As we all prepare for the zombie apocalypse, we start to have a serious need to understand just what “going viral” means. Or, maybe you’re a game developer who’s curious how it impacts your business. I suppose that’s a good reason as well.
“Virality” is exactly what it sounds like--it’s how, when, how often and how much one thing spreads through a group. Things that are really virulent like a bird flu--Lululemon sportswear for women or “What Does the Fox Say?”--spread because lots of people experience them and lots of those people pass them along. Other things like Spam, brussel sprouts or the latest Tesla might not spread as fast due to cultural or practical reasons. How and why things spread is important to understand.
There’s science for this, and it’s one dose epidemiology and one dose network theory. There’s a great and brief primer on this via a couple of blog posts from my colleague Brian Keegan to be found here.
Epidemiology is the science of diseases. Again, maybe you’re aware of this because the CDC was in that Walking Dead episode. Some diseases spread and some don’t. Infection may be easy for a flu virus and hard for say, dengue fever. So, one of the things we look at carefully is the rate of infection. If a person exposes themselves to 10 people, how many of those 10 will become carriers themselves?
We’re of course using an analogy here. In your business, the “spread of the disease” is the same thing as the spread of an idea, or most likely, an invitation. If you invite 10 friends to join you in Super Squirrel Motocross, how many of them are going to say yes? And if some of them join, how many of them will go on to invite their friends?
OK, so let’s briefly take a look at the idea of social networks. You are, as the cliche goes, truly connected to everyone else on the planet via six “degrees” or six people linked in a chain. If you don’t believe me, pretend you only have five friends, and that each of them have five other friends you don’t know, and each of them have five friends, etc., etc. You start quickly getting into really big numbers which surpass the population of the planet. This was most famously proven by Stanley Milgram in 1967 when he tested out how hard it would be--how many degrees--to go from a random person in Nebraska to a random person in Boston. It wasn’t hard.
1 There are probably a dozen fields each with hundreds of papers proposing statistical models of why some things go viral. Each of them focuses on some features over others, but a couple common themes stand out: Whether the contagion itself reproduces rapidly, and whether it is introduced in a place that will let it spread.
2 That story, along with some great other background--including the math behind the “Six Degrees of Kevin Bacon” idea are in Duncan Watt’s “Six Degrees.”
Looking at them tells you a lot about the “position” of a person. Imagine you’re one of the people more at the center of this graph. You’re connected to a lot of people, and within just a degree or two you may be connected to everyone. Or, imagine you’re on the periphery. You may be connected to only one or two others and it may be many more degrees to reach everyone.
You may be intimidated by the idea of using these networks in a practical way. Maybe you’re thinking “how the hell would I know the networks in my game?” Well, you may or may not have access to the tools, but they’re open-source and you have the data to populate them. The connections between people we call “edges,” and you can use log events for many player-player events to populate them. The best ones are those that signal a social connection. For example, “invited friend to friend list,” “joined party or game,” “sent message,” etc. These are all good sources of edge data. You’ll need a social networks expert or an analytics service that works with the data. We automate this process via an SDK because it’s so fundamental to our approach, and we find that it’s more than most developers want to tackle. But maybe you have a grad student who knows this stuff or you want to go learn it. Your mileage may vary.
OK, so what’s with the zombies, the dengue fever and Stanley Milgram when this is about making money using video games? Well, your game has a network of players and an extended network of non-players. Let’s imagine that your game is the bottom big circle. It’s got nine players and you’re interested in getting more. Smart cookie that you are, it’s occurred to you that these existing players are likely to be a good conduit to new players.
So, if you were fortunate enough to actually be able to see the entire picture, you certainly wouldn’t look at all of your players the same way. No, you’d zero in on that yellow or purple player at the left edge because they are connected to the outside world. You can intuit that their position is more important than the others without any kind of higher math or training.
So, now we go back to the viral metaphor. What can you do to get that person to infect--err, invite--their friends to play your game? And now you start thinking about the many mechanisms present in social games, most of which feature truly annoying approaches like pestering your friends, posting on their Facebook walls, and begging them for near-worthless in-game currency because you’re too cheap to spend $2.99 on a stack of 10 gems/energy pods/monster eggs/whatevers.
As you think about these several good and bad options, you should consider which are in fact more “organic,” i.e. which are more about the friendship than the game mechanic. Those are going to be more effective. Why? Because if they are friends, they value each other, and probably more than they value their relationship with you. Give them something real that reinforces that friendship, something that says “I like you,” rather than “I need to exploit you to reach level 13 of a game you haven’t heard of.”
Say you’ve settled on some promotion or tactic that makes sense in the context of your title, and you’ve deployed it on maybe everyone, or maybe just on those key people. Let’s assume that it’s something they send to their friends, some sort of invitation. How do we know if it worked? We have a lot of potential data points now. Consider the ingredients on hand: the social network structure, the sender of the invitation, the receiver, the action on the part of the receiver that means success or failure, and the timestamp of these actions.
If you mix and match those pieces you can get a wide range of metrics to tell you what worked and what didn’t. The absolutely most basic and simple of these is k-factor. You may have heard of this from marketing materials and it might have sounded difficult or mysterious, but it’s simply the average success rate. The formula is average invites sent X average success rate. Yeah, that’s really it. Here’s the I-don’t-believe-you-it-must-be-more-complex-than-that-reference. It’s a pretty handy shortcut number in that it tells you about the effectiveness of a campaign. If you are only running one kind of campaign, and you’re running it on everyone, then a k-factor tells you basically how well it’s performing. We provide it, and so does anyone with their act together. It doesn’t even require you to see the network.
But, k-factor is a pretty broad number. Like “average” it condenses a lot of information into one number and can seem more precise than it is. Looking at individuals is always better than looking at everyone. In games, and particularly in microtransaction games, all players are far from equal. That goes for their spending and play, but also for their network position. Those yellow and purple players noted above are much more important than the others. If they were also connected to many other potential players, that’d make them more important still. So, the moral of the story is that segmenting by network position or importance is a big deal.
In addition to k-factor, I like to focus on speed and probability. How long a campaign takes matters because it tells you something about the nature of the relationship that you’ve tapped and about how patient you should be using future promotions. Many developers and their CRM teams have a ton of noise with overlapping promotions and it’s impossible to know what worked. Let the thing play out so you know for next time.
Predictive analytics make things more interesting still. Since you can model just about anything, why not model the probability of infection/invitation success?
A precise approach will take on an A-B testing flavor. Try multiple promotions, be patient, and see which generates the best response rate. Then extrapolate up to what would happen if you did it en masse. Of course, if you also have insight into the network graph, you can use a scalpel rather than a hammer. It’s as common sense as tailoring advertising to the right demographic groups. You wouldn’t advertise cars to children or a winter coat to people living in Florida because they’re unlikely to do anything. So, why would you try to go viral via isolated or barely-connected people? You’d be much smarter to have some kind of indicator of social importance because it’s more than simply how many people you are connected to.
You can use degree centrality and dozens of other metrics for that. I prefer a number based on actions and cascades of actions across the network, which is why we developed the Social Value metric. Those with a high value are going to be much more likely to be the people who’ll successfully go viral because they’ve already shown that social pattern for playing time, spending, etc. Those who are simply isolated whales may seem important, but they aren’t viral at all. If I’m out to start the zombie apocalypse, I’m going to choose my patient zero with care and science.