The Numbers Game

Follower numbers are either the most or least important metric to a stream depending on who you are talking to at the moment. This isn’t unique to streaming as we’re perfectly happy to make scorecards for anything: Citation counts, CPU benchmarks, salaries, horsepower, awards, sales, years, credentials, ELO, ancestry, land, height, if it can be measured, we’ll press it into the service of gaining status. Such measurements are never in short supply and are convenient to employ since arguments over methodology don’t carry the same appeal as the blood sport of showing someone they’re not worth as much as they thought they were. This post is going to be about the boring discussion of methodology.

Defining the question

If the question is “What measure will allow me to lord my status over another and make them feel inferior?” then the answer is specific to the person you’re talking to. The fact that we insist on asking the wrong questions results in the untrue and unhelpful assertions that follower counts don’t matter or are the only thing that matters. We are awash in data, but we seem to be short on interesting questions to answer with it.

Is my stream a success? Give me a definition of success, and I’ll find you a measure that can answer that. Fame and popularity are perfectly legitimate goals, and are more likely to correlate with measures like follower count. Does success mean respect? This seems like it would correlate with fame or follower counts, but clearly Rowan Atkinson’s fame is quite different from Stephen Hawking’s. Of course, if one is streaming in order to be respected, there may be more fundamental questions as to whether or not this is the best way to pursue that goal.

For most people streaming is a hobby. The fact that it is a hobby does not mean that a streamer cannot be ambitious or aspire to turn that hobby into a profession, but it does help to provide some perspective. Model trains are a hobby. It is conceivable that one might aspire to own the best model train set. While I’m not clear on what the criteria for the best model train set might be, I can at least hypothesize its existence. The pursuit of the best train set may provide tremendous satisfaction despite the fact that a comparison to all other train sets is impossible. Here the measurement would be against whatever Platonic ideal of train set the hobbyist aspires to. The only time we would say this ambition was a problem was if the failure to own the best train set brought more distress than enjoyment, and this would be because hobbies are supposed to be enjoyable.

Unlike train sets, streaming necessarily involves a social dimension, and so followers seem to align more closely with certain goals. Enjoyment seems to be the main goal but let’s entertain the possibility of a goal like having the best stream for a moment. What does that look like? Maybe we can warm up with an easier question: What’s the best movie? Well that’s just, like, your opinion, man. We can’t definitively answer this because it’s a matter of taste. If streaming is like any other broadcast medium then popularity is not a necessary or sufficient condition for quality, though it is for commercial success.

If we fail to keep a clear question in mind then we are unlikely to find the data we need and we will be susceptible to others telling us what we should care about. Priorities can change, but in addition to keeping us grounded, well formed questions enable us to benefit the most from the information we have and achieve our goals.

What followers mean

Someone following your channel means that at some point they pushed in a button to select into seeing your channel listed on a subset of all Twitch channels should the user select that tab. Usually this is not what people think when they see that someone followed, but this is the only certain knowledge we have. It says nothing about their intentions to ever visit the channel again (although on average it is more likely they will return), nor does it say anything about their state of mind when following or unfollowing. For instance, I assume that behind every follow there is a remarkably attractive woman who has been driven wild with lust through the quality of my content, and I am happy to report I have seen no evidence to the contrary.

We are usually safe making a few assumptions about followers. It’s not unfair to see it as a vote of support that says “I like your stuff enough to want to know when you’re live again” but this is not always the reason. When another channel (especially a larger channel) raids, there will be an influx of followers who will never see your content again, since the follow is more a show of support for the channel that brought them there. On the rare occasions I draw the ire of a larger streamer on a social media platform, I will inevitably see a few additional followers on the channel. Even if these were benign (they usually aren’t), my conduct outside of Twitch does not always align with my behavior on stream, and so the channel may not be a good fit.

A follow is basically a probability. It’s the probability that someone will come back and you’ll have a chance to entertain them again. They may keep coming back, they may not, and sometimes it may even be because of something you had direct control over. When someone watches a stream they become a concurrent viewer, which will determine the ranking of the game on the browse page, and the ranking of the stream for the subset of streamers who are playing that given game (or category). Of course, concurrent viewers can only be measured when a stream is live, and so while this is the measure Twitch cares the most about with regards to partnership and their own sorting, the measure is not as convenient when attempting to use it to compare yourself to others.

The attention paid to the probabilistic viewers in the form of followers is out of proportion to the actual increase in probability that someone will watch. There are billions of people in the world right now who will never know about a given broadcast. The probability of a given Twitch visitor watching your stream specifically is infinitesimally small. While the probability for a given follower is higher than a given Twitch viewer, our inclination assign a super-fan probability to each follower is clearly mistaken. In addition to the inflated probability we assign, we overvalue the measure in general since it is the most convenient point of comparison to our peers.

When I started on cast I had no followers, and occasional views. Sometimes people would watch and sometimes they would follow. The likelihood of them returning was somewhat high since there were no reasons for them to follow other than enjoying the content, but even then it was a while before I could stream with the knowledge that there would be at least one concurrent viewer throughout the cast. Part of my growth also came from connecting with other streamers, becoming involved in their casts as a viewer, and moderating some of them. JessyQuil has and continues to move heaven and earth to draw attention to my cast, but despite our similarities there will be a subset of our viewership who likes one but does not like the other and this is fine. What this is intended to illustrate is that if I was unconcerned with the viewers who were not watching when I started casting (my only potential source of viewership), then it doesn’t make a lot of sense to start worrying about them once they include me in the subset of broadcasts they’re informed of when they hit the following tab.

This does not mean that followers are without value or aren’t desirable to have more of. It is intended to be more precise as to exactly what it means to have a follower and illustrate how easy it is to let poorly formed questions assign an undue importance to this metric. If you are looking for a proxy of success defined as popularity, then followers can provide some estimate of what the concurrent viewership will be. But concurrent viewership is not an unknown variable as Twitch will report this to you as a broadcaster directly, display it live during the stream (and services will report their estimate of a concurrent viewership for some casters). This is a lamentably common example of a frequently asked (and highly overrated) question being answered in a misleading way with inappropriate measures.

Measures of success on Twitch

Twitch’s preferred metric for success is concurrent viewers. We know this because this is how they determine placement on the list of streamers playing a given game and is the measure that determines partnership (beyond “knowing a guy” which isn’t measureable in the same way). These align with Twitch’s goals of having as many people as possible on the site, watching ads, subscribing to channels, and cheering with bits. Intuitively it’s not the worst way to identify channels with broad appeal either since, all things being equal, people will leave channels that aren’t entertaining them, while remaining in a channels means you at least consider it suitable background to whatever else you’re doing.

On the other hand very little of this actually has to do with quality. If you are looking for quality, go watch a Stanley Kubrick movie. If the evening’s entertainment simply must be Twitch then you are likely to notice that the streams you like (likely the ones you follow) are not necessarily the most popular on the platform or even in their category. Twitch’s interests are not the same as the viewer’s interests and this is reflected in the kinds of channels you are driven to. Looking closer at the simple ranking by viewer count we see that the penalty for failing to keep a viewer in your channel (and potentially on the Twitch platform) is fairly steep in the sense that there is an active penalty in terms of placement on the channel listings. In contrast, the behavior that is rewarded is relatively minor. A viewer does not need to be engaged, entertained, or really even present so long as their browser remains on the channel.

A very common measure of success on Twitch would be partnership. If this is true for you, then yes, concurrent viewers matter quite a bit and you need to consider how you will maximize this. There are a number of well known games that people who are pushing for partner status will play to boost these numbers, though I think many partners will tell you that gaining the partner status is the start of the journey, not the destination. It’s worth noting that in addition to fairly well known exploits like bot viewers, streamers will employ tricks like embedding their streams in ads on game related wikis to boost their concurrent viewers. While I think for some set of streamers (even hobbyists), wanting to become a Twitch Partner is an understandable goal, the singular focus on concurrent views avoids asking the interesting questions about why this role is more important than any other potential ambition.

The only information partnership conveys is that an Amazon affiliate has decided to share ad revenue, enable subscriptions, and emoticon slots at some point in the lifetime of the channel (noting that the requirements for partnership have changed over time but at present there are no requirements for maintaining that status). This status does not imbue the broadcaster with unique insights, work ethic, or any special moral authority beyond whatever the audience willingly surrenders to them. There are also no assurances that people will actually use those subscription options. Audiences need to be built and maintained regardless of partnership status.

Partnership likely has the best alignment with Twitch’s priorities (though, of course, the health of Twitch is not directly tied to a given partner or broadcaster except in the aggregate), but there are other groups who have different goals. While developers may not want to broadcast on the platform, they specifically would specifically like to see the largest number of people who did not know about the game wind up buying the game, which is roughly correlated with viewership (though more on this in a future article). Hobbyists will likely have private goals that are unique to their casts, or more abstract ones as “Be the best.” To speak to my own case (and this has partly been shaped by changes I’ve made over time), I stream to show people things that interest me and hopefully convey why I find them interesting and help them to see things that way too. Streaming is also a decent way to improve on your presentation skills, and so the purpose of a channel might just simply be to provide these opportunities to strengthen these abilities.

Using measures profitably

One of my least favourite features of getting involved in communities of streamers is to see otherwise nice people behave in selfish or dishonest ways in hopes of gaining a few new followers. Some parties can be depressing affairs where it feels like a secret game was announced where you need to give away all your cards and accept as few as possible, or that having a conversation with a person below a certain follow count will transmit a terrible disease. Online this kind of behavior is seeps through into broadcast content through careless talk about ‘loyalty’ to the stream, view bots, or unwelcome self-promotion in other channels. These are depressingly common cases of people who have mistaken the measure being used with the end goal itself.

There are obvious problems with the kinds of behaviour outlined above, but they also carry with them their own corrective mechanisms. A streamer who is constantly complaining about viewer or follow numbers is simply not going to be that fun to watch, and either they will grip tighter on their remaining views and lose them, or they’ll take the hint and lighten up. Beyond simply being unpleasant to be around, this kind of focus on a follower count, or a similar metric, also leaves a streamer vulnerable to unscrupulous actors. Follow for follow schemes, view bots, and stream teams that bear a resemblance to pyramid schemes are examples of this and seem to have a robust customer base of people who have mistaken their metrics for their accomplishments.

Considering my own goals, a high number of concurrent viewers is desirable as it means the message is getting out there and I have an opportunity to share things with a larger audience than before. However, this goal also provides some very good limits on the composition of this viewership. Since I want to share the game with this audience, one that wants to tell me how to play the game or do some kind of ‘Twitch plays’ scenario is not going to fit well with this format. This does not mean they do not have a place on Twitch, but it does mean that they would be better suited to another channel, since I am going to be frustrated by them, and they will likely be frustrated not to have their expectations met. I also do find comparing concurrent views useful, but only against other casts I have done.

Comparing broadcasts is a particularly interesting case as the following dilemma will hopefully illustrate. I generally like playing adventure games in the style of Life is Strange or The Wolf Among Us on cast as it lets me talk about the story with people who may or may not have experienced it before. However, these kinds of games also tend to have a lower average concurrent viewership than a strategy game like X-COM or Crusader Kings II. The most likely reason for this is that I am competing with the game’s dialogue and so have extended periods of camless silence followed by periods of wandering so I can say what’s on my mind (so people wanting gameplay are bored to tears, and people who want to hear me talk have to wait until after the dialogue). To complicate things, the chat activity is also quite a bit higher, especially at the end, indicating that while the viewership is smaller, those who stay are engaged with the content. Viewership is down (on average), but the engagement is deeper than the aforementioned strategy game which does not inspire the same kind of chat response.

I have the data, but what’s the question I’m asking? Probably something to the effect of “Should I keep doing adventure games?” This is a simplified account of how I came to decide that these games should be ones I do as a special event or at a time I know regulars will be able to watch, since it will have the most impact. Concurrent views alone were not sufficient to measure the total engagement with the ideas behind the stream, and so chat activity (as well as more general feedback) needed to be taken into consideration as well.

The most popular measures tend to be ones that do not convey a lot of information. Take two random streamers as the top of games currently in the top ten on Twitch and compare the number of concurrent viewers. Is the one with more viewers better than the other? Assuming you even have an answer to this, the concurrent views are not likely going to be the deciding factor. In reality, the comparison by views is meaningless simply because there are so many other variables, and yet this comparison is made far more by streamers than any other mentioned so far.

Sadly, there are no shortcuts to doing this kind of analysis effectively. It requires honesty, both in terms of asking proper questions and considering the appropriate measures. This is not an exercise in moving the goalposts. Streams can shrink and die, and this information will be seen in the numbers as well as successes. The most effective questions are going to be ones that are measureable and oriented towards specific goals. What is the best time to stream is a good question to ask (and note that if you do radically change the time you broadcast, know that you’re likely going to see a dip in viewership while you re-establish yourself at the new time). Quality indicators can be tricky, but measuring things like how much you talk, the number of times you found you were lost for words, or the number of books and articles you read relevant to a game or topic you were going to cover for cast are all things you can use to identify if you are becoming a better presenter.

Moving beyond high scores

Broadcasters are best served when they stop looking at Twitch as some terrible massively multiplayer online arcade game with leaderboards, and instead use measurements against well defined personal goals. Using numbers to compare yourself against others only means bad questions are being asked and that the measurements will always be changed so you come out on top. Streamers can instead use the information provided by Twitch, their viewers, and their own private measurements to throw aspects of their broadcast into sharper relief.

This change in emphasis is also helpful when interacting with other broadcasters. A comparison of concurrent views or follower counts is adversarial and not a great conversation starter. Talking about unexpectedly successful casts, or communities that have active and positive Twitch viewers are conversations more in line with people’s original intentions in broadcasting, and are more likely to result in a friendship or at least a collaboration. Since communities and networking remain one of the most effective ways to grow an audience, this shift in perspective also carries a greater chance at channel growth in the long run.