REVERSE MARKET SEGMENTATION WITH PERSONAS

Harri Ketamo

Satakunta University of Applied Sciences, Pori, Finland

Kristian Kiili

Tampere University of Technology, Pori unit, Pori, Finland

Jarkko Alajääski

University of Turku, Department of Teacher Education, Rauma, Finland

Keywords: User Experiences, Digital Learning Materials, Market Segmenting, Adaptive Systems.

Abstract: In this study the user experiences of commercial educational product were gathered in order to build the

personas that can be used to revising the market segmentation. Personas are empirically formed archetypical

characters representing distinct behavioural clusters, goals and the motivation of end users. Usually

personas are used in different production phases as tools that help designers and marketing people in

decision making. In this study, the personas were formed by applying k-means cluster analysis into

quantified user interviews. According to the results of the study, qualitatively formed personas showed their

strengths as decision making tools: They helped publisher to maintain the focus on a learner's needs, wants

and requirements during the whole process of development.

1 INTRODUCTION

According to Cooper (1999) personas are

empirically formed archetypical characters

representing distinct behavioural clusters, goals and

the motivation of end users. Usually personas are

used in different production phases as tools that help

developers, producer and publishers to make

reasonable and empirically based design decisions

(e.g. Cooper, Reimann & Cronin 2007).

In practice, personas and market segmentation

has been seen as complementary methods (e.g.

Kujala & Kauppinen 2004; Grudin & Pruitt 2002).

Traditional market segmentation can help the

designers to build commercially more effective

personas for product development.

However, in this study, the real user experiences

were used to build the personas that can be used

when revising the market segmentation. In other

words, the “from segments to personas” -process is

reversed to “from experiences to personas to

segments”

The reason, why the reversed market

segmentation is used relays to the complexity of

adaptive/personalized content. The idea of adaptive

educational systems is to produce individual and

optimized learning experiences (Eklund &

Brusilovsky 1999; Brusilovsky 2001) and the high

end user models that are relatively complex. In terms

of complexity, a definition for a complete adaptive

system can be based on the capability of self-

organization.

The persona approach provides a sound design

framework for complex adaptive learning materials.

One of the main benefits of using personas is that it

helps to eliminate the problem of the "elastic

learner". An elastic learner stretches and adapts

during the design process, allowing designers to

implement almost anything. However, real learners

are not elastic. Thus, the persona method aims to

design of learning materials that will stretch and

adapt to the learner's needs - not the other way

around. Such an approach is the key to effective

learning experience.

Ketamo H., Kiili K. and AlajÃd’Ãd’ski J.

REVERSE MARKET SEGMENTATION WITH PERSONAS.

DOI: 10.5220/0002781300630068

In Proceedings of the 6th International Conference on Web Information Systems and Technology (WEBIST 2010), page

ISBN: 978-989-674-025-2

2 RESEARCH TASK AND

METHODS

2.1 Research Tasks

This study focuses on the 1) evaluation of the

product and 2) finding most suitable customers

(segments) for Mathematics Navigator. The

collected data includes qualitative and quantitative

variables. In this study the focus is on qualitative

data. Quantitative data, such as test scores and

improvement measures in learning results are used

to support the decision making.

In this study, user prototypes are formed

empirically according to qualitatively measured user

experience. The research tasks are:

1. To form personas from the user experiences

gathered with interviews.

2. To find most suitable market segment for

Mathematics Navigator.

3. To evaluate the usefulness of the persona -

method.

The sample (n=74) consists of first year class

teacher students at the University of Turku,

Department of Teacher Education in Rauma. All of

the participants had an upper secondary school

degree on mathematics – Basically all of the

mathematics related to this study is based on upper

secondary school mathematics curriculum.

2.2 Materials

Mathematics Navigator is a product family

published by Otava Publishing Company Ltd.

Mathematics Navigator is designed to operate as a

personal tutor and guide in the studies. It supports

the development of a student's mathematical skills

and abilities and helps the student in recognizing

his/her mathematical strengths and weaknesses.

Mathematics Navigator gathers information on

the student's actions while studying. On the basis of

these, Mathematics Navigator adjusts the set of

exercises and/or content presented to the student in

order to optimally support his/her development.

Thus, the learning paths are formed individually.

Furthermore, the student is able to follow the

development of his/her competence profile.

The user interface of Mathematics Navigator

consists of a menu-bar and three main areas (Figure

1). At the left side of the interface is the table of

contents that have two different views of the

content: 1) The traditional book-like table of

contents and 2) an exercise adapted table of

contents.

Figure 1: User interface of Mathematics Navigator and the

basic mathematics course (in Finnish).

The exercises are presented in the right-bottom

corner of the user interface one at a time. The

exercises can't be neglected, changed or left behind

without giving an answer. The learning profile is

always (real-time) adjusted on basis of the answer to

the current exercise. The exercises are selected to

support the individual user's needs. There are no

fixed paths through the topics: the path is based on

the student's learning profile and by Mathematics

Navigator estimated need for exercises and contents.

The guiding factors in the exercise selection are: 1)

The course structure (the traditional table of

contents). 2) Student’s learning abilities and areas of

weaknesses measured and estimated by Mathematics

Navigator during the studying process. 3) Critical

points, derived from the learning community's

actions, derived from generalized success patterns in

answering the exercises.

The competence profile values are indicated with

colours that vary from red (insufficient skills or not

yet estimated skills) to green (good skills). Those

skills mastered and measured within a certain theme

will be transferred with certain estimates to other

themes requiring similar (by proximity or by

hierarchy) skills.

2.3 Measures and Analyses

In the beginning of the experimental period, a pre-

test, and at the end of it, a post-test, were arranged

for measuring the educational effect of the

Mathematics Navigator. The tests were based on the

Finnish lower secondary school and partly upper

secondary school mathematics curricula. After the

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

post-test feedback was gathered through the

following two open questions (originally in Finnish):

1. What kind of good or positive experiences

did you have during your work with

Mathematics Navigator?

2. What kind of development needs or negative

experiences did you encounter during your

work with Mathematics Navigator?

At first, feedback was analyzed with a quantitative

method – a thematic analysis was conducted. The

thematic analysis started with forming main themes.

The themes were formed as archetypes of enough

similar comments. The themes were reviewed by

two independent professional learning material

producers. One has a background in social sciences

and the other has a background in natural sciences.

Finally 16 themes were determined. Several themes

could occur in one feedback. The number of times a

theme was emphasized was not calculated. The

theme either exists in the feedback or then there was

no such theme in the feedback. The number of

different themes in a single feedback varied from

zero to eight themes per feedback.

The quantification of the qualitative data was

done as a binary data matrix, where each test person

was represented by a case (as done traditionally).

Each theme was represented by a variable, the value

of which is either 1 or 0. The variable got a value of

1 only if the theme that variable described was

expressed in the feedback. In the other case, the

value of the variable was 0. The quantified data was

extended by the results from the pre- and post test.

In this study, k-means clustering was used to

build the personas according to the quantified binary

data. Thus, k-means clustering was designed to work

with continuous variables. It is note worthy that k-

means clustering has been successfully used with

binary data (e.g. Postaire, Zhang & Lecocq-Botte

1993; Ordonez 2003). The differences between

continuous and binary data should be taken into

account while analyzing the results. Finch (2005)

has shown that there are small differences between

different proximity calculation methods with k-

means. In this study, k-means cluster analysis was

used to maximize the distances between clusters. In

this approach, the proximity calculations are not as

crucial as compared to cases where we are aiming to

form tight clusters and extract all the outliers.

In this paper, the clustering result is discussed

along with literature on behavioural theories in order

to ensure the validity of clustering. Secondly, we use

clustering to characterize the personas, the clustering

result is not in and of itself a result alone - it is a

starting point for building personas.

3 RESULTS

3.1 Thematic Analysis

The written feedback was classified into thematic

categories in three iteration phases (table 1).

Table 1: Theme frequencies and descriptions.

In the first phase all the different expressions were

extracted. In the second phase, the expressions were

integrated into 14 themes. In the third iteration

phase, the themes were reviewed by two external

people (not the authors) who are professional

REVERSE MARKET SEGMENTATION WITH PERSONAS

learning material producers. One has a background

in social sciences and the other has a background in

natural sciences. During this review, two themes

were divided into four themes and finally there were

16 themes with relatively unambiguous definitions

(table 1).

Most of the feedback expressions contained

several themes. The number of different themes in a

single feedback varies from zero to eight themes per

feedback. The number of times a theme was

emphasized was not calculated. The theme either

exists in the feedback or then there is no such theme

in the feedback.

3.2 Clusters based on Thematic

Analysis

Quantifying the qualitative data is described in

method -section. The visualization of k-means

clustering is shown in figure 2. In the clustering, the

distances between clusters were maximized in order

to build personas that 1) form a proximate group and

2) are as different as possible to other clusters in

order to support further development design.

The clusters are interesting when comparing the

learning achievements and skills between the

clusters. Generally, the average improvement in test

scores was 5 points (median, when N=74), which

achieves a statistically significant difference (t=-

2,054, df=146, p=0,042). When focusing on learning

achievements by clusters, there was only one group

with a statistically significant improvement.

Table 2: Learning outcome for each cluster.

In Table 2 the learning outcome (improvement in

test score) and skills (test score) measured in pre-

and post-test are shown. According to the results,

only members in cluster 5 have had a statistically

significant improvement (t=-2,082, df=22, p=0,049)

in test scores.

Figure 2: Plot of means for clusters received from

quantified feedback data.

Next we build qualitative descriptions of the

personas according to the classification of features

of the clusters. For this we determine the major

feature, which is represented by the theme that is

mentioned by more than 60% of the cluster

members. A minor feature is represented by the

theme that is mentioned by more than 40% of the

cluster members, but less than 60% of the cluster

members.

According to these limits, we can build the

following personas and their descriptions (Table 3).

The personas (each cluster) were given a name in

order to highlight their definitions as user

archetypes. Short descriptions of the personas are

following:

Matti (representing cluster 1) values freedom and

independence in digital learning systems. However,

he still wants clear teaching and tutoring features.

Teaching and tutoring make him feel more confident

due to his average skills in mathematics.

Kirsi (representing cluster 2) prefers traditional

classroom teaching. She thinks that digital learning

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

materials are good for revising, but are not adequate

as the only method. She does not like learning

systems that penalize too much for mistakes made,

but she values positive encouragement. Kirsi does

not like to work much for success that usually leaves

the learning outcome poor.

Table 3: Personas and their descriptions.

Aapo (representing cluster 3) is relatively good at

mathematics. He masters the use of digital products

quite well and easily masters the use of new

systems. Aapo values challenges that motivate him.

Non-challenging tasks do not engage Aapo.

Eero (representing cluster 4) is a student that

wants everything to be easy. He values freedom and

independence in studying, but does not like to work

much for success. Eero does not like learning

systems that penalize too much for mistakes made,

but values positive encouragement.

Anna (representing cluster 5) values freedom and

independence in digital learning systems. She needs

a lot of theoretical content and specific examples of

exercises. Furthermore, she needs extensive, clear

instructions and explanations about how learning

systems operate. She also values complete and

correct solutions for exercises. Although, Anna's

mathematic skills are relatively good, working with

good digital learning materials encourages her to

reach an even higher skill level.

3.3 Analysis about Personas

The formed personas are classified as primary,

secondary and tertiary personas in accordance with

their importance to define the potential customers.

Anna (5) is selected as a primary persona

because of her good learning outcome and

constructive design ideas. Also Matti (1) is classified

as a primary persona. Anna and Matti will also be

the main sources, when further developing

Mathematics Navigator. By accepting all three needs

for further development suggested by persona 5, we

can also support persona 1's suggestion for more

teaching and tutoring: Adding detailed solutions to

exercises with more detailed content can be expected

to help persona 1 in some way. Because of his lower

skills at the beginning, we cannot be sure if this is

enough.

Kirsi (2) is selected as a secondary persona: By

developing instructions as suggested by Anna, we

can also deliver ideas on how Mathematics

Navigator can be used during classroom teaching in

a pedagogically meaningful way. However, several

pedagogical tests without a scientific context have

been done in classrooms by teachers. We can

produce directions about good practices in

classrooms after we have collected feedback from

those teachers.

Aapo (3) and Eero (4) are classified as tertiary

personas. Aapo has relatively good skills with

relation to the objectives of the learning material.

We can hope that he learned something that we have

not measured. Aapo's positive feedback about

Mathematics Navigator allows us to determine that

there was nothing critical in Mathematics Navigator.

However, when developing Mathematics

Navigator we have to ensure that the user interface

remains as easy to use as it is now. Eero has a

negative attitude against learning, self-studying in

particular. Learning is not always easy. It can

certainly be fun, but according to previous studies, a

good learning outcome requires work. It would be a

crucial mistake to make Mathematics Navigator

easier or more abbreviated. Hopefully more

informative directions will help this persona to better

understand the nature of self-studying and learning.

Kirsi's and Eero's concern about the competence

profiling system being too penalizing is valuable

from a motivational point of view. In order to

engage and motivate users, the profiling system will

REVERSE MARKET SEGMENTATION WITH PERSONAS

be fixed. In the future, the exercise selection will be

based on a real competence profile, while the visible

competence profile will be designed to be more

humane: it will not immediately penalize for one

mistake.

4 CONCLUSIONS

Personas are powerful design tools if they are used

correctly. They help designers, producers and

publishers to maintain focus on a learner's needs,

wants and requirements during the whole

development process. Furthermore, personas enable

the whole production team to achieve a shared

understanding of the requirements and the context

within the learning taking place. Production team

can make decisions based on user archetypes rather

than basing the decisions on their own intuition or

personal likes.

In this study personas were constructed in order

to ground publishing decisions of Mathematics

Navigator. Qualitative user feedback was analyzed

thematically at first. Secondly the users were bound

to a certain clusters according to the proximity of

their feedback. Finally strict clusters with

meaningful common nominators were named as

personas. Personas in this study are based on the

mathematical modelling of quantified user

experiences and therefore they are highly valid

archetypes of the tested population. If the archetypes

had been formed only according to thematically

analyzed feedback, the outcome of the study would

have been different.

Furthermore, several decisions about further

development have been made according to results of

this study: 1) The quality and quantity of feedback

from Mathematics Navigator to the learner will be

improved. Complete solutions to exercises will be

added. Also, tools for accessing completed exercises

and solutions will be designed. 2) General

instructions will be rewritten in accordance with the

feedback. However, this could have been done

without the personas -method, but the importance of

the task would have not been so clear to the

developers. 3) The competence profile that was

experienced as being too penalizing will be fixed.

Exercise selection will be based on a real

competence profile, while the visible competence

profile will be designed to be more humane: it will

not penalize immediately for one mistake.

During this study a new research challenge

emerged: Is it possible to construct artificial test

users according to personas? According to this idea

artificial users represent archetypes of human users

with a certain variance in behaviour. In other words,

the artificial users are computational representations

of personas: They will be constructed according to

the behaviour of real users in digital environments

by analyzing the behaviour as quantitative

phenomena and designing a representation of a

system, corresponding to the behaviour. Such a

system can be implemented as a software agent. As

a test person, a software agent can communicate

with the educational systems by e.g. Web Services

interfaces. An interesting question is related to the

behaviour of the artificial user: Is its general level

comparable to the behaviour of human users?

REFERENCES

Brusilovsky, P. (2001) Adaptive Hypermedia. User

Modeling and User-Adapted Interaction. 2001 vol 11,

pp. 87-110.

Cooper, A. (1999). The inmates are running the asylum.

New York: Macmillan.

Cooper, A., Reimann, R. & Cronin, D. (2007). About Face

3: The Essentials of Interaction Design. Indianapolis,

Indiana: Wiley Publishing Inc.

Eklund, J. & Brusilovsky, P. (1999). InterBook: an

Adaptive Tutoring System. UniServe Science News;

1999 vol 12.

Finch, H. (2005). Comparison of Distance Measures in

Cluster Analysis with Dichotomous Data. Journal of

Data Science, vol 3(1), pp. 85-100.

Grudin, J. & Pruitt J. (2002). Personas, participatory

design and product development: an infrastructure for

engagement. In Proceedings of the 7th biennial

particpatory design conference, Malmö, Sweden, June

2002, pp. 144-161.

Kujala, S. & Kauppinen, M. (2004). Identifying and

Selecting Users for User-Centered Design. In

Proceedings of the third Nordic conference on Human-

computer interaction, October 2004. Tampere,

Finland, pp. 297-303.

Ordonez, C. (2003) Clustering Binary Data Streams with

K-means. In Association for Computing Machinery

(ACM) Special Interest Group: Management of Data

(SIGMOD) Workshop on Research Issues on Data

Mining and Knowledge Discovery (DMKD), pp. 10-

17.

Postaire, J.G., Zhang, R.D. & Lecocq-Botte, C. (1993).

Cluster Analysis by Binary Morphology. IEEE

Transactions on Pattern Analysis and Machine

Intelligence, vol. 15(2), pp. 170-180.

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies