Game Creator Success on Twitch: Hard Numbers

This story begins when the Hearthstone beta exploded on Twitch 3 years ago. My co-workers were calling in favors for beta invites. Broadcasters were spending hundreds of dollars on packs. It became one of the largest communities on Twitch even though fans wouldn’t be able to play for 3 months. No one who followed Hearthstone’s rise on Twitch would be surprised to learn it made $250 Million this year.

Hearthstone’s success raised the line of inquiry that I’ve only recently acquired the data to really explore. Just how much does exposure on Twitch do for game creators? Forget views, uniques, and other reach metrics. We attribute 25% of Punch Club and The Culling’s sales to Twitch. And we attribute increases in week over week player retention as high as 15% to Twitch [1, 2]. Pick up the money you are leaving on the table. Invest in your Twitch community.

Watch then Buy

Twitch provides the most authentic preview of a game you’ll find on the internet. That’s because it replicates the experience of watching your friend play from their couch. Is the hype real? Do you kind of want the controller? Should you buy this game right now? Contrast that with watching a two minute trailer loaded up with cut scenes. That’s my intuition for how Twitch helps viewers make purchasing decisions.

Showcasing gameplay is great if you’ve made a quality game. Let’s take a look at some games Twitch has helped.

Huge! I found that when a Steam connected viewer watched a game on Twitch, their odds of purchasing the game within 24 hours went up substantially. So I attribute purchases that fit this “watch then buy” pattern to Twitch. Want more discussion of methods [1], assumptions [2], and the causality argument [3]? I totally get that, check out the footnotes.

Build your Community Pre Launch

If you missed it, tinyBuild launched a Twitch plays Punch Club channel with the novel slant that the game would be released when the people in chat beat the game. Super fun and creative. After Twitch beat the game, tinyBuild kept doing things right, and they gave a bunch of broadcasters keys right after the game launched to keep up momentum. “Getting press is now more of an afterthought” at tinyBuild. All this is old news, so let’s get to the numbers.

Within 6 weeks, 1.2 Million viewers watched Punch Club on Twitch. The viewing experience was so compelling that 2.8% of steam connected viewers went on to buy the game [1]. Given the assumption that steam connected viewers (0.53% of views) behave similarly to Twitch’s global viewership, I estimate 25% of Punch Club sales are directly attributable to Twitch. [2] Punch Club’s wildly successful launch showed that Twitch can be the end all and be all marketing plan. This has become so apparent that developers have begun to build their games from the ground up for Twitch ushering in the Stream First genre.

Part of Punch Club’s success was its novelty, so it doesn’t provide an off-the-shelf template you can duplicate. What you can mirror is making sure you strongly engage your community on Twitch pre-launch. Give out early access (Hearthstone), get game design feedback from broadcasters (Squad), or consider a deep integration with Twitch (Punch Club).

Engage Mid Tier Influencers

The natural focus is on the absolute top tier influencers, but I attribute 46% of sales to broadcasters that average between 33 and 3333 concurrents.

Smaller channels are more like your friend’s couch and less like a stadium. Mid tier broadcasters convert views into purchases 13 times more effectively than top tier broadcasters, and small broadcasters convert views into purchases 1000 times more effectively than top tier broadcasters. This is strong evidence that Twitch does more than just remind potential buyers that your game exists. Some channels are much more likely to convert a viewer into a buyer. The biggest channels on Twitch have massive reach. Much of their reach comes from their ability to influence other broadcasters. Let’s get specific and talk about how Lirik helped Hurtworld.

Lirik’s audience voted him to play Hurtworld, a hardcore multiplayer survival FPS on a Sub Sunday. After Lirik, one of Twitch’s most prominent broadcasters, played the game many mid-sized broadcasters followed suit. I estimate 10% of Hurtworld sales on Steam are attributable to Twitch, while 52% of those sales were from channels that averaged between 33 and 3333 concurrents. I credit less than 10% of sales to channels that average 3333 concurrents or more. Hurtworld is the perfect example of how important the fat middle is to move your game.

There’s no way around it, you need to start by making a game people want. 83% of sales were from broadcasters that streamed Hurtworld on at least 3 separate days, and 80% of sales were from broadcasters that streamed it for at least 10 hours. You are going to need to find the influencers that like your game. Lirik’s influence comes from his ability to get a game in front of mid-sized broadcasters, not just his viewing audience. If Lirik isn’t in the cards there are other paths to that goal. Developers can ask for our help working with broadcasters at dev.twitch.tv

Twitch Increases Player Retention

Retention matters a lot. Products that can’t retain users stop growing and eventually disappear. Twitch users engage in a number of complementary activities including viewing, chatting, and broadcasting. All of these activities lead to increased retention in the associated game. The more invested someone is in your game, the more likely they are to buy cosmetics, DLC, and sequels. I can’t put a monetary value on it, but I’m sure you know what retention lifts like this are worth to you.

Twitch is helping all sorts of games retain their players. And almost half of the observed impact is attributed to the effect of watching, rather than the increased engagement of watchers (number of days played). I separated these two effects by looking at what the improvement in retention would be if viewers had the same in game engagement as non-viewers (the same days played distribution). I’ll show how and why I did that using Dota 2, a personal favorite of mine, as an example.

I got excited when I found that viewers who watch Dota at least once on Twitch had a 5% improvement in week over week retention [1, 2]. Meaning a user who played Dota in week 1 was 5% more likely to show up in week two if they watched the game at all on Twitch in week 1. That’s super exciting, because moving retention even a single percent is crazy hard, and has an exponential upside.

This is a huge effect, but that could be because well retained Dota users are more likely to watch Twitch, not that Twitch led to increased Dota retention. I hoped that I could control for that by bucketing players based on how many days a week they play Dota. Good news! As you’d expect, the more you play week one, the more likely you are to play the week two and so on!

It’s not surprising that if a player played Dota everyday last week, they are extremely likely to play at least once this week. Let’s further segment players on whether or not they watched Dota on Twitch the previous week.

But regardless of how much you play, Twitch viewers are still more likely to play next week than non-Twitch viewers. So, assuming the relative improvement above, I looked at what the improvement in retention would be if viewers had the same days played distribution as non-viewers. Boom! 80% retention up from 77%. I attribute about half of this boost to the complementary activities Twitch viewers engage in. Moving retention up 3% is really hard and worth a lot over time. So figure out your Twitch strategy and build up your community. Get started by promoting your existing broadcasters, focusing your community manager on Twitch, and creating original content. Collaborate with us, we’re here to help.

Conclusion

There is now strong public evidence that game creators should foster their Twitch community both pre and post launch in order to move units and retain players. This means building out your community by giving out early access, getting game design feedback from broadcasters, and considering a Stream First approach. All game devs should get their community manager focused on Twitch and start a dialogue with us. The Developer Success team is committed to helping developers leverage Twitch to everyone’s benefit. We’d love to help. Reach out to us through dev.twitch.tv.

Thanks

Kathy Astromoff who first presented many of these findings at GDC.

Max Lee for his engineering work.

Alex Matzner who designed the visualizations.

Brooke Van Dusen, Soucindar Hoche, Aaron Krasnov, Drew Harry, Brendan Rocks, Chase, Kevin Lin, Lenny Simon, and Jason Maestas who helped edit this post.

Footnotes

1. Methods

• Hearthstone was a top 3 community means it sustained viewership

• Data source: Steam API. Sample of 100,000 users that linked accounts for chance at random CSGO and Dota 2 drops. Our sample is 0.53 % of Twitch views.

• Attributed games sales were taken from a 5 month window. Retention looked at over a 3 month window. Steam only.

• Actual game sales estimates are from steamspy.

****• Attribute a game sale to Twitch if a game purchase happens in the next 24 hours after a view of the game on Twitch. If multiple broadcasters were viewed, split attribution evenly.

• I didn’t find a minute watched or video play threshold that substantially outperforms a single video play in predicting retention or sales. My intuition for this is that a single minute of authentic game play can convince you that you want to be playing this game, not watching it.

• Retention boosts are given in terms of the affected viewers. The overall boost in retention will be proportional to the percent of players that watch the game in a given week.

• Community size on Twitch is based on minutes watched by game.

2) Assumptions

• Assume Steam connected Twitch views are equally likely to lead to a game sale as all other twitch views. Users connected accounts for the chance to win Dota 2 or CS:GO items, so they are definitely more engaged in Twitch, Steam, and gaming than our typical viewers. This assumption is definitely the most likely source of error; it’d be completely reasonable to attribute uncertainty of +/-25% to just this assumption for the all the game estimates.

3) Causality

• Punch Club viewers were 4.6 more likely to buy the game within 24 hours than non viewers. The converse, purchasers were much more likely to watch within 24 hours is not true. That’s evidence we don’t have correlation with causality flowing in the opposite direction.

• The strongest alternate hypothesis to viewing influencing purchase intent is that users seek out games they are already very likely to buy. The strongest evidence that Twitch is influencing purchase intent, is the variance different broadcasters have in selling games. Mid tier broadcasters convert views into purchases 13 times more effectively than top tier broadcasters, and small broadcasters convert views into purchases 1000 times more effectively than top tier broadcasters. This is strong evidence that Twitch is more than a pitstop on the way to checkout. Twitch does more than just remind potential buyers that a game exists.