Data Science

Sustainable Gamification: Playing the Game for the Long Haul

Michael Wu, Ph.D. is Lithium's Principal Scientist of Analytics, digging into the complex dynamics of social interaction and group behavior in online communities and social networks.


Michael was voted a 2010 Influential Leader by CRM Magazine for his work on predictive social analytics and its application to Social CRM.He's a regular blogger on the Lithosphere's Building Community blog and previously wrote in the Analytic Science blog. You can follow him on Twitter or Google+.



Little announcement before I begin today. I've been invited again to teaching the Rotman Executive CRM program at the University of Toronto, Nov 9--11. Since this is the third time I teach this course with Paul Greenberg, Ray Wang, and other CRM luminaries, I’ve decided to create a brand new lecture on gamification and what it means to Social CRM. It should be fun. If you are around Toronto that time, it would be great to meet if we can beat the scheduling gods.


This is the sequel to my previous post: The Gamification Backlash + Two Long Term Business Strategies. If you missed that, I recommend taking a quick read before moving forward.


Last time, we talked about the overjustification effect (or the moral hazard of game play). It is one of the biggest challenges for any gamification strategy. As Prof. Jesse Schell said, many studies have shown that “if you bribe someone to do something, they always come to hate that thing.” So the use of external incentives (e.g. points, badges, perks, money, etc.) will decrease a person’s intrinsic motivation and ultimately lead to the resentment of the gamified behavior (i.e. gamification backlash).


So gamification can’t possibly work over the long term. However, it doesn’t need to work long term to bring sustainable value. Last time, I described two strategies that can make gamification more sustainable:

  1. Use gamification to jumpstart activities that have either intrinsic or long term values for the player
  2. Use gamification to jumpstart a series of activities and enables the discovery and inference of the player’s intrinsic motivation, and then use extrinsic rewards to reinforce the player’s intrinsic motivation

Today, I will illustrate how these strategies work with some examples.


Strategy 1: The Value Mechanism

This strategy is not too difficult, because it can be implemented using extrinsic rewards. The key requirement is that the gamified behavior must have long term value to the players; otherwise, success is unlikely. I will cite three examples using this strategy, one in non-commercial setting and two in business.


In all three examples, players usually start the gamified behavior due to some aspects of gamification. Some people adopt the behavior because their friends are doing it (i.e. social facilitation, epic meaning, etc.). Others do it because they want to win and be on the leader board (i.e. social competition, badges, achievements, leader boards, etc.). Still, others do it simply because it is fun and gives them a sense of mastery (i.e. points, feedback, progression, level up, etc.). However, once the players realize the long term value, they no longer depend on gamification for motivations to perform the behavior. At that point, the value becomes the primary motivator. And gamification becomes a secondary reinforcement system that strengthens the correlation between value (effect) and behavior (cause).


I’d like to emphasize that the creation and realization of value usually don’t happen immediately, but that is the point of this strategy. Gamification will keep you motivated extrinsically until you realize the values from the gamified behavior. So gamification doesn’t need to work long term, it just has to work long enough for the players to realize the value. Subsequently, the reinforcement function of gamification can last indefinitely until the cause/effect cycle is interrupted. This conditioning effect of gamification may even help the players to learn to be intrinsically motivated by the gamified behavior over the long run.


Example 1: Education and learning

What is the gamified behavior(s)?

Good attendance, reading books, practice doing math, turning in homework/assignments on time, etc.

What long term values does it create?

  1. Students receive better grades
  2. They may get more peer respect, recognition and self esteem
  3. They may even gain more affirmation and trust from their parents, friends and teachers


Example 2: Nike+ sensor with Apple’s iPod/iPhone

What is the gamified behavior(s)?

Running and work out

What long term values does it create?

  1. Runners lose weight, look better, feel better, have a healthier life, and boost their self-confidence
  2. They see their own progression through history and see their mastery of running
  3. They may even enter flow sometimes if they join the challenges


Example 3: Lithium-powered communities

What is the gamified behavior(s)?

Community participations: answer questions, vote up good ideas, participate in conversations, etc.

What long term values does it create?

  1. Community members establish relationships and build tie strength, which can last a long time
  2. They co-create valuable knowledge that serves the entire community, the company and beyond
  3. They may even self-actualize by generating great ideas that gets implemented into products that can benefit themselves and future consumers.


Strategy 2: The Motivation Mechanism

This strategy is much harder because figuring out the intrinsic motivation of players is not easy. It requires an environment that allows the players to discover what motivates them intrinsically. Moreover, the gamification system has to infer a player’s intrinsic motivation through his activity data, and then focus on reinforcing his motivation with extrinsic rewards while the user is still engaged.


Returning to the examples above, we can see that this strategy won’t work for the Nike+ example. Because Nike only gamifies a single behavior (e.g. running), there is no room for the players to explore and discover their intrinsic motivation. For players that are already motivated to run intrinsically, it works wonderfully, but if the players are not intrinsically motivated by running, there is not much that they can do. That is probably why they didn’t try to use this strategy.


In the two remaining examples (e.g. education/learning and community participation), they gamified a large number of desirable behaviors in schools and in communities respectively. In these examples, players have the autonomy to explore among the gamified behaviors and potentially discover what motivates them intrinsically. However, it is very hard to track all the gamified activities for every single student in a school. Even if a school can pull it off, it may not have the analytics capacity to infer the intrinsic motivation of the students accurately. At best, they can give students the autonomy to self-select for their intrinsic motivation. Teachers would have to pay close attention to recognize the students’ distinctive and exceptional behaviors. This is like Montessori education, where the students always have a choice of activity from within a prescribed range of options.


Although this strategy can work in school, it is not very efficient. There is also a potential for rewarding the wrong behavior, and this circumstantial reward can turn into a Skinnerian feedback loop, which will ultimately lead to overjustification. In the example of a Lithium-powered community, however, the platform automatically tracks all actions by all members. The platform also has cross community activity data for all possible actions. This enables the gamification system to estimate the distribution of behaviors for each member (e.g. his social fingerprint) and the activity distribution for each type of behavior. Subsequently, the analytics engine can infer the most probably intrinsic motivators for all members and reward them accordingly. Provided that we can collect enough data fast enough to do the inference accurately, this strategy can be applied efficiently in an online community.



Although gamification doesn’t work over the long term, they can still bring sustained value via two routes:

  1. The value mechanism
  2. The motivation mechanism

We have seen three examples of these mechanisms at work. The value mechanism is definitely easier to implement, and most non-commercial usage of gamification tend to use this strategy (e.g. education, environment, government, health, social good, etc.).


Implementing the motivation mechanism is very challenging. It requires several ingredients that make this strategy technically feasible and practical only in an online community. Although I have yet to see this strategy use elsewhere, I believe it can be applied in mobile gamification, enterprise 2.0 systems that gamify work and collaboration, or any platform that offer a large selection of potentially motivating experiences. So despite overjustification, gamification can bring long term sustained values if done right. If done right, gamification is here to stay!




Thanks for the great post.  It connects to your presentation at the Like to Love roadshow...

Hi Mike,

First of all, I'd like to express how excited I am about how seriously gamification is being taken by people such as yourself. Getting people to do things by making the more fun is certainly not a new idea, but taking the scientific rigor of psychology and neuroscience and applying it across the board, combined with the power of data analytics that is becoming available is the first step on a very important road,

I would be extremely interested in hearing more about the second strategy, specifically on how you construct the model you use to analyze user actions to come to the model of intrinsic motivation. If the task were thrown at me I would probably do something along the lines of creating a model that learns (using some machine learning algorithm) what the user prefers to do, maybe based on which activity is repeated, or what the very first thing the user looks at is upon entering the page, maybe even some sentiment analysis on posts relating to a certain activity (these could be coupled by time or maybe posts directly from within the activity, for example posting directly to one's facebook wall from farmville).

Either way I would be making some massive assumptions. Let's take each into account.

Looking at repeated behaviors
If we take this to be a measure of intrinsic motivation we would assume, at least, that intrinsic motivation is more likely to lead to repeated activities than extrinsic motivation. Is there any research on this?

First action
There are many motivations for first action. By taking this to be a metric we assume that intrinsic motivation is at least as powerful, if not more, than the next most powerful motivation.

Sentiment analysis
I think the value of this would depend heavily on the quality of the analysis. The classic pitfalls apply of course, such as interpretation of sarcasm and humor, but more importantly: how will you be able to differentiate between an intrinsically motivated post and an extrinsically motivated post? "I'm so proud of myself for doing xy - I got z points!!" Regardless, we're assuming some kind of ratio across time of the influence of extrinsic vs intrinsic motivation on posts, which may not necessarily be the case.

I understand this is the gold at the end of the rainbow, and trade secrets must be kept. These are also simply the first few metrics that spring to mind; however I can't help but think it's not the variables I would track that would cause the model to be incorrect, but my approach.

So, what's your approach to this question?

Thank you!


Data Science



Hope to see you next time.

Data Science

Hello Zack,


Thank you for the comment and question.

Again, I’m on the road now. Actually teaching at U of Toronto now, so I apologize for the late reply.


I’m glad that you are excited about the rigor. Sometimes I feel that is not appreciated enough. So sincere thanks.


Your thinking is quite good. Those are the things that we look at initially, but then refined, iterate, and modified from analyzing tones of data. Although the precise algorithm is something that I can’t go too much into detail, but I will say a few thing.


Our approach is not to only look at the user’s interaction data, but we rather we should look at the distribution of all activity and interaction data across all users. This distribution is actually a power law distribution, and it turns out that only those that are at the tail of the activity distribution are intrinsically motivated to do that activity. Repeated behavior IS required, but not sufficient to infer intrinsic motivation.


Let me illustrate what I mean with an example. Suppose we want to find people who are intrinsically motivated to post messages in a community. Just because someone posted 10,000 post, it doesn’t mean he’s intrinsically motivated to do it. He may do it because he need to. Even if he is the top posting guy in the community that is still not  enough. He may do it because he want the attention from the community. Intrinsic motivation is something that people would do for its own sake, not for any other reasons.


Second, how susceptible the user’s behavior is to reward mechanism. Intrinsically motivating behavior are much more susceptible to positive reinforcement. This is actually a pretty important factor.


Last point is that first action actually is too noisy and not very predictive of their a user’s intrinsic motivation. Most of the first action is driven by the goal of initial engagement. But intrinsic motivation is something that people would do just for their own sake. Not the reason that drive them to engage initially.


Alright, I hope this helps a little bit. If anything is unclear, please let me know. I’m happy to discuss further.


Thanks again for the question. And I hope to see you next time.



Lithy Awards 2017

Voting is now closed. Winners will be announced on June 14th!!

See the nominees!