Goals
1.
Calories Burned per Day.
2.
More consistent nights rest (same # of hours from night to night).
3.
Increased # of moments where heart rate is sustained for at least 30 min.
Mission
We will be using new
features to improve the current process of tailoring effective messages to
users in order to sustainably engage as well as effectively improve their
health. We will be using the Hook Model (Trigger-Action-Reward-Investment), as
designed by Nir Eyal, to develop new short term and long term habits.
General User Profile
This general user is one
who cares about living a healthy life but does not come to fitbit with specific
goals of his/her own. They may care a lot about tracking their runs, or showing
off new technology, or sending social signals that they are a “healthy person”.
Regardless of their individual goals, they are there initially because of the
health halo surrounding the device. Therefore we have leveraged the default
bias to enroll them with our default goals since they are validated
strategies that are effective in promoting health and are applicable to most of
society. For those that want to explore goals and add new ones, they still have
that capability, these are merely what are provided at sign in in order to
promote the most amount of health for their population.
Added Features
We recommend integrating
Fitbit’s application with Facebook and Google Calendar. These allow us to
integrate new features into the current system that will greatly improve the
efficiency and accuracy of our targeted moments of engagement. The following is
a list of the algorithmic components that shape the nature and content of the
engagement.
Algorithm
1.
Goals: Depending on which of the
three default goals the user would prefer to pursue (as determined by analytics
of user engagement on the app), the Fitbit platform reinforces their
motivation.
2.
Events: (from facebook events and google calendar). We will use these
events as the primary structure of our intervention, thereby anchoring the
three default goals to meaningful events in their lives.
3.
Emotion: Sentiment analysis of their facebook posts allow the app to
estimate the user’s current emotional state (short term emotions), mood (longer
term emotions), and personality (considered permanent emotional states). These
factors are critical in determining the emotional framing of the
engagement messages. For example, if somebody is sad, it is not as effective to
use guilt to motivate them when compared to other possible forms of emotion
elicitation.
4.
User App Engagement: This monitors where the user spends their time when
they are using the app. Are they focused on their sleep or their calories
burned? These insights can shape the types of goals that are set by the
engagement messages and how frequently they should run.
5.
Current User Health-Related Patterns: These are determined by monitoring
the metrics used as proxies for the three default health goals we set.
Therefore the user’s current patterns in calorie burning, consistency of hours
slept per night and number of moments with sustained heart rate for 30 minutes
can all determine the nature of the engagement messages.
6.
GPS: This can be used to determine where the user might be and the form
of the message that is created to engage them. Therefore if a user is out at a
bar and has an exam the next day, the nature of the message can tailor it to
their situation to more accurately leverage the right tone and prompt.
7.
Survey: This information is collected by impromptu surveys of users who
failed to meet their goals that day/week/month. As a way to make of for lost
points that they would have gained by meeting their goal(s), users have the
option to fill out the survey in exchange for these points. These surveys can
reveal reasons why the user did not meet their goal(s) and therefore prepare
the next set of engagement messages accordingly.
Our Model (See Picture 1)
How the Components of the
Algorithm Inform the Hook Model (See Table 1)
Triggers: These will either be written or sensory messages sent to the
user through push notifications, texts, or the fitbit itself. For the written
triggers; their development will be determined by the various components of are
algorithm. The following are the components that make up a single trigger, all
of which are shaped by the algorithm.
1.
Content – The elements of the algorithm can be utilized to determine what the
actual content of the trigger message says.
2.
Framing (Cognitive) – This element of the trigger indicates whether the trigger
message emphasizes educational, behavioral, or otherwise cognitive salience.
3.
Framing (Emotional) – This element of the trigger indicates whether the trigger
message incorporates a certain emotional tone. Examples could be a guilt driven
message, one of humor, or any other appropriate tone for the moment.
4.
Channel – This is how the trigger message is delivered and through what medium.
5.
Time/Frequency – This is when and how frequently the trigger message is
delivered.
Actions: This
is the desired change in behavior of the individual. Therefore it is either an
increase in calories burned, an increase in consistency of hours slept,
sustained heart rate for 30 min.
Rewards:
These are what are given to the user in return for accomplishing their action.
It can take the form of points, badges or positive emotional feedback, like
“great job, you are getting more rest!”.
Investment: If
the user succeeds in their action (accomplishing their goal), then they can
invest by turning in their points for information or new app features. Another
way they may invest, assuming they succeeded, would be to share their
accomplishments with their friends. If the user did not accomplish the action,
then they are given the opportunity to achieve award points/badges/pos feedback
by filling out the survey explaining why they did not, and this form of
information sharing also acts as investment for the next round of Trigger-Action-Reward-Investment.
USER CASE:
In this experience we
follow the user, Bill, through a week involving 2 days of class and a quiz. Graph 1 shows how Bill would have slept during the week without the
Fitbit, while the second graph shows the points at which our intercession
would change his behavior for the better. Ordinarily, Bill sleeps inconsistently and averages less than 8 hours sleep per night. This is not behavior conducive to maximizing performance and if Bill follows Fitbit’s nudges, he will perform better in his classes and in the quiz at the end of the week than he otherwise would have done.
Table 1 shows the elements going into the algorithm to decide the content and
tone of the messages sent to Bill. He is relaxed about class but becomes more stressed as the quiz draws nearer. This affects the amount of sleep that he would ordinarily get – sacrificing sleep for more study time. The Fitbit encourages him to sleep more consistently throughout the week, using guilt as a motivator when he is more relaxed and the GPS places him away from home, late at night. As the test draws nearer and Bill gets more stressed, the tone of the messages changes from guilt inducing to encouraging and factual, suggesting to Bill that he focus on sleeping more and sending him links to articles that illustrate the positive effects that sleep has on academic performance.
The third and fourth
graphs are an illustration of Bill’s daily experience and of the nudges
provided by the Fitbit. The graphs are plotted with regard to movement,
which here serves as a proxy for wakefulness. The difference in message
timing and tone is a result of the algorithm output from Table 1. In each message cycle there is a Hookian trigger, action, reward and investment, the tone and type of which are decided by the algorithm. The goal of this is to keep Bill engaged with the app so that it can help him to best achieve his goals.
Graph 5 shows what
Bill’s sleep patterns might look like if he engaged with the app and used it
to help him prepare for his quiz. Sleep is more consistent from
night to night and Bill got his maximum amount of sleep, more than 8 hours, the night before the quiz. This is an improvement on Bill’s initially
predicted behavior and should result in him getting a better test score on the quiz than he otherwise would have done.
