Predictive analytics for games are here, accessible, and getting easier to use. Here is what you need to understand what they are, how they work, and what it takes to get them going
Are Developers Asking the Right Questions?
Most game development companies use player behavior data to inform their game development process. Primarily they use historical data to suss out which parts of their game is working and which are not.
They use traditional metrics to ask questions like:
- What’s my average revenue per user (ARPU)?
- How many players do I have today (DAU) or this month (MAU),
- What’s the most I have on at one time (peak concurrency)?
- What’s my K-factor in games?
Good answers to these questions is all very useful information to the game developer. It’s good to know what happened last week, and yes it’s good to know what’s happening today. Both are important for understanding why some parts of your game are working better than others, how you’re doing, and whether you are trending in the right direction.
But of course this is all reactive, and as your mom told you, it’s always better to be proactive.
In other words, if you can figure out not only what's happened but also what could happen with a reasonable degree of certainty, that would be something. That would be actionable. You can plan a response.
For example, what if you knew that player X is not going to spend any money for at least the next 10 days, you might think about taking some sort of action, right? If you knew that player Y is going to quit in 4 days and she's worth $100 to you, you’d do something to intervene. right?
Of course you would....if you only knew about it beforehand.
It's Like Predicting the Weather...But With People
As you delve into predictive analytics for games, you quickly discover that it's not radically different from forecasting the weather. You have data, you have trends, and you have an estimate for what’s going to happen tomorrow. The smartest meteorologists use scientific data-driven models. And over time you learn which ones you can trust.
As a game developer, you care about making predictions about people; specifically their likelihood to take certain actions. For example:
- Where do people quit the game?
- Why do they quit at that point in the game?
- Why has churn rates increased (or decreased) recently?
- What makes players go from observer to freemium?
- What makes players go from freemium to paying?
- What makes people decide to first start spending money?
- What makes them spend more?
- What makes them click on ads?
Player Lifetime Value
- How much do players spend in total?
- What is the value of player's networks?
- Are some players worth more than others?
Asimov's Predictions Are Your Reality
If you've read Isaac Asimov's Foundation Trilogy, you may remember that there's an advanced civilization that made a science out of understanding humans and societies so well that they could accurately forecast the actions of individuals.
But that’s science fiction. It's impossible to really predict the future, right?
Technically, no, it’s not impossible. And it's not magic either. It's math.
Luckily, the science here is getting easier to deal with as more and better tools become available. Do you need to hire a bunch of PhDs? You can but it’s not terribly practical (or cheap). Still, it’s important that you understand what actually happens so you understand how usable and actionable the results are.
Recognizing Patterns with Confidence
To start, let's keep it simple and imagine that only two patterns are possible in an adventure game.
- Players login and play Level 1. Some go into the Dark Forest and 20% of them quit.
- Players login and play Level 1. Some go into the Blue Bayou and 40% of them quit.
Now imagine a more realistic scenario where players are doing maybe 35 things and they do them in a wide variety of sequences. From this seemingly randomn series of events, sophisticated algorithms can mine the data and start to “learn” patterns.
This is why these programs are dubbed “machine learning.” Those programs then start to be able to look at a player and say, “based on what other players have done, there’s an 80% chance of this player converting or quitting or clicking an ad this week.”
And bam, you have predictive analytics!
In future posts, I’ll spell out some of the implications around percentage accuracy and looking into that black box. But for now you know that it’s not science fiction, and that the techniques exists today.