A Model for Designing Personalized and Context-Aware Nudges
Randi Karlsen and Anders Andersen
Department of Computer Science, UiT The Arctic University of Norway, Tromso, Norway
Keywords:
Smart Nudging, Digital Nudging, Nudge Design Model, Personalization and Context-Awareness,
Adaptive Nudge Design, Behavioral Change.
Abstract:
Nudging is a popular approach used to influence people to change their behavior towards a desirable goal.
To be effective, nudges should be tailored to the user’s specific needs based on user profile, current behavior,
and context information. In this paper, we target the questions of what to nudge for and when to nudge, and
how nudges can be automatically designed at a time when the user needs nudging. We present a model for
personalized and context-aware nudge design that is adaptive in that it continuously tailors the set of relevant
activities and the time for creating a nudge to the user’s needs for behavioral change. The model follows
a just-in-time approach, where nudges are created at the time when nudging is needed, based on the user’s
current situation.
1 INTRODUCTION
Nudging is a technique used to influence people to
change behavior towards a desired outcome, e.g.,
to adopt healthier or more environmentally friendly
habits. Behavioral change is the process of adopt-
ing new habits or modifying existing ones to achieve
a specific goal, and nudging is an effective tool for
promoting behavioral change involving psychologi-
cal factors that influence human behavior without re-
stricting their freedom of choice (Thaler and Sunstein,
2008).
Our focus is on personalized and context-aware
digital nudges, called smart nudges, where digital
technology is used to influence or guide people’s be-
havior (Karlsen and Andersen, 2019). Personalized
nudging refers to the use of tailored nudges based on
the user’s specific characteristics, activity history, and
preferences, while context awareness uses people’s
situation and environment to identify opportunities,
limitations, and obstacles, to further tailor nudging to
the needs of the user.
To facilitate behavioral change, nudging will chal-
lenge the user by motivating for activities that brings
the user closer to the nudging goal. To do so, a
nudging system will continuously keep track of the
user’s activities, preferences, achievements, and envi-
ronment, to design relevant and timely nudges that are
presented to the user, e.g., on a mobile device.
The idea behind tailored nudging is that people
are more likely to respond positively to nudges that
are relevant and meaningful to them, rather than us-
ing generic or “one-size-fits-all” nudges (Schneider
et al., 2018; Peer et al., 2020; Mills, 2022). How-
ever, a challenge when designing tailored nudges is to
determine what to nudge for and when to nudge.
This paper presents a practical model for how to
design smart nudges based on the user’s current be-
havior and situation. We focus on what to nudge for
and when to nudge, and suggest a general approach
to select activity and time frame for a nudge. We ex-
emplify how the proposed solution can be used in two
different use cases; for physical activity nudging and
green transportation nudging.
In the following, we first present smart nudges,
nudge design, and the two use cases. We describe
several factors that influence nudge design, including
level of behavior, behavioral progress, activity pat-
terns, reactions to previous nudges, capability, and
opportunity. This is followed by a description on how
activity and time frame for nudges are selected. Fi-
nally, we discuss our approach and conclude.
2 BACKGROUND
This section describes smart nudges, principles for
smart nudge design, and two use cases that will ex-
emplify how our generic model for nudge design can
be applied to different nudging systems.
Karlsen, R. and Andersen, A.
A Model for Designing Personalized and Context-Aware Nudges.
DOI: 10.5220/0012882800003838
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 16th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2024) - Volume 3: KMIS, pages 151-162
ISBN: 978-989-758-716-0; ISSN: 2184-3228
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
151

2.1 Smart Nudges and Nudge Design
A smart nudge is a personalized and context-aware
digital nudge, described through a set of components,
including an activity selected for the nudge, a time
frame for when the activity is suggested done, some
influence that motivates the user to do the activity, and
some practical information useful when following the
nudge.
Every component of a smart nudge can be tailored
to the user’s need for nudging. A user profile (includ-
ing, e.g., user interests, capabilities, activity history,
and reactions to previous nudges) and context infor-
mation (describing the user’s environment and sur-
roundings) are used to determine which activity and
practical information to include, which time frame to
target, and how to influence the user.
Nudges are typically offered to the user through an
application on a mobile phone, and a combination of
back-end and edge computing determines when and
what to nudge for through monitoring, selecting, and
analysing data from various sources, e.g., IoT sensors,
third-party data sources, mobile phone sensors, and
stored user data (Andersen et al., 2018).
Designing a nudge involves several tasks, as illus-
trated in Figure 1. In this paper, we focus on the de-
tails concerning selecting activity and time frame for
nudges. In paper (Karlsen and Andersen, 2024) we
describe the task of triggering smart nudges, paper
(Dalecke and Karlsen, 2020) focus on the influence
part of nudge design, while we are currently work-
ing on details concerning content selection and nudge
presentation.
The design tasks are influenced by a set of nudge
design principles, presented in Table 1 and previously
described in (Karlsen and Andersen, 2022), that guide
the selection of activity and time frame.
2.2 Use Cases
2.2.1 Physical Activity Nudging
Nudging people to be sufficiently physically active
is a desirable goal since regular physical activity
is a known protective factor for the prevention and
management of physical and mental diseases (WHO,
2016).
A nudge will present the user with a suggestion for
an activity and a time frame for doing it. For the user
to improve, nudging must suggest behavior that goes
beyond the user’s current behavior, e.g., by nudging
the person to (i) be active more often, (ii) be active
for longer periods, or (iii) engage in more challenging
activities.
Table 1: Principles for smart nudge design.
Challenge Nudging must challenge the user to
choose activities that improve the user’s
behavior.
Consolidate After improving behavior, the user must
be stabilized on the new level of behavior.
For a period the user is not challenged.
Progress Challenging the user will continue after
a consolidate period. Given there is still
room for improvement.
Variation The user should over time be given a vari-
ety of nudges, to make nudging interesting
and introduce new activities.
Timeliness A nudge should be given at a time when
the user can react to it and when the nudge
can be effective.
Safety and
feasibility
A nudge must avoid activities that are dan-
gerous or impossible for the user to do.
We distinguish between generic and specific
nudges. A generic nudge suggests an activity (such as
walking or cycling), and leaves it to the user to decide
how and where the activity is performed. The nudge
can optionally suggest a time frame, distance, or du-
ration of the activity. An example nudge is: “It’s time
for a walk. The weather is nice and you have time this
afternoon.”.
A specific nudge suggests a trail, that has a loca-
tion, route, and destination, and is characterized by
properties such as activity type, distance, estimated
duration, and difficulty. For example, a nudge can
suggest a hike to Mount X, which has a 5.8 km long
trail, elevation gain of 553 m, an estimated duration
of 2.5 hours, classified as hard, and has a spectac-
ular view from the top. Description and properties
of such trails can be found on online sources such as
AllTrails.com
1
, and can be used for creating specific
nudges.
2.2.2 Green Transportation Nudging
The goal of green transportation nudging is to mo-
tivate the user to choose environmentally friendly
transportation means. This is desirable because of
the urban challenges of increased traffic, conges-
tion, air pollution, and global-scale issues of climate
change and global warming (Comission of the Euro-
pean Communities, 2007).
When the user needs to move between two lo-
cations, origin (O) and destination (D), nudging can
motivate the user to choose alternatives to, e.g., pri-
vate car usage. Relevant alternatives depend on, e.g.,
the availability of public transportation, whether the
1
https://www.alltrails.com
KMIS 2024 - 16th International Conference on Knowledge Management and Information Systems
152

Select activity
and time
Select
influence
Trigger
nudge design
Select
content
Present
nudge
Figure 1: Tasks in a nudge design process.
distance between O and D allows walking or cycling,
and the user’s ability to use the different transporta-
tion means.
Alternatives to using the private car include walk-
ing, cycling, public transportation, and the use of mo-
torbikes and electric scooters. An alternative can also
be to optimize and limit private car usage through car-
pooling, incentive parking (park and ride), and finding
the most efficient driving route to the destination.
Available transportation means are ranked accord-
ing to how environmentally friendly they are, and a
nudge will suggest that the user selects a transporta-
tion alternative that is better than what the user nor-
mally prefers between O and D.
2.2.3 Related Work
The goal of nudging is to motivate people to change
their behavior in some direction. As motivation is an
important feature of nudging, much focus has been
on the influence component, as seen in, e.g., (Cara-
ban et al., 2019; Forberger et al., 2019; Caraban
et al., 2020; Villalobos-Z
´
u
˜
niga and Cherubini, 2020;
Bergram et al., 2022). However, it is equally impor-
tant to focus on other components of a nudge. In this
paper, we target the questions of what to nudge for
and when to nudge, and how timely nudges can be
automatically designed. A just-in-time nudge design
approach is necessary since we tailor nudges to the
user’s behavior, situation, and environment, which is
known only at the time of nudging.
Different authors have described digital nudge de-
sign through theoretical models or frameworks in-
tended for designers or practitioners (Meske and Pot-
thoff, 2017; Mirsch et al., 2018; Schneider et al.,
2018; Purohit and Holzer, 2019). In contrast, we
describe a model that targets automatic and adaptive
nudge design in that it continuously tailors nudges to
the user’s current needs for behavioral change.
The main focus for nudging has up until recently
been on one-size-fit-all nudges (Anagnostopoulou
et al., 2020; Peer et al., 2020). However, the im-
portance of personalization in designing effective
nudges has been recognized by several authors, e.g.,
(Sch
¨
oning et al., 2019; Peer et al., 2020; Mills, 2022),
and experiments (Peer et al., 2020; Mills, 2022) show
that personalized nudges can lead to more effective
nudging compared to non-personalized nudges. Per-
sonalized nudging is a topic in need of more re-
search (Caraban et al., 2019; Jesse and Jannach, 2021;
Bergram et al., 2022), and we target this important
topic through our model for nudge design.
Our approach to personalized and context-aware
nudging is different from other approaches we are
aware of, in that nudges are automatically designed
and adaptively tailored to the current need of the user,
based on continuous monitoring of user behavior and
environment. The user is gently challenged to im-
prove, based on an adaptive goal, and the nudging
system will continuously distinguish between impos-
sible and possible activities for the user.
3 FACTORS INFLUENCING
NUDGE DESIGN
Smart nudging requires knowledge about the user’s
current behavior and behavioral change. This is
obtained by monitoring user activities relevant to
the nudging goal. This section describes how the
user’s Level of Behavior is measured, how Behavioral
Progress is determined, how Activity Patterns are use-
ful for nudge design, and how reactions to previous
nudges influence nudging. Nudge design is also in-
fluenced by user capability (i.e., the ability to do an
activity) and opportunity (i.e., how context can make
an activity possible or not). Table 2 summarises these
factors and describes how they influence smart nudge
design.
3.1 Level of Behavior and Behavioral
Progress
This section presents a general description of Level
of Behavior and Behavioral Progress, followed by ex-
amples of how the concepts are used in the two use
cases.
3.1.1 General Description
Level of Behavior (LoB) refers to how well a user
behaves with respect to the nudging goal, e.g., how
physically active or environmentally friendly the user
is. LoB is measured using an ordinal scale, e.g., Very
Low, Low, Medium, Good, Excellent.
The user’s current LoB is determined by monitor-
ing and assessing all activities relevant to the nudg-
A Model for Designing Personalized and Context-Aware Nudges
153

Table 2: Factors influencing the selection of activity and time frame during nudge design.
Factors Description Effects on nudge design
Level of
Behavior
(LoB)
A measure of how well the user behaves
regarding the nudging goal.
LoB is used as a basis for setting an ActivityGoal, determin-
ing behavioral progress, and keeping track of how user behavior
changes over time.
Behavioral
progress
As the user’s behavior change, the
progress is described as improved, stable,
or decreased.
Determines if the nudging system should challenge the user to
improve or help the user to stay on the current activity level.
Activity
pattern
Describes recurring activities that happen
in a systematic way.
Can detect activities the user is predominantly doing, when and
under which circumstances activities are normally done. Identi-
fies preferred activities and time frames, and can guide how to
improve behavior.
Reactions
to nudges
Monitor reactions to nudges and register
which nudges the user accepts and rejects.
Knowing which activities, time frames, and influence types the
user responded positively to, contributes when selecting compo-
nents for new nudges.
User
capability
The individual’s physical and psychologi-
cal capacity to engage in an activity.
Distinguishes between activities the user is unable to do (i.e., im-
possible activities) and activities the user is capable of doing.
Opportunity The factors that lie outside the individual
that make an activity possible or not
An activity can be impossible because of permanent/long-lasting
circumstances (e.g., lack of equipment) or temporary circum-
stances (e.g., challenging weather conditions).
ing goal. As behavior changes over time, LoB is de-
termined for discrete time periods (e.g., per week or
month), and LoB for the latest period determines if
the user needs to improve or if the best LoB is reached
and the user should be nudged only to continue on the
same behavioral level.
User activity is in our use cases measured using
a numerical value. Each activity is given a value
that reflects how well it supports the nudging goal
and UserActivity(P) represents the total value of all
the user’s activities in period P. Level of Behavior
for P, i.e., LoB(P), is determined by mapping the
UserActivity(P) value to the LoB scale.
To follow the principle of challeng the user, the
nudging system automatically sets and adjusts a goal
(denoted ActivityGoal) for the next time period P. If
the goal is met for period P
i
, ActivityGoal is slightly
increased for the next period P
i+1
.
Behavioral Progress guides the “aggressiveness”
of the system, by determining if the system should
challenge the user to improve or help the user to stay
on the current activity level. Three states describe
behavioral progress, i.e., improved, stable, and de-
creased, and enables the system to design nudges so
that the principles of consolidate and progress are fol-
lowed.
When the user reaches a higher LoB (e.g., when
going from Medium to Good on the LoB scale), the
progress state, PrState, is set to improved and the
system suspends, for a period, the automatic adjust-
ment of the goal, as nudging will focus on consol-
idating the user on the newly reached level. After
a while, the user enters a stable stage and the sys-
tem can again challenge the user by gently increasing
the ActivityGoal value. If user behavior declines, the
state is set to decreased, the value of ActivityGoal re-
mains unchanged, and the user will be challenged in
order to improve.
Formula 1 describes how the ActivityGoal value
is automatically adjusted with an improvement factor
Impr. The ActivityGoal is increased only if the user
has reached the activity goal the last time period, and
the user is not in an improved state.
I f UserActivity(P) ⩾ ActivityGoal
AND PrState ̸= “improved”
then ActivityGoal = ActivityGoal + Impr
(1)
To keep track of how user behavior changes over
time, historic values of Level of Behavior are stored
in a History of Behavior. The history can, e.g., show
whether user behavior improves over time or if the
time of year has any impact on behavior.
3.1.2 Physical Activity Nudging
Physical activity can be measured using step count
and , UserActivity(P) represents the total number of
steps for all activities performed during P. For activ-
ities such as walking or running, a pedometer is used
for simply counting steps. Other types of activities
(e.g., swimming or climbing) can be manually regis-
tered by the user and converted to steps using a step
conversion factor. In Formula 2 the number of steps
are calculated for an activity a.
Steps(a) = StepConversionFactor(a) ∗ Duration(a) (2)
KMIS 2024 - 16th International Conference on Knowledge Management and Information Systems
154

The number of steps obtained from a pedometer
does not measure user effort or exercise intensity. Us-
ing number of steps per minute or complementing
the pedometer with a heart rate monitor are two ap-
proaches to get indications of effort or intensity. If
effort or intensity is available, number of steps can
be normalized, by multiplying with an effort factor
before being added to the UserActivity(P) value, as
described in Formula 3.
NormalizedSteps(a) = steps(a) ∗ e f f ort(a) (3)
The ActivityGoal is also given in number of steps,
and the user can be challenged by expecting more
steps the next period. To reach the new goal, the sys-
tem may nudge for longer walks, more challenging
exercises where user effort increases, or nudge more
often to increase the frequency of physical activity.
Based on health organizations’ recommen-
dations for physical activity, mapping between
UserActivity(P) value and the LoB scale can be
adapted according to, e.g., age and capabilities of the
user.
3.1.3 Green Transportation Nudging
Determining Level of Behavior regarding green trans-
portation includes monitoring or registering the type
of transportation chosen in every travel situation.
A travel is described through a set of properties,
including (O, D, Time, TMeans), where O and D rep-
resent origin and destination, Time represents depar-
ture time, and TMeans the transportation means used
for the travel. Each type of transportation means is
characterized by its environmental friendliness (EF)
using an EF-value in the range [0,1], where the
higher the value, the more environmentally friendly
the transportation means is.
UserActivity(P) represents the average EF-value
over every transportation means used in period P.
This is expressed in Formula 4, using the set
Travel(P) that includes all EF-values of travels made
by the user in P.
UserActivity(P) =
∑
x∈Travel(P)
x
|Travel(P)|
(4)
Travel(P) = {x|x is the EF-value of a travel in P}
Both UserActivity(P) and ActivityGoal represent
EF-values in the range [0,1]. To help the user reach
the goal, nudging can suggest more environmentally
friendly transportation means (with a better EF-value)
compared to what the user has previously chosen.
The user may have different habits when traveling
between different locations, e.g., the user may prefer
to walk to work, while using the car when traveling
with kids or going shopping. Knowing the user’s LoB
for recurring (O,D) pairs can be helpful for identify-
ing where improvements can be suggested.
Consequently, user activity is determined per
(O,D) pair (i.e., UserActivity
(O,D)
(P)) using Formula
5, where Travel
(O,D)
(P) ⊆ Travel(P). This makes it
possible to register LoB per (O,D) pair and have an
(O,D) specific goal, i.e., ActivityGoal
(O,D)
.
UserActivity
(O,D)
(P) =
∑
x∈Travel
(O,D)
(P)
x
|Travel
(O,D)
(P)|
(5)
Travel
(O,D)
(P) = {x|x is EF-value of a travel be-
tween O and D in P}
Generally, taking the bus or a car gives a lower
EF-value compared to walking and cycling. How-
ever, if the distance is too long and walking/cycling
is not practical or possible, the bus (or other public
transportation means) can be considered the greenest
choice. Also, the time of day may determine what
to expect from the user and what the nudging system
can recommend. Walking late at night may not be
safe, and therefore not an option. Taking the bus, or
perhaps even a taxi, will in this situation be consid-
ered the greenest alternative. In each situation, the
greenest alternative is given the best EF-value.
3.2 Activity Patterns
This section presents a general description of activ-
ity patterns used in nudging, followed by examples of
how activity patterns are detected in the two use cases.
3.2.1 General Description
Activity Patterns represent information on user activ-
ities detected through monitoring and/or manual reg-
istration of activities. Generally, an activity pattern
describes recurring activities that happen in a system-
atic and predictable way and, with respect to nudging,
targets activities that are relevant to the nudging goal.
As nudging should support the user to improve
behavior, we expect activity patterns to change over
time. Therefore, activity patterns are detected for
different time periods (e.g., weekly or monthly pat-
terns), where differences between patterns can un-
cover changes in behavior.
An activity pattern identifies activities the user
is predominantly doing, characteristics of these ac-
tivities, when and under which circumstances (e.g.,
weather conditions) activities are normally done.
Knowing the user’s current activity pattern and level
of behavior is the key to knowing how to nudge the
user. The activity pattern is used as baseline when se-
lecting activities for a nudge while avoiding activities
A Model for Designing Personalized and Context-Aware Nudges
155

or circumstances that represent a radical change for
the user.
User preferences can to some extent be favored
in nudge design, given that it does not conflict with
the goal of improving behavior. The user may, for
example, be inclined to accept a nudge suggesting a
familiar activity or an activity with properties that do
not deviate too much from the user’s current behavior.
3.2.2 Physical Activity Nudging
Activity patterns provide information on which activ-
ities the user prefers, habits concerning when to ex-
ercise, and the duration and/or distance of previous
activities. When designing new nudges, this is used
as a basis to determine which activities to nudge for
and when to nudge.
To identify activity patterns, we focus on activities
(or exercises) that last more than a certain amount of
time (e.g., 15 minutes). This is different from deter-
mining Level of Behavior, where we include all steps
during a time period.
Exercises can be identified by analyzing i) step
count data, where an exercise is identified as contin-
uous activity (i.e., steps) exceeding a certain amount
of time, and ii) manually registered exercises.
To manually register exercises, many tracking de-
vices (such as FitBit
2
, Oura ring
3
, and Apple Watch
4
)
allow users to select an activity type (such as walking,
cycling, swimming, or climbing) and register start
time and duration. An activity tracker can also sup-
port the user by automatically detecting certain types
of exercises.
The two complementary approaches; step count
analysis and analysis of manually registered exer-
cises, may both find exercises not detected by the
other. For example, step count analysis will not de-
tect swimming and climbing, but may detect (walk-
ing) exercises overseen or forgotten by the user, and
therefore not manually registered. Also, some exer-
cise types (such as walking and running) may be de-
tected by both approaches. As part of a data cleaning
process, duplicate exercises are identified and merged
by comparing the time property.
An exercise is described through a set of prop-
erties, including activity type, number of steps, dis-
tance, duration, elevation, and time of day/week
when the exercise took place. Activity patterns can
provide frequency of activity types (to identify fa-
vorites or activities the use never do), at which time
the user tends to be active (indicating when it may
2
https://fitbit.com
3
https://ouraring.com
4
https://www.apple.com/watch/
be useful to nudge), and relations between activity
and time (e.g., which activities are preferred at cer-
tain times of week/year).
For each type of activity, distance, duration, and
elevation can be described using statistical values
such as average, distribution, minimum, and max-
imum, where minimum and maximum identify the
currently easiest and most challenging experience
(e.g., shortest and longest distance), while average
and distribution are used for identifying the user’s
normal activity.
Activity patterns can be combined with context
data (e.g., weather conditions or level of pollution)
to detect which activities are predominantly done at a
given condition.
3.2.3 Green Transportation Nudging
Activity patterns provide information on recurring
travels between (O,D) pairs, including preferred
transportation means and departure times. Recurring
travels represent traveling habits, such as going to
work at approximately the same time every weekday,
picking up kids at school, or going to the gym at cer-
tain times during the week. As users may have differ-
ent habits when traveling between different locations,
an activity pattern for each recurring (O,D) pair (de-
noted ActivityPattern
(O,D)
), is identified.
The user’s movements, from one location to an-
other, departure, and traveling time can be monitored
using some activity tracking devices. Also, trans-
portation means can to some extent be inferred based
on monitored data. However, to obtain accurate travel
information, the user must manually register the se-
lected transportation means and departure time.
The activity patterns are used as a basis for se-
lecting transportation means and time frame when
designing new nudges, where the user can be chal-
lenged to use more environmentally friendly trans-
portation means compared to the preferred choice,
while patterns concerning departure time indicate
when a nudge can be useful.
3.3 Reactions to Nudges
By registering reactions to nudges, the system can
learn which nudges were accepted or rejected by the
user. Accepted nudges reveal which nudges the user
found useful, while rejected nudges reveal what the
user did not find tempting or useful, where, e.g., the
timing was wrong, or the activity too challenging or
not interesting. Since there are many reasons for re-
jecting a nudge, it is useful with a response from the
user on what caused the rejection.
KMIS 2024 - 16th International Conference on Knowledge Management and Information Systems
156

For both physical activity and green transportation
nudging the set of accepted nudges may over time re-
veal patterns of which activities tempted the user, and
which influence types the user reacted positively to.
The set of rejected nudges may, on the other hand,
identify activities the user is not interested in, and
possibly also influence types that do not motivate the
user. An activity that is repeatedly rejected should be
suspended from nudging for a period.
3.4 Capability and Opportunity
The work of (Michie et al., 2011) describes a behav-
ior system, where capability, opportunity, and motiva-
tion interact to generate behavior. Capability and op-
portunity are factors that influence motivation, while
behavior is influenced by all three components. Mo-
tivation is in our work represented by the influence
component in a nudge, while capability and opportu-
nity impact the selection of activity and time frame
for a nudge.
3.4.1 Capability
Capability is defined as the individual’s physical and
psychological capacity to engage in a suggested activ-
ity, and includes having the necessary knowledge and
skills to do the activity (Michie et al., 2011).
In a nudging system, the user can register per-
sonal capabilities and/or capability constraints, such
as physical disabilities and lack of skills. The user’s
status regarding capability makes certain activities
permanently or temporarily impossible. Activities
can become possible if the capability status change,
e.g., if the user recovers from a temporary disability
or obtains the necessary skills. User capability con-
tributes to distinguishing between activities that are
impossible for the user, and activities that can be in-
cluded in a nudge.
3.4.2 Opportunity
Opportunity is defined as all the factors that lie out-
side the individual that make a behavior possible or
not (Michie et al., 2011). This includes physical cir-
cumstances (such as access to resources or facilities),
environmental and social factors that support or hin-
der activities. Opportunity covers a large number of
different circumstances that affect the user’s ability to
perform an activity, and therefore affect the choice of
activity when designing a nudge.
We identify an obstacle as a situation which
makes it impossible to go through with an activ-
ity. The reason can, e.g., be the lack of equipment
(such as bicycle or skis), lack of facilities or services
(such as safe outdoor spaces or public transportation),
or harmful or hazardous situations (such as severe
weather conditions) that make activities potentially
dangerous.
An obstacle can be long-lasting or temporary. A
long-lasting obstacle is removed only if some funda-
mental change is made, e.g., if new equipment or fa-
cilities are made available, or if the user change loca-
tion where circumstances are different. A temporary
obstacle ends as soon as the harmful situation is over.
Long-lasting obstacles can be manually registered by
the user, while temporary obstacles are detected based
on monitoring the user’s surroundings.
4 SELECTING ACTIVITY AND
TIME FRAME FOR THE NEXT
NUDGE
This section describes factors relevant when selecting
a suitable activity and time frame during nudge de-
sign. In the following, we describe i) how the need for
nudging is detected, ii) how the set of possible activi-
ties for a nudge is identified, and iii) how the selection
of activity and time frame can be done.
4.1 Identifying an Activity Gap
4.1.1 General Description
A nudge can be triggered by certain situations that in-
dicate the need for a gentle push towards the nudg-
ing goal. An important reason for nudging is that
the user’s current Level of Behavior is insufficient to
reach the ActivityGoal.
Formula 6 calculates ActivityGap(P), which rep-
resents a measure of the activity needed to reach the
goal for the current (unfinished) time period, P.
ActivityGap(P) = ActivityGoal − PredActivity(P) (6)
PredActivity(P) represents the amount of activ-
ity predicted to be done in P, including already com-
pleted activities and expected activities in P. An ex-
pected activity can be a recurring activity, detected
through the user’s activity pattern, or an activity the
user has committed doing in P.
The ActivityGap(P) can be calculated from the
first day of P, and as both completed and expected
activities will change during P, PredActivity(P) and
ActivityGap(P) must be regularly recalculated to re-
flect the user’s achievements during P.
Knowing the ActivityGap(P), makes it possible
to plan and design a set of nudges for P that collec-
tively helps the user to reach the activity goal. As the
A Model for Designing Personalized and Context-Aware Nudges
157

ActivityGap(P) value is updated during P, the set of
planned nudges must also be updated.
The generic Formula 6 applies to both use cases.
However, as described in the following Sections 4.1.2
and 4.1.3, ActivityGap(P) and PredActivity(P) are
calculated differently in the two cases.
4.1.2 Physical Activity Nudging
For physical activity nudging, the ActivityGap deter-
mines the additional number of steps needed for the
user to reach the ActivityGoal.
Predicted activities, PredActivity(P), is calculated
as seen in Formula 7, where CompActivity(P) and
ExpActivity(P) represent the number of steps for
completed and expected activities in P, respectively.
PredActivity(P) =
CompActivity(P) + ExpActivity(P)
(7)
The ActivityGap(P) is calculated using Formula
6, and nudging is needed if ActivityGap(P) > 0.
4.1.3 Green Transportation Nudging
For green transportation nudging, ActivityGoal,
ActivityGap, and all travels are measured using the
EF-value. Predicted activities, PredActivity(P), is
calculated as a combined average EF-value over all
completed and expected travels, see Formula 8. The
calculation is based on two sets of EF-values, repre-
senting the completed travels, CompTr(P), and the
expected travels, ExpTr(P).
PredActivity(P) =
∑
x∈CompTr(P)
x +
∑
y∈ExpTr(P)
y
|CompTr(P)| + |ExpTr(P)|
(8)
CompTr(P) = {x|x is the EF-value of a completed
travel in period P}
ExpTr(P) = {y|y is a predicted EF-value of an ex-
pected travel in period P}
The expected transportation means, and thus the
expected EF-value, for a travel between O and D is
based on the user’s previous behavior, such as the
preferred transportation means for (O,D) the last few
time periods.
Nudging is needed if ActivityGap(P) > 0, and,
at the end of P, the ActivityGoal is reached if
ActivityGap(P) ⩽ 0.
When using (O,D) specific activity goals, i.e.,
ActivityGoal
(O,D)
, activity gaps can be determined
per (O,D) pair (i.e., ActivityGap
(O,D)
(P)) using For-
mula 9, where PredActivity
(O,D)
(P) is calculated as
in Formula 8 but in this case using the two sets
CompTr
(O,D)
(P) and ExpTr
(O,D)
(P) which only in-
cludes completed and expected travels between (O,D)
in period P.
ActivityGap
(O,D)
(P) =
ActivityGoal
(O,D)
− PredActivity
(O,D)
(P)
(9)
4.2 Identifying Possible Activities
4.2.1 General Description
In (Karlsen and Andersen, 2022) we classify activi-
ties as impossible, possible, unlikely (to be accepted
by the user), and probable (that have the potential to
be accepted by the user). Table 3 describes each class
of activity, while Formula 10 shows the relation be-
tween the classes, where U represents a user and A all
available activities.
Table 3: Classification of activities.
Impossible Activities the user is not capable of doing.
Identified based on user capability and op-
portunity.
Possible Activities the user is capable of doing.
Possible(A,U)=Probable(A,U)
S
Unlikely(A,U)
Probable Activities the user has done in the past and
is likely to do in the future. Likely to be
accepted when suggested in a nudge.
Unlikely Activities the user has never done, chal-
lenging and/or repeatedly rejected activi-
ties. Identified based on the user’s activity
pattern and reactions to previous nudges.
Possible(A,U) = A − Impossible(A,U)
Probable(A,U ) = Possible(A,U ) −Unlikely(A,U)
(10)
The classification is user-specific, based on the
user’s activity pattern and reactions to previous
nudges, where a challenging or repeatedly rejected
activity can cause the activity to be classified as un-
likely. Also, capability and opportunity can make an
activity temporarily or permanently impossible.
4.2.2 Physical Activity Nudging
For the physical activity use case, activity patterns
provide frequency of activity types (such as walking,
cycling, and swimming), and identifies for each activ-
ity type, minimum, average, and maximum values for
properties such as distance, duration, and elevation.
Physical activities can be conditionally possible,
depending on properties of the activity. For example,
a user may have a disability that makes short walks
possible, while longer walks are impossible.
To distinguish between probable and unlikely ac-
tivities, we identify activities that are challenging for
the user. For each activity property, p, minimum
(min) and maximum (max) values are used to distin-
guish between normal and challenging activities for
user U, see Formula 11.
Normal(p,U) = (min
p
, max
p
− d]
Challenge(p,U ) = (max
p
− d, limit)
(11)
KMIS 2024 - 16th International Conference on Knowledge Management and Information Systems
158

Normal(p,U) is an interval reflecting the user’s
previous achievements with respect to property p. For
example, if p represents walking distance, max
p
and
min
p
represent the longest and shortest distance, re-
spectively, walked by user U . Challenge(p,U ) is an
interval for p which represents efforts that are possi-
ble for the user, but currently beyond the user’s max-
imum. An absolute limit for p can be registered by
the user, and if p of an activity exceeds the limit, the
activity is classified as impossible.
The d in (max
p
− d) represents a deviation from
the max
p
value and is used to adjust the upper bound-
ary for normal activities. For example, if max
p
is
far above the average value for p, max
p
may be too
high to count as normal activity. Deviation d may be
adapted to the user’s current activity pattern by, e.g.,
being calculated as a function of average.
An unlikely activity is an activity the user has
never done, has been repeatedly rejected when sug-
gested in a nudge, or has one or more properties that
are within the Challenge(p,U) interval. An activity
is impossible if capability or opportunity issues hin-
der it, or if at least one activity property is above the
limit. Otherwise, it is a probable activity.
4.2.3 Green Transportation Nudging
As the user may have different habits when travel-
ing between different (O,D) pairs, we identify prob-
able, unlikely, and impossible transportation means
per (O,D) pair. Walking may, e.g., be classified as
unlikely for user U when going shopping, while it is
possible when going to work.
Impossible transportation means are identified
based on user capability and opportunity, and
impossible(A
(O,D)
,U) can, e.g., include i) private car,
if the user does not have a driver’s license, ii) train, if
there are no train services for (O,D), or iii) cycling, if
the user does not have a bicycle.
Transportation means where the user is physically
active (e.g., walking or cycling), can be conditionally
possible. Similarly to the approach in Section 4.2.2,
an absolute limit for an activity property p can be reg-
istered by the user. If walking distance between O and
D is above the limit, walking is impossible for (O,D).
To determine what is physically challenging for a
user, a collective activity pattern covering all (O,D)
pairs is used to determine how frequent physical ac-
tivity is used as transportation means, and average and
maximum values for distance and/or duration proper-
ties. These values determine similar intervals to the
ones presented in Formula 11, and is used as a ba-
sis for classifying activities, as described in Section
4.2.2.
Unlikely(A
(O,D)
,U) includes possible transporta-
tion means that U has never used, repeatedly rejected
for (O,D), or, if it is a physical activity transportation
means, has a property p within Challenge(p,U).
4.3 Selecting Activity and Time Frame
4.3.1 General Description
To help the user reach the ActivityGoal, a nudge can
either suggest a new activity that contributes to filling
the ActivityGap(P) or suggest an improvement of an
expected activity (e.g., suggest a longer walk or im-
prove a predicted transportation means).
An unlikely activity may be selected to challenge
the user to do something new or more demanding,
while a probable activity, representing the familiar or
preferred, can be selected to make it more likely that
the user accepts the nudge.
Time frame and activity are closely connected, as
the time frame for a nudge determines which activi-
ties are possible to suggest. An obvious requirement
is that the selected activity must be possible to do
within the targeted time frame. For some nudges, the
time frame is first selected, and an activity that fits the
time frame is subsequently chosen. For other nudges,
the order is reversed, and activity is chosen before the
time frame.
To fill the ActivityGap(P), the activity must take
place in P, meaning that the time frame for the nudge
must be within P. If a nudge is given with a time
frame outside P and the nudge is accepted, the sug-
gested activity will later be included as an expected
activity in a future time period.
4.3.2 Physical Activity Nudging
The ActivityGap(P) represents the number of steps
that remains in period P to reach the ActivityGoal.
To fill the gap, the user needs to be more active, and
nudging will suggest additional activities to the user.
Different activities (e.g., frequent short walks or
a long hiking trip) can be equally useful as long as
the suggested activity is possible for the user to do.
That is, if a nudge suggests an activity with cer-
tain properties p, such as distance or duration, each
property must be within the user’s Normal(p,U) or
Challenge(p,U) interval.
What to nudge for also depends on the user’s sit-
uation (such as available time) and preferences, and
which nudges the user reacts positively to. There-
fore, when selecting an activity from either of the two
sets Probable(A,U) and Unlikely(A,U), these factors
must be considered. Additionally, the principle of
variation (described in Section 2.1) can be supported,
by, over time, nudging for a variety of activities.
A Model for Designing Personalized and Context-Aware Nudges
159

When selecting a time frame for a nudge, the time
since the last activity and time frame for committed
activities are important factors. Time frames should
not overlap, and there must be a sufficiently large in-
terval between the end of one activity and the time-
frame for the next.
4.3.3 Green Transportation Nudging
While physical activity nudging can fill an
ActivityGap(P) by adding more activities,
green transportation nudging can only reach the
ActivityGoal by improving the EF-value of activities,
i.e., selecting a transportation means that is more
environmentally friendly than what the user normally
prefers. The number of travels cannot be adjusted to
fill an ActivityGap(P). Travels are triggered only by
the user’s need to change location.
A location change is detected by i) monitoring
the user’s activity pattern to detect regular location
changes, ii) using calendar information to detect ap-
pointments that require location change, and iii) let-
ting the user register a location change.
The goal is to nudge the user to choose
transportation means so that, when period P is
over, ActivityGap(P) ⩽ 0 (i.e., the ActivityGoal is
reached). A nudge should suggest a transportation
means that brings a positive ActivityGap(P) value
closer to 0, or keeps ActivityGap(P) ⩽ 0.
The system must detect which (O,D) pair the user
can improve, and determine which improvement to
suggest. If user U has previously traveled (O,D), there
exists an expectation of which transportation means
the user will choose and a corresponding predicted
EF-value. To improve, a transportation means with
a better EF-value must be selected from either of the
sets Probable(A
(O,D)
,U) or Unlikely(A
(O,D)
,U).
Available transportation means are partially
ranked based on their EF-values. This means that
some transportation means (e.g. walking and cycling)
are equally environmentally friendly and can have the
same EF-value.
5 DISCUSSION
This section discusses some observations made dur-
ing the work on smart nudge design. It includes how
relevant activities can be determined ahead of nudge
design, to make the design process more efficient, and
how a system can create plans for nudging, either as
a set of individual nudges or a succession of linked
nudges. This section also shows how the smart nudge
design fulfills the design principles presented in Sec-
tion 2.1.
5.1 Predetermining Relevant Activities
To simplify the task of choosing an activity, a set of
only relevant activities (denoted RelActivities) can be
identified. To make nudging more efficient, the con-
tent of RelActivities can be determined in advance,
before nudge design time. Activities that are perma-
nently impossible for the user, are the most obvious
activities to be excluded from RelActivities.
A nudge must include an activity that represents
an improvement, or at least a status quo, with respect
to user behavior. In general, this means that all ac-
tivities that represent a decrease in behavior should
never be nudged for and are consequently excluded
from RelActivities.
For green transportation nudging, the partial rank-
ing of activities makes it possible to identify activi-
ties that represent a decline in user behavior and dis-
regard them as not relevant for nudging. For exam-
ple, if the user is predominantly walking to work, a
nudge can suggest walking or cycling, but will never
suggest public transportation or carpooling (since this
represents a decline in behavior).
For physical activity nudging, where every activ-
ity contributes to reaching the ActivityGoal, it may
be more difficult to discard activities as not relevant.
However, for a relatively active user, the system may
choose never to nudge for short or very easy activi-
ties, as these may not be considered a push towards
improved behavior.
5.2 Creating a Nudging Plan
The ActivityGap(P), described in Section 4.1, repre-
sents a prediction of how much the user must improve
to reach the ActivityGoal. When predicting a defi-
ciency in user behavior, the system can set up a plan
for nudging so that the user can reach the goal during
period P.
A nudging plan consists of a set of nudges
{n
1
, . . . , n
m
}, where each nudge can be predesigned,
including a tentative activity, time frame, and other
nudge components. When it is time to nudge the user,
the final nudge design is done, and tentative compo-
nents may be replaced if the user situation at nudge
design time makes it necessary or beneficial. The
nudging plan may be adjusted during P as the user sit-
uation may change or the ActivityGap(P) is updated
and the prediction changed.
A physical activity nudging plan can be created
by detecting periods, during P, when the user is avail-
able for being active, and planning a nudge for each
KMIS 2024 - 16th International Conference on Knowledge Management and Information Systems
160

selected time frame. The planned nudges are spread
over the period, and design and presentation of a
nudge are done close to the selected time frame.
A green transportation nudging plan can be cre-
ated by identifying planned travels (i.e., (O,D) pairs)
during P, where the user’s transportation habits can
be improved, and planning a nudge for each selected
(O,D) pair.
5.3 A Sequence of Nudges
Ranking of activities can be used as a basis for design-
ing a sequence of linked nudges, where the first nudge
suggests a higher-ranked activity than the next nudge
in the list. When one of the linked nudges is accepted,
the following nudges in the list are discarded.
For example, when nudging for transportation
means between O and D, the system can first nudge
for walking, and later, if the user did not walk, issue
a new nudge for taking the bus. This gives a list of
linked nudges, where walking has a better EF-value
compared to public transportation.
5.4 Fulfilling Nudge Design Principles
The nudge design approach presented in this paper,
follows the design principles described in Section 2.1.
The principles of challenge, consolidate, and progress
describes how nudging should stepwise challenge the
user to improve behavior. The need to challenge
the user is supported by automatically increasing the
ActivityGoal value as the user’s behavior improves.
The consolidate and progress principles are handled
using the three behavioral progress states (improved,
stable, decreased), where the consolidate principle is
followed when the user is in the improved state, while
the progress principle is used for the states stable and
decreased.
The variation principle is supported by keeping a
set of possible activities (i.e., Possible(A,U)) and rec-
ognizing that an activity can be selected from either
of the two sets Probable(A,U) and U nlikely(A,U ),
as long as the activity represents an improvement or a
status quo with respect to user behavior.
Timeliness is supported by detecting time frames
when nudging is needed or can be effective for the
user. For physical activity nudging, an opening in
the calendar can be utilized for exercising, while for
green transportation nudging an arrival time at the
destination, together with the selected activity, sets a
required time frame for a nudge.
Feasibility is supported through classification of
activities, where impossible activities will never be
nudged for, while safety is supported by recognizing
harmful or hazardous situations as obstacles that iden-
tify an activity as potentially dangerous and classify it
as impossible.
6 CONCLUSION
This paper presents a model for adaptive nudge de-
sign, that provides personalized and context-aware
nudges tailored to the user’s current need for a gen-
tle push towards a desirable change in behavior. The
model follows a just-in-time nudge design approach,
where tailoring of nudges is based on the user’s be-
havior, situation, and environment at the time of nudg-
ing.
The design process adapts according to the user’s
change in behavior, by continuously challenging the
user to improve based on an adaptive activity goal,
which is automatically adjusted as user behavior im-
proves. Activity and time frame for a nudge is se-
lected based on what is needed for the user to reach
the activity goal, and what currently is possible for the
user.
This paper targets the questions of what to nudge
for and when to nudge. Other aspects of nudge design,
such as details regarding which influence type to se-
lect and how to present the nudge are left for future
work.
The model is described through a general ap-
proach to selecting activity and time frame for a
nudge, followed by examples of how the proposed so-
lution can be used in two different use cases; physi-
cal activity nudging and green transportation nudging.
Future work includes applying the general approach
to other use cases. Presenting practical experiments
using the design model is also left for future work.
REFERENCES
Anagnostopoulou, E., Urban
ˇ
ci
ˇ
c, J., Bothos, E., Magoutas,
B., Bradesko, L., Schrammel, J., and Mentzas, G.
(2020). From mobility patterns to behavioural change:
leveraging travel behaviour and personality profiles to
nudge for sustainable transportation. Journal of Intel-
ligent Information Systems, 54(1):157–178.
Andersen, A., Karlsen, R., and Yu, W. (2018). Green trans-
portation choices with IoT and smart nudging. Hand-
book of Smart Cities: Software Services and Cyber
Infrastructure, (1):331–354.
Bergram, K., Djokovic, M., Bezenc¸on, V., and Holzer, A.
(2022). The Digital Landscape of Nudging: A Sys-
tematic Literature Review of Empirical Research on
Digital Nudges. Conference on Human Factors in
Computing Systems - Proceedings.
A Model for Designing Personalized and Context-Aware Nudges
161

Caraban, A., Karapanos, E., Gonc¸alves, D., and Cam-
pos, P. (2019). 23 Ways to Nudge: A review of
technology-mediated nudging in human-computer in-
teraction. Conference on Human Factors in Comput-
ing Systems - Proceedings, pages 1–15.
Caraban, A., Konstantinou, L., and Karapanos, E.
(2020). The nudge deck: A design support tool for
technology-mediated nudging. DIS 2020 - Proceed-
ings of the 2020 ACM Designing Interactive Systems
Conference, pages 395–406.
Comission of the European Communities (2007).
COM(2007) 551 final: Green Paper - Towards a
new culture for urban mobility. Green paper.
Dalecke, S. and Karlsen, R. (2020). Designing Dynamic
and Personalized Nudges. ACM International Confer-
ence Proceeding Series, Part F1625:139–148.
Forberger, S., Reisch, L., Kampfmann, T., and Zeeb, H.
(2019). Nudging to move: A scoping review of the use
of choice architecture interventions to promote phys-
ical activity in the general population. International
Journal of Behavioral Nutrition and Physical Activ-
ity, 16(1):1–14.
Jesse, M. and Jannach, D. (2021). Digital nudging
with recommender systems: Survey and future direc-
tions. Computers in Human Behavior Reports, 3(Jan-
uary):100052.
Karlsen, R. and Andersen, A. (2019). Recommendations
with a Nudge. Technologies, 7(2):45.
Karlsen, R. and Andersen, A. (2022). The Impossible, the
Unlikely, and the Probable Nudges: A Classification
for the Design of Your Next Nudge. Technologies,
10(6):1–17.
Karlsen, R. and Andersen, A. (2024). How to Trigger Per-
sonalized and Context-Aware Nudges. Presented at
the 28th Int. Database Engineered Applications Sym-
posium (IDEAS 2024), Bayonne, France, 28-31 Au-
gust, 2024.
Meske, C. and Potthoff, T. (2017). The dinu-model – A
process model for the design of nudges. Proceedings
of the 25th European Conference on Information Sys-
tems, ECIS 2017, 2017:2587–2597.
Michie, S., van Stralen, M. M., and West, R. (2011). The
behaviour change wheel: A new method for character-
ising and designing behaviour change interventions.
Implementation Science, 6(1):1–11.
Mills, S. (2022). Personalized nudging. Behavioural Public
Policy, 6(1):150–159.
Mirsch, T., Lehrer, C., and Jung, R. (2018). Making
digital nudging applicable: The digital nudge design
method. International Conference on Information
Systems 2018, ICIS 2018, (2009):1–16.
Peer, E., Egelman, S., Harbach, M., Malkin, N., Mathur, A.,
and Frik, A. (2020). Nudge me right: Personalizing
online security nudges to people’s decision-making
styles. Computers in Human Behavior, 109(Novem-
ber 2019):106347.
Purohit, A. K. and Holzer, A. (2019). Functional digi-
tal nudges: Identifying optimal timing for effective
behavior change. Conference on Human Factors in
Computing Systems - Proceedings, (May).
Schneider, C., Weinmann, M., and vom Brocke, J. (2018).
Digital Nudging: Guiding Online User Choices
through Interface Design. Communications of the
ACM, 61(7):67–73.
Sch
¨
oning, C., Matt, C., and Hess, T. (2019). Personalised
nudging for more data disclosure? On the adaption of
data usage policies format to cognitive styles. Pro-
ceedings of the Annual Hawaii International Confer-
ence on System Sciences, 2019-Janua(January):4395–
4404.
Thaler, R. H. and Sunstein, C. R. (2008). Nudge: Improving
Decisions About Health, Wealth, and Happiness. Yale
University Press.
Villalobos-Z
´
u
˜
niga, G. and Cherubini, M. (2020). Apps That
Motivate: a Taxonomy of App Features Based on Self-
Determination Theory. International Journal of Hu-
man Computer Studies, 140(April):102449.
WHO (2016). WHO Guidelines on physical activity and
sedentary behaviour. WHO.
KMIS 2024 - 16th International Conference on Knowledge Management and Information Systems
162