The Fruit Basket Factor: Rational Churn Prevention

Posted by Dmitri Williams

Find me on:

Dec 4, 2013 8:08:00 AM

All players are not created equally, and frankly, not all are worth keeping. How can a developer decide which is which, and what to do about it?

There’s a very logical answer here that combines 1) the worth of the player with 2) the likelihood of them leaving. So first brief notes on those two things.

1) The worth of a player is their future impact on the system.

In other words, it’s their predicted future LTV (Lifetime Value). Looking at what they have done up until now is the next-best option, but it gives you no insight into how how their behavior will continue. For future LTV, we use a combination of their value plus their impact on others (Social Value) because we think it represents the best possible picture of total worth. If you don’t have these kinds of measures, you need some other way to estimate why player X is any different than player Y. If all you have is a proxy for value and you can at least separate the population into, say, high value vs. low value, that’s a lot better than assuming everyone is equal. But you really shouldn’t skimp on this. In a world with highly skewed spending patterns, it’s typical for only a small % of players to be worth anything, let alone a lot. High vs. low is just OK, and predictives are best.

2) The likelihood of a player leaving is also known as churn.

Here again, there’s a high-end solution using predictive analytics that tells you the % likelihood of any given player leaving in a specified time period. We typically use 3 weeks, so when we say someone’s churn probability is .76, what we’re really saying is “There’s a 76% chance that this person will not return within the next 3 weeks.” If you don’t have access to such tools, the next-best thing is to guess based on larger-scale patterns. Maybe you’ve done some analysis and found that players who reach level 8 and don’t come back are mostly not going to. That’s not great, but it’s still better than random.

OK, so you have some measure of worth and some measure of churn probability. Now what?

Let’s assume that you aren’t just looking at big aggregates of people. You’re interested in doing something for a specific set of users and you have the ability to act on just them. This may be with in-app messaging, email campaigns, push notification, etc. (and if you can’t do any of these things, you need to!).

Take user 1234, who has a value of $40 and a churn probability of 50%. What kind of effort should you expend to keep this person? The logical answer is that you should spend just up to how much value is at risk. Another term for this is “expected value.”

In this case, the expected value is the product of the worth and the probability, so $40 x .50 = $20. So, user 1234 is logically $20 at risk. How? Let’s say there are 10 users just like 1234, and each of them is worth $40 and is a 50% probability to churn. Of those 10 users, 5 are definitely going to churn and 5 are not, but of course you don’t know which. So, in actuality, each of those churned accounts is $40 lost and each that stays is $40 that wasn’t. But since you don’t know which is which, the only way you can take action is to assume they are all average. You’re going to be right half the time.

Now let’s say you spent $15 on the retention effort and it succeeded. For those who were going to leave, you spent $15 and they go on to spend $40, meaning you netted $25 for each of them, or $125 total. For those who were going to stay, you spent $15 you didn’t need to and they still spent $40, meaning you wasted $75. Overall, you came out $50 ahead. Make that $15 effort smaller, like $10 or $5 or $1 and you’ll come out better still.

In practice of course you’re unlikely to spend $15 on anyone. More likely you are sending an email that’s essentially free, maybe with a code for a free item. How chintzy should you be? You now have a rational way of deciding when it is or isn’t worth expending resources, exactly how much, and (if your retention effort works) you can be confident that you are going to come out ahead.

Social Whale - A Super-Valuable Player

Lastly, let’s consider the case of a super-valuable player. Maybe this is a big whale, or a social whale, whose departure would cost you say $500. The metrics tell you this person is 70% likely to leave. So, your math says this person is an at-risk value of $350. This means that logically, you can spend up to $349 to retain this person. Are you going to do that? Probably not, but if you’ve identified this person you understand exactly how much effort they are worth. And, at some point you’re going to find players who are worth a call from the lead developer or community manager, or some great incentive, or a special place in the community--or, yes, a fruit basket sent to their door.

Topics: Churn Analysis, Game Analytics, Video Game Analytics, Churn Prevention

    

Subscribe to Email Updates

Ninja Metrics Katana Demo

Ready for a live Demo of Katana?

If you have a social game or application, we can help you learn who matters most in your network and what features are affecting the health of your community