Predictive analytics is a broad term that describes a variety of statistical and analytical models in order to predict future events. These predictive models generate a score for each variable with a higher score indicating a higher possibility of the given behavior happening. Today what interests most companies is what their customers are going to do, not what they’ve done in the past.
If you have a gaming company that owns and operates a game, you are probably aware of the fact that you have many different types of players that may play a particular game of yours. For example, you may have players that if they win too easily they will get bored with your game, you may have some users that if they fail more they will spend more on in-app purchases, and then you may have some users that will spend more money when they win more often.
In order to increase your game’s revenue, you may have to make decisions regarding your users’ playing experience, like to help them pass or fail a level, according to their past behavior in order to increase the amount or time or money they spend on your game, or in other words, their lifetime value. You may choose to help some players by providing in-app bonuses or gifts in order to help them pass a level, while giving others nothing.
For each player you need to find out the “next best action” for them. You can build a statistical model that will enable you to understand each of your users and then build a path for them that will maximize your game’s kpi.
Here’s an example:
For example, let’s say your users fall into the following six paths when they play your game:
1. They can spend up to 5 minutes on the level before they pass
2. They can spend 5 to 15 minutes on the level before they pass
3. They can spend 15 to 30 minutes on the level before they pass
4. They can spend 30 to 60 minutes on the level before they pass
5. They can spend 60 to 120 minutes on the level before they pass
6- They can spend more than 120 minutes on the level before they pass
You will need to build a regression model for each of the six categories that calculates for each user what will be his life time value (LTV) according to the category he falls into, which will then help you determine the “next best action” for that user. You will then give them a game score that will encourage him to take the action that will ultimately increase his LTV.
For example, let’s say you calculated that the best LTV for a certain user is between 15-30 minutes of playing. You will make the level more difficult for him in the first 15 minutes and then make it easier in the next 15 minutes or give him added bonuses so he can buy stuff that helps him complete the level during that playing session.
The way to build this model is create a user attributes table for each one of the users anytime he starts a new level. This attribute table contains both demographic and behavioral attributes like RFM (see below), as well as the LTV for that user and how much time it took him to pass the level according to the path categories above. Then for each path I run a regression model and that determines a score for each attribute.
What is RFM? It stands for:
Recency: How much time did it take for a user to pass a level? How much did it take to pass the level before that? What is the ratio between the two? What is the average amount of time spent in the last three levels?
Frequency: when a certain user passes a level, how much time does it take for them to move on to the next level? How many attempts are there on a particular level in a session before he drops off? What is the ratio between time spent during a session when he passes a level and when he doesn’t pass a level? What is the average amount of time it takes this user to return to a game during a session when he passes a level and one when he doesn’t?
Monetary Value: What is the amount of money a user spent on the most recent level? What is the ratio between the money spent on the last level and last 3 levels? What is the ratio between the amount of time he spends in the game and the amount of money he spends?
At this point you can use aggregation function via sql to build a panel of attributes and past history. Once you know what happened you can assign each of the attributes a score to that refers to the weight of the probability of each event happening again. In other words, you can use regression in order to look at a certain point in time in a user’s gaming history, find out the particular attributes of that user, look ahead of that particular point in time to see what happened and then make a prediction according to what has already happened.
Finally, You will look at today’s users and according to their attributes, calculate their expected LTV for each of the 6 paths. Take the highest possible score for this path and decide the best course of action for that particular player and determine whether or not they should pass a level.
CoolaData’s platform enables you to collect, transform and connect easily with statistical platforms like R and python in order to build predictive models. We also have a real time users’ table which can hold calculations of segmentations and scoring and give answers in less than 300 milliseconds.