This article is the second in a three-part series on prediction and using predictive metrics in the gaming industry.
If you haven’t read my first article on prediction in the gaming industry, stop here: go back to part one for a background on predictive analytics, modeling and confidence estimates.
Now that we’ve covered the basics, it’s time to delve into how game developers, specifically, can use prediction to their advantage. Prediction can be a crystal ball of sorts for game developers, but only if you have the right information in place. So the first step is to start collecting data -- and a lot of it. If your game doesn’t have data hooks, you aren’t going anywhere.
Once you have your data ready to analyze, analytics get to work. There are countless ways to slice and dice the information you have -- basic metrics like looking at Daily Active Users, Average Revenue Per User and Average Session Length -- but the metric that’s most pertinent to prediction and gaming is player lifetime value (LTV). This is basically a sum of how much a player is expected to spend in a game before they leave, or churn out. Some people look at spending to date, but what we really care about is either total, or future spending.
What we’re trying to get at here, of course, is who your most valuable players are. Not every player is created equal: For every hardcore gamer who has an LTV of thousands, there will be many more players who never spend a dime. And from a developer’s perspective, you obviously want to target the players with a high value to keep them happy, longer.
Back to metrics. LTV is a great metric to use to get the lay of the land (assuming you’re using a good model, of course), but it only goes so far. It’s kind of like looking at your players in a bubble, and just looking at how each player interacts with the game.
If you’re a developer, you’re probably shaking your head right now. Players don’t just sit down and play the game in a vacuum! They interact with other people and talk to the other players. They have clans, guilds, networks. That has to count for something, right?
This is incredibly important for prediction, and is known as Social Value. Like LTV, Social Value is a dollar amount. How can you quantify a player’s social network? If you have an online game, think of your player network like a giant web, with players all connected to each other based on their social connections. Traditional LTV metrics look at each node in the web to determine who’s the most valuable to you, and social value takes into account the entire web and all of its interactions.
Say you have a player, Player A, who spends $5 in game, and another player, Player B, who spends $1. Who’s worth more? Traditional metrics would say Player A. But your analytics program discovers that for every $1 Player B spends, Player C spends $3, Player D spends $2, and Player E spends $3. Add it all up, and Player B doesn’t look so little anymore.
Prediction is a science, but you can’t underestimate the power of human influence -- and Social Value takes that into account. Add the Social Value to LTV and you get the true Total Value of a player, which is the best of both worlds. And this is where you can see who the really big players are.
It’s worth noting that the most valuable players often aren’t who you think. Players who can spend a lot, and whose LTVs are high, are almost never most valuable players in terms of influence. The ones predicted to have the greatest net worth based on social connection and LTV don’t spend that much, but are highly influential and cause a huge ripple effect with the little they do spend. They’re known as Social Whales, and are the ones you should be targeting.
And this, for a game developer, is the million dollar question (or maybe more, depending on the game) is: Who will quit, and how can I stop them?
Of course, you can’t stop everyone from quitting. And, as you probably guessed, an answer to the first half of the question lies in predictive analytics. This is where churn rates come in. Aside from telling you who’s already quit, predictive models can tell you who’s at risk of quitting, and how much is at stake if they do. That’s an expected value number where the amount at risk is the total value of the player X their likelihood of quitting. For example, Suicide Bob has a total value of $100, and we think he’s 65% likely to quit. If there were 1,000 of him, we’d know that over time the amount at risk for any one of them would be $65 on average. So that’s the number we assign to Bob, and on average, we’re going to be right.
From there, you can take action. A casual gamer with no social upside is at medium risk of churning? Well, maybe you can let that one slide. If you have a Social Whale about to churn, though? It’s time to pull out the big guns and rethink your retention strategy, because they’ll take their whole network with them. And what number do you use? The expected value ($65 for Bob, for example).
The opposite of churn is conversion. This is when your players start to bring in money and monetize, the dream for any freemium developer. Basic analytics programs will be able to show which players have converted, and predictive analytics takes it a step further and asks who will convert, and then how much revenue they’re likely to bring in. With the right system this can also predict who will become the Social Whale, and who to target to reach everyone else.
This is by no means a comprehensive guide to all the capabilities of predictive analytics, even within the gaming industry. Eventually, prediction will get to a point where developers will be able to predict virality, monetization, and even return on advertisements.
So, where is the industry heading? I’ll cover that in the final article of the series.