gaming have a mutually vital relationship. While Twitch has helped create a new kind of professional gamer in general, they have also been instrumental in ushering a new age in competitive gaming. Their passion for esports is clear and backed by real money. They are even hiring a director to build a team focused on tournament organization partners . But there’s a cheaper, simpler, and arguably more important thing Twitch can do to help: give competitive broadcasts a directory. If the biggest tournament running is in World of Tanks with 300 viewers, anyone who loves esports should be able to see it simply by dropping in on the Competitive directory to see what’s shaking.
- Speed Running: absurd numbers of people tune in to AGDQ twice a year and raise crazy money for baby cancer or whatever. And Twitch have been great partners to them, making sure to spotlight them on the front page and promote them in social media. I quite enjoy watching AGDQ but I couldn’t name a single speedrunner. The utility of this category is self-evident.
- Talk Shows: broadcasters who run talk shows on Twitch currently have two options: list themselves in the Talk Show directory or, if they are focused on a single game, possibly list themselves in that directory for better exposure. The way you maximize exposure for such content is to give them the capability to be listed in both.
- Creative: this currently functions like Talk Shows. LiveWorkshop, as a standout example, has the most extreme production value to viewer count ratio on Twitch. Manny and Anuxinamoon do incredible work and have very enjoyable streams. It's too bad no one who doesn't browse the Dota2 directory will ever find them. And if they sign off you aren't breadcrumbed to the unofficial "Creative" directory. They are in a group called Game Artists which has the advantage of being curated by professionals but the disadvantage of no one using group pages.
- Charity: with the caveat that being listed in this category should require pre-approval from Twitch, this seems like the natural next step in the way Twitch supports charity streams, and it would make it easier for smaller ones to get noticed.
- Professional: professional competitive gamers drive a huge percentage of Twitch’s viewership even when they aren’t officially competing. The biggest argument against a category like this is how it would be administrated. There’s a pretty clear case for Twitch to have a Verified system like Twitter or – relatedly – Dotabuff, but who gets to list themselves as a professional and who doesn’t is a bit more complicated. Still, it is worth considering.
- Game Studio: I don’t know how often studios smaller than, say, Nintendo or Epic have streams but there’s a clear audience interest in such streams and making it easier for them to be found will certainly encourage more of them to take place.
As obvious as this seems, nothing is accomplished simply by adding these directories. They have to be meaningful implemented into the website. A film strip element on the front page directly underneath the featured streams section is a good first step. Show the directories with active listings and the viewer count. Are there 30,000 people watching Competitive? Maybe you should take a look.
The front page has two purposes: it acts as a new user's first experience with the site, and it acts as the primary launching point for finding content. An Explore page focused on the latter would be redundant with the current state of Twitch, but that itself is enough proof of the need for these changes.
It's elective! It's elective...
The above categories all have one thing in common: while not every category is available to every user, they are self-selected. They also leave out the greatest bulk of streams (though they capture the overwhelming bulk of viewers). A very smart boss I once had told me the easier you make it for people to do the right thing, the more likely they are to do it. Electing a category is quite easy, and the way the provided examples are arranged, most users will seldom change theirs.
But, while much would be accomplished with categories alone, far more can be done with a deeper approach. Let users drill down.
First up is the fairly situational application of the CS:GO approach. It isn't right for most games but where it applies it can be a great tool. This would give you the ability to subsort a game directory by factors such as:
- Map / Mode
- Ranking
- Solo / Group queue
- ????
The problem is that in order to not create a headache for Twitch admins and streamers alike, this needs to be done via API, the way it is with CS:GO. (Disclosure: my SO works for an API firm). Let's pretend for a minute that Blizzard aren't API-hating assholes. A WoW stream could be sorted by queue/zone type, and subsorted again. Battlegrounds? Arenas? What kind of arenas? RBGs? Raids? What size raid? What zone? PVP rating? Class? Dota2 and LoL have fewer meaningful options but can still be sorted this way.
I don't know if Twitch has a developer liaison evangelizing the merits of creating Twitch-friendly APIs and interfaces (do you have a safe corner for a camera?) but if they don't they really need to hire one. The case is fairly easy to make for taking streaming into consideration if you're a developer and if you're Twitch you have a lot of input to give.
So far, the above methods of categorization are all limited to one data point per category. This is simpler and better for macrocategorization and the only option for microcategorization (what we just went over). The final step is to add tags, but these, again, are not selected by the user. They're selected by the game and administered by Twitch.
The primary tag type would be genre - and you would use tags here because many games cross genres and let's not make this too hard on ourselves.
- MOBA
- RTS
- FPS
- MMORPG
- Fighting Game
- Puzzle
- Strategy
- RPG
- Card (online)
- Card, Board, Tabletop, and PnP but like IRL idk -_-? (I really don't know what to call this but it really needs some love)
- Mobile (this is coming eventually, right?)
- ETCETERA
Another tag to associate with a game would be the publisher(s). And as you go to the directory of each of these you are once again shown the film strip at the top displaying the top games in that directory by viewers, and then the usual array of broadcasts in that directory from every game, sorted by rank. While search by publisher is niche enough not to warrant a spot on the front page, top genres does. Obviously MOBAs will be the top genre the vast majority of the time and battle lines have long been drawn over who loves which, but most genres aren't so violently territorial. Encouraging people to browse a genre will likely lead to them finding content they want.
The End of the Line
As I put the finishing touches on this post, YouTube has announced their entrance into the game streaming market. Twitch's first-mover advantage is huge, but not insurmountable. So before concluding it's important to note that most people already search for VODs on YouTube. From Twitch's standpoint this might be OK - YouTube pays for streaming that Twitch might be doing at a loss. But it also points to the fact that VODs need work. There's not a reasonable world in which Twitch's video browsing is expected to be as polished as YouTube's, but Twitch has the advantage of specialization. Apply the above logic to VODs and add the ability for users to add free-form tags.
I was watching Hafu last year when SUDDENLY, a WILD METEOS APPEAR. I don't watch LoL so I didn't understand all the nerdgasming in chat. I just searched for a VOD of that on Twitch. Guess how much luck I had? Now, Hafu and whoever is working with her, have now, and will always have, responsibility for how well-curated their VODs are. But assuming they do an even remotely good job, finding something like that should be extremely easy. Assuming they're given the tools.
And giving users tools is what this post is all about. I'd love to talk about basic user friction, and the limitations of monetization for streamers, and the outsider's perspective of how limited the options for data analysis must be at Twitch, and possibly try to talk them out of the idea that monochromatic superflat is a good design language on 27" monitors. But the most important thing for Twitch to do at this juncture is give users the power to find the content they want to see, and give broadcasters more power to be found. Because as popular as the top .1% of streamers are, for the average user, when their favorite streamers are offline there's almost always something they'd enjoy watching more if only they could find it. That keeps them watching, and keeps the discovered streamers streaming. And in the end, that's what we want.
