<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.0 20120330//EN" "JATS-archivearticle1.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink">
  <front>
    <journal-meta />
    <article-meta>
      <title-group>
        <article-title>Product Configuration in the Wild: Strategies for Conflicting Decisions in Web Configurators</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <string-name>Thomas Thüm</string-name>
          <xref ref-type="aff" rid="aff0">0</xref>
        </contrib>
        <contrib contrib-type="author">
          <string-name>Sebastian Krieter</string-name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <string-name>Ina Schaefer</string-name>
          <xref ref-type="aff" rid="aff0">0</xref>
        </contrib>
        <aff id="aff0">
          <label>0</label>
          <institution>TU Braunschweig</institution>
          ,
          <country country="DE">Germany</country>
        </aff>
        <aff id="aff1">
          <label>1</label>
          <institution>University of Magdeburg</institution>
          ,
          <country country="DE">Germany</country>
        </aff>
      </contrib-group>
      <abstract>
        <p>Customization is omnipresent in our everyday live. There are web configurators to customize cars, trucks, bikes, computers, clothes, furniture, and food. At first glance, customization using configurators appears trivial; we simply select the configuration options that we want. However, in practice, options are usually dependent on each other. Reasons for dependencies are manifold and are typically specific for the particular domain. Dependencies can be simple, such as one option requiring or excluding another option, but also arbitrarily complex, involving numerous options. In this study, we aim to understand how today's web configurators support users in their decision making process. In particular, we are interested in understanding how configurators handle decisions that are in conflict with dependencies. To abstract from different visualizations, we classify the existing strategies of web configurators and discuss advantages and disadvantages of them. While we identified eight strategies, a single configurator typically uses several of those strategies.</p>
      </abstract>
    </article-meta>
  </front>
  <body>
    <sec id="sec-1">
      <title>-</title>
      <p>
        Mass customization is the vision that customized products are
produced to a price similar to that with mass production [
        <xref ref-type="bibr" rid="ref13 ref15 ref3 ref4">3, 4, 13, 15</xref>
        ].
This vision requires that customers specify their needs
explicitly [
        <xref ref-type="bibr" rid="ref10">10</xref>
        ]. To this end, there are thousands of product configurators
available in the world wide web,3 which guide customers during the
decision making process. Unfortunately, customers do not only need
to understand which options are available, but also which
combinations of options are valid. For the success of mass customization it is
crucial that potential customers are able to easily explore valid
combinations and to make compliant decisions. Our personal experience
with product configurators suggests that we are not there yet, as
configurators often require a considerable mental effort from their users.
      </p>
      <p>In Figure 1, we show an excerpt of a configurator for a Lenovo
ThinkPad. The excerpt contains three configuration options, of which
each stands for a different display being available for the notebook.
However, these three options cannot be chosen freely due to existing
dependencies. First, the options are alternatives to each other,
meaning that a notebook must have exactly one of those three displays.
For that purpose, the three options are arranged in a category called
Display. Second, the green hint below the last option reveals that this
display is only available if option WWAN (not shown in the excerpt)
is chosen. While the first dependency is enforced by the configurator,
the second must be taken care of by the user. We argue that
dependencies are one of the main challenges for customers, as they heavily
influence the decision making process.</p>
      <p>
        While there are several approaches to implement configurators [
        <xref ref-type="bibr" rid="ref15 ref5">5,
15</xref>
        ], we aim to understand how configurators handle dependencies
from a user’s point of view. That is, we want to identify strategies to
guide the decision process, which are currently applied in real-world
configurators. Our insights may be used to uncover gaps between
research and practice on product configuration. In particular, we
analyze advantages and disadvantages of each strategy, which may
support development of more sophisticated strategies in the future.
      </p>
      <p>By studying a corpus of seven web configurators (cf. Section 2),
we identified eight strategies to handle dependencies and discuss
their advantages and disadvantages (cf. Section 3). We focus on web
configurators, because they are freely available and are used by
thousands or even millions of customers. Interestingly, of each
configurator uses a different subset of all identified strategies. Moreover,
configurators with fewer categories seem to be simpler in a sense
that they apply fewer strategies. In particular, we even studied a
configurator applying all eight strategies.
2</p>
    </sec>
    <sec id="sec-2">
      <title>Subject Configurators</title>
      <p>As considering all product configurators on the web is certainly not
feasible, finding a representative selection of configurators is crucial
for our study. Initially, we googled the term “configurator” and
inspected the first 100 search results to understand which configurators
are used frequently on the web. Almost all entries that we found were
configurators for cars, but we wanted to cover more than just the
automotive industry. We though of further industries where we
experienced mass customization before. Then, we tried to find
representative configurators for those industries that are likely to be used by
a wide audience. Ultimately, we have chosen configurators from the
domain of automotive, mobile computers, smartphone accessories,
and clothing.</p>
      <p>In our study, we do not distinguish between configurators that
are used for customized production opposed to a selection of
preproduced goods (aka. selectors), as this distinction is typically not
visible to customers anyway. Even though the number of potential
combinations and length of the delivery period might be good
indicators. For instance, in order to speed up delivery, t-shirts are
typically available in different sizes and colors, which are often produced
before customers make their decision.</p>
      <p>As we aim to provide evidence for our study by means of
screenshots, we decided to use the UK version of all configurators.
Nevertheless, we have drawn samples in other languages, too. The only
differences we identified were options being available only in some
countries, whereas the principle strategies have been identical. We
accessed all configurators in May 2018, whereupon we experienced
that configurators and their behaviors change on a daily basis. That
is, dependencies are updated frequently, whereas we did only
experience changes in the strategy for one configurator (cf. Section 3.2).
BMW 1 (3 Door) BMW is a German car manufacturer
providing individual configurators for 49 different car models.4 As cars are
known to have hundreds of configuration options, we decided to use
the configurator for the cheapest car model to find a comparatively
small, and thus manageable number of configuration options for our
study. There are 42 alternative options for engines and gearboxes,
eleven exterior colors, five alloy wheels, three options for upholstery,
four interior designs, two packages, and 81 further options with
respect to optional equipment.</p>
      <p>Toyota AYGO x-play 5 Door Hatchback With about 9 million
sold cars a year, Toyota is one of the largest car manufacturers.
Toyota offers 19 models, which are available in a total of 84 editions
of which each can be configured separately. Just as for BMW, we
wanted to select the cheapest model and edition available, namely
Toyota AYGO x3 Door Hatchback. However, users cannot even
choose anything in the first configuration step for this edition. Thus,
we have chosen the cheapest edition for this model for which users
can choose between manual and automatic gearshift, namely Toyota
AYGO x-play 5 Door Hatchback.5
HP Velotechnik Streetmachine GTE Besides cars, many other
transportation means exist that can be configured. In particular,
bicycles can often be configured by customers. We investigated the
configurator of HP Velotechnik, as we already had experience with
this configurator prior to this study. Customers can choose one of 14
recumbent bicycles and then start the configurator. We have chosen
the model Streetmachine GTE which provides 34 categories of which
customers have to chose one option each.6
Lenovo ThinkPad X1 Yoga (3rd Gen) ThinkPad X1 is one of the
most expensive and powerful convertibles of Lenovo. Before users
can configure it on Lenovo’s website,7 they have to choose between
three models in black and one model in silver. Of those, we have
chosen the cheapest as it seems to provide the most configuration
options (i.e., many options could not be downgraded in the more
expensive models). The configurator provides 39 configuration options
in eleven categories, such as processor, operating system, display,
hard drive, and keyboard. Besides those, there are another eleven
categories with a single option (i.e., cannot be selected by the user).
4 https://www.bmw.co.uk/configurator#/
5 https://www.toyota.co.uk/new-cars/aygo/build
6 http://hpvelotechnik.velocom.de/step-1.jsf
7
https://www3.lenovo.com/gb/en/laptops/thinkpad/xseries/ThinkPad-X1-Yoga-3rd-Gen/p/22TP2TXX13Y</p>
      <p>Microsoft Surface Book 2 Surface Book 2 is the latest high-end
convertible by Microsoft. With 13 configuration options in five
categories there is only some rudimentary support for customization. On
Microsoft’s website there is a link to pre-configured products,8 but
also to different versions of a configurator to which we later refer
to as a version with a default configuration9 and a version without a
default configuration.10
T-Shirts at Amazon Online shopping is an increasing market and
Amazon is one of the most popular web stores. Although not all
products can be configured, there is huge amount of products for which
at least certain properties, such as color and size, can be specified.
Even though there are many products, the configurator technology
seems to be rather generic at Amazon. For our study, we have
chosen a t-shirt configurator, as t-shirts are a rather common consumer
good and typically have two categories, which is necessary to study
dependencies. The subject configurator offers 21 colors and 8 sizes,
whereas 123 of 168 combinations are valid (i.e., 73.2%).11
Smartphone Cases at Ebay Another popular web store is Ebay,
which appears to be similar in terms of configuration compared to
Amazon. Again, the underlying configurator seems to have the same
behavior for very different kinds of products. Nevertheless, we have
to decide on one product for our case study, for which we have
chosen a configurator for smartphone cases.12 The configurator supports
the selection out of 36 smartphone models and eight colors. Overall,
208 out of 288 principal combinations are valid (i.e., 72.2%).
3</p>
    </sec>
    <sec id="sec-3">
      <title>Strategies for Conflicting Decisions</title>
      <p>We investigate all seven configurators to explore how they deal with
conflicting customer decisions. For that purpose, we randomly
selected options and observed whether those decisions had any
consequences for the selection of other options. We identified the following
eight strategies, whereas we abstract from the visual representation
of selected and deselected options.
3.1</p>
    </sec>
    <sec id="sec-4">
      <title>Automatic Deselection in Alternatives</title>
      <p>A simple but effective strategy exists for alternative options within
one category. If an option A is already selected and a user attempts
to select an alternative option B in the same category, a configurator
may automatically deselect option A. Users are typically not notified
about this change.</p>
      <p>All configurators that we studied apply this strategy. As an
example, we refer again to Figure 1 and the display options of a ThinkPad.
If a user selects the second or third display option, the first display
option is deselected (i.e., the price difference is shown).</p>
      <p>The automatic deselection avoids further burden on the user for
these kinds of simple conflicts, as the conflict is immediately solved.
It is also an intuitive strategy, especially if accompanied with a
graphical representation that suggests that these options are alternative to
8 https://www.microsoft.com/en-gb/surface/devices/
surface-book-2/
9 https://www.microsoft.com/en-gb/store/config/
surface-book-2/8MCPZJJCC98C/17NG
10 https://www.microsoft.com/en-gb/store/config/
surface-book-2/8MCPZJJCC98C
11 https://www.amazon.co.uk/gp/product/B00V3IDB3Q
12
https://www.ebay.co.uk/itm/Sports-Running-GymCycling-Jogging-Armband-Case-Cover-For-HuaweiMobile-Phones/181809023183
each other (e.g., a drop down box as in Figure 3 or radio buttons as
in Figure 5). Nevertheless, the deselection is problematically if the
user does not recognize it (e.g., if the deselected option is not on
the screen). Furthermore, the deselection itself may result in further
conflicts with other decisions.
3.2</p>
    </sec>
    <sec id="sec-5">
      <title>Starting with a Default Configuration</title>
      <p>Dependencies are typically unsatisfied when no options are selected
at all. For example, a notebook needs exactly one kind of display. A
very common strategy to avoid invalid states when starting the
configuration process is to start with a default configuration. The default
configuration satisfies all dependencies and may be changed by the
user during the process.</p>
      <p>For instance, the first display is already selected when opening
Lenovo’s configurator (cf. Figure 1) and in BMW’s configurator a
default engine is selected from the start. Interestingly, Microsoft’s
configurator can be accessed via different links of which some start
with a default configuration and others do not. However, even when
starting with a valid default configuration in Microsoft’s configurator,
any change resets decisions on later categories. As a consequence,
changing the display to 15 inch removes all default selections and
makes the configuration invalid (cf. Figure 2). It requires decisions
on all subsequent categories to finish and to come to a valid
configuration again. Note that all decisions of subsequent categories are
reseted independent of whether they are actually conflicting or not.13</p>
      <p>Starting with a valid configuration means that a user can also skip
the configuration process completely. It may also help users to avoid
making some decisions, if they do not care about all options. They
simply have to change options they care about. However, the
downside of starting with a default configuration is that users can often
not recognize which options they have chosen and which have
already been chosen by the default configuration. As a consequence,
they may forget to examine some of the decisions and may end up
with a product not customized to their needs.
3.3</p>
    </sec>
    <sec id="sec-6">
      <title>Hiding Invalid Combinations</title>
      <p>Another simple strategy to deal with conflicts is to hide all options
that are in conflict with previous decisions. At the start, all options of
13 When we tested the configurator in April 2018, the configuration in
subsequent categories was even re-assigned automatically. This behavior was
even worse from a user’s point of view as decisions on all subsequent
categories are not just requested again, but even overridden without notice. It
is likely that this behavior was replaced as being too unintuitive or even
questionable from a legal perspective.
all categories are visible and no option is selected by default. Once
a user makes a selection in one of the categories, all other categories
are filtered for compatible options. As a result, no decision of the
user leads to a conflicting state.</p>
      <p>This strategy is followed by the configurator available to products
at the Ebay web store. All models and all colors are shown when
nothing is selected. For a given color, between 22 and 33 models are
visible, and for a given model, there are between one and all eight
colors. In Figure 3, we selected the Huawei Ascend P7, for which a
case is only available in black, blue, and gray.</p>
      <p>The advantage of hiding invalid combinations is that users are not
overwhelmed by a large number of options that are actually not
available for their prior decisions. Also, there is no need to solve any
conflicts and users only need to decide on the remaining valid options.
Nevertheless, there are also major drawbacks. This strategy seems
to be rather incompatible with a default configuration and, thus, it
will not be possible to simply continue and skip decisions. Requiring
users to make every decision is fine if there are only a few categories,
but can be infeasible for large configuration spaces. Nevertheless, we
found such a combination in the configurator by Toyota: a default
car color is selected when entering the configurator for which black
wheels are filtered out. Hence, users will only find out about black
wheels if they change the car color first. Furthermore, this strategy
might be unintuitive to users who do not expect such a filtering.
Indeed, it took us a while to find out that the lists in the Ebay
configurator are changed depending on other selections. Finally, it seems that
users should start with selecting the most relevant criteria for them,
because otherwise they will simply not notice that there are other,
potentially better options for later selected categories.
3.4</p>
    </sec>
    <sec id="sec-7">
      <title>Alternatives of Compound Options</title>
      <p>Most product configurators have categories with alternative options
(i.e., options of which exactly one or at most one option can be
chosen). If there are dependencies between these categories, one
strategy is to build pairs of the corresponding options that are wanted.
Those alternative pairs are then grouped under one compound
category. Pairs not allowed by dependencies are simply omitted.</p>
      <p>For instance, in the ThinkPad configurator, dependencies between
the categories Microsoft Office and Adobe Acrobat are handled this
way. Office is available as 365 Home, 365 Personal, Home and
Student, Home and Business, and Professional. However, what is
available does also depend on whether Acrobat is selected: Professional
requires Acrobat and Acrobat requires Home and Business or
Professional. Instead of presenting six options for Office and two options
for Acrobat with two dependencies, there are seven compound
options without dependencies, as illustrated in Figure 4.</p>
      <p>BMW uses compound options for dependencies between 23
engines and three gearbox options. Out of 69 combinatorial
combinations, only 42 are considered valid. In particular, four engines are
not available with manual gearbox and engines are either available
with Automatic Gearbox or with Sport Automatic Transmission (i.e.,
4 + (23 4) 2 = 42). A possible reason for compound options
besides eliminating constraints is that the choice of engine and gearbox
affects the acceleration, consumption, CO2 emission, and pricing,
which are properties shown next to each option (cf. Figure 5).</p>
      <p>Compound options can essentially be used to get rid of all
dependencies, which results in an enumeration of all products. Microsoft
gives customers both options, a configurator and a simple selection
among pre-configured products as shown in Figure 6. Interestingly,
our comparison of available products revealed that the keyboard is
always pre-configured to be British Qwerty and else only one
products is missing opposed to the product configurator (i.e., Intel Core i5
with 128GB storage). That is, ten configuration options in four
categories have so many constraints that they only result in eight different
products for customers, instead of 32 theoretical combinations.</p>
      <p>Compound options are well-suited if the combined options have
many dependencies and customers can directly explore the few
available combinations. In particular, it would not be feasible to present
all 123 combinations of colors and sizes in the Amazon configurator
(cf. Figure 10). However, compound options also typically lead to the
problem that one original option is represented by several compound
options, which increases the effort for customers when comparing
compound options. Furthermore, compound options may complicate
the handling of dependencies to options in other parts of the
configurator. For instance, a simple rule requiring another option may then
need to be duplicated for all occurrences in compound options.
3.5</p>
    </sec>
    <sec id="sec-8">
      <title>Continuing with Invalid States</title>
      <p>A further strategy to deal with conflicting options is to let users
continue with the configuration process although the current
configuration is invalid. That is, users may even select an option that is not
allowed according to dependencies and already selected options. The
action required by the user to fix the configuration is basically
postponed to a later point during the configuration process.</p>
      <p>As illustrated in Figure 7, Lenovo’s configurators issue a warning
to the user once a conflict was detected. The user can continue with
the configuration process, but the same warning appears after every
selection until the problem is fixed. Unfortunately, the only hint on
how to solve this problem is the written-down dependency along with
one of the interacting configuration options. Thus, the dependency
may happen to be written next to the option whose selection lead to
the problem or next to any other option.</p>
      <p>Letting users continue with invalid states seems to be a flexible
way for configuration, as users can decide when to handle conflicts.
In particular, a user can focus on the most relevant options first and
then solve conflicts with the remaining options. However, especially
problematic is that postponing a fix can result in multiple conflicts,
which are even harder to fix. Furthermore, it seems that tool support
for invalid states is not straightforward, as we have only found this
strategy in the two configurators by Lenovo and Toyota, of which
neither provided useful support.
3.6</p>
    </sec>
    <sec id="sec-9">
      <title>Subsequent Configuration Steps</title>
      <p>Configurators with many options often split the decision process over
several steps. The decisions in one configuration step are confirmed
before continuing with subsequent steps. While configurators
typically allow to go back and forth in those steps, the configuration
decisions of subsequent configuration steps may be reverted in this
process. In some configurators, certain steps can even be skipped based
on the configuration in prior steps.</p>
      <p>For instance, the BMW configurator consists of seven steps and
depending on the choice of engine, a further additional step called
Model Variants may appear (cf. Figure 8). However, in the studied
configurator, the additional step was only used to show an inclusive
option called Sport package, which cannot be deselected.</p>
      <p>Configuration steps seem to increase the overview over complex
configuration processes with hundreds of features. Hiding irrelevant
configuration steps is also an interesting concept to deal with
constraints. Nevertheless, hiding certain steps may also be a challenge
for users that are actually looking for a given step and do not
understand why it is not visible. In addition, users could be unaware
of configuration steps that would be helpful in their decision making
process. Furthermore, although guiding the user and hiding options,
configuration steps do not solving the problem of conflicting
configuration options. Even worse, constraints between different
configuration steps could be a particular challenge for users, if they have to
switch back and forth over and over again.
3.7</p>
    </sec>
    <sec id="sec-10">
      <title>Automated Reconfiguration</title>
      <p>Another strategy to deal with constraints is to automatically select
and deselect options after every user decision. In particular,
constraints are used to identify conflicting decisions. Those conflicts are
then resolved automatically, if possible.</p>
      <p>Automated reconfiguration is applied by the BMW configurator. If
already chosen configuration options are affected by the current
decision, a configuration assistant is shown to the user summarizing the
automated changes (cf. Figure 9). There is also an automatic mode
in which the configuration assistant will not inform users after every
reconfiguration. Unfortunately, we were not able to provoke a
situation in which the configurator obviously has to chose one of several
alternative fixes. Such a situation would be interesting to see whether
some kind of user interaction is supported in this process.</p>
      <p>In the Amazon configurator, we can distinguish available from
unavailable combinations visually, as illustrated in Figure 10.
Nevertheless, we can select also grayed-out options, such that a conflicting
state arises. How the conflict is solved dependent on the latest
decision. If a new size is selected that is not available for the current
color selection, then another color that is still available is
automatically selected. The selection appeared to be arbitrary, but it could
be based on a recommender knowing more common combinations
(e.g., red for smaller sizes and blue for larger ones). Interestingly, if
a new color is selected the configurator does not automatically select
another size, but instead resets the size selection. A reason for this
different strategy might be that people can come to terms with
another color, but typically are not willing to choose a t-shirt in a great
color with a size that does not fit them.</p>
      <p>One advantage of automated reconfiguration is that conflicts are
directly solved when they occur. Furthermore, users may have the
chance to confirm or abort the current decision propagation.
However, an issue with this strategy is that reverting decisions that
triggered automated reconfiguration may not revert the automatically
introduced changes. This behavior can be confusing to the user and can
even introduce further conflicts. Furthermore, automated
reconfiguration can interact with other applied strategies in undesired ways,
which can again lead to the introduction of conflicts.
3.8</p>
    </sec>
    <sec id="sec-11">
      <title>Interactive Resolution of Conflicts</title>
      <p>Once a conflict occurred, it can either be resolved automatically, as
discussed in Section 3.7, or with user interaction. With interactive
resolution, we refer to a process in which the configurator informs
users about conflicts in the current configuration and guides them to
a resolution by proposing possible changes. Thus, interactive
resolution is especially helpful if there are several meaningful resolutions
to a conflict and the users input is required.</p>
      <p>We experienced interactive resolution of conflicts in the
configurators by Toyota and by HP Velotechnik. For Toyota, we identified
a dependency between car colors and wheels. Black wheels are only
available for three of six car colors, namely White Flash, Electro
Gray, and Bold Black. Only when of those colors is selected, black
wheels are visible (cf. hiding invalid combinations). However, when
black wheels are selected, we can still choose one of the three
incompatible car colors, as those are not hidden. In that case, the
configurator selects the incompatible car color and shows the dialog for
wheels in which the previously chosen wheel is hidden. If users
select another wheel, the conflict is resolved. However, we experienced
bugs when users decide to cancel the interactive process. We were
actually able to configure a car with a wrong price computation (cf.
Figure 11). If we continue and try to actually buy that car, the reason
for the wrong computation becomes clear; the configurator
automatically selected two wheel types, for which eight instead of four wheels
are charged in terms of costs (cf. Figure 12). When we tried to fix the
conflict manually, we were not even able to open the configuration
dialog for wheels anymore. However, the Toyota configurator has a
rather unique feature allowing users to undo and even redo previous
decisions by going back and forth in the browser.</p>
      <p>For HP Velotechnik, interactive resolution is applied to all
conflicts that occur between any pair of two categories. When opening
the configurator, each category is set to the first option, whereas
options are sorted in alphabetical order. When changing the tires to
Schwalbe Kojak for the default configuration, a dialog opens asking
the user to select one of two options as a drivetrain (cf. Figure 13).
Although we opened the UK version, there is a hint in German
saying that an unrelated drivetrain (i.e., not available and not selected
as default) has some conflicts with particular brakes. After studying
the configurator for a while, we found out that this hint is always
shown if there is a conflict with drivetrains and the hint is most likely
supposed to help with another conflict. Selecting one of those two
options leads to another conflict which is resolved with the same
technique. While, in this case, there are only two interactive
resolution steps due to one selection, we experienced cascades of up-to
five conflicts. With such cascades, there is a considerable burden on
the user to understand why all those changes need be done.
Configuration is especially challenging as users easily get lost which options
were the default, which did they chose on purpose, and which options
had to be selected in the interactive conflict resolution.</p>
      <p>The advantage of resolving conflicts interactively is that the user is
essentially making decisions on conflicting options. Hence, opposed
to automatic reconfiguration users have the control over conflict
resolution and opposed to the hiding of invalid combinations they can
actually see and decide on all options. The downside is, however,
that users have to take decisions in response to a conflict directly. In
particular, this may lead to a frustrating cascade of numerous
conflict resolutions, which may even need to be reverted, if the user is
not satisfied with the options in one conflict resolutions. A postponed
interactive resolution would probably need some kind of support for
continuing with invalid states (cf. Section 3.5), which seems to be
hard to be implemented properly.
In the previous sections, we discussed each identified strategy
independently. However, in practice, none of the studied configurators
applies only a single strategy. Instead, we found that configurators
apply between two and eight strategies, as illustrated in Table 1. HP
Velotechnik, Ebay, and Amazon only apply two or three strategies.
The configurators of Ebay and Amazon did only come with a few
categories of alternative options. The most strategies are applied in
the car configurators by Toyota and BMW, which also comprise the
largest number of options and categories. The configurator by Toyota
even applies all eight strategies.</p>
      <p>We can further distinguish between strategies that are applied
globally to all options or only locally to some options. Automatic
deselection of alternative options is not only applied in all
configurators, but also for all categories with alternative options. Similarly,
configuration steps and default options have been applied only
globally. In contrast, all other strategies have typically be applied locally
to some options in the configurator. One exception is that the Ebay
configurator uses hiding of invalid combinations in all parts, but there
were only two categories of alternative features. Other exceptions
tf
o n
BMW taoyoT leoPHV enovoL icM azAm abyE
s o
o
r</p>
      <sec id="sec-11-1">
        <title>Strategy</title>
      </sec>
      <sec id="sec-11-2">
        <title>Automatic Deselection in Alternatives y y y y y y y</title>
        <p>Default Configuration y y y y y/n y n
Hiding Invalid Combinations n y n n n n y</p>
        <p>Compound Options y y n y y n n</p>
        <p>Continuing with Invalid States n y n y n n n
Subsequent Configuration Steps y y n n n n n</p>
        <p>Automated Reconfiguration y y n n y/n y n
Interactive Resolution of Conflicts n y y n n n n
are that BMW applies automated reconfiguration globally and HP
Velotechnik uses interactive resolution for all categories.</p>
        <p>The configurator by Microsoft is especially interesting as there
are multiple links to configurators for the same product, which apply
different strategies. For instance, there is a version of a configurator
using default configurations and another version that opens without
any decisions. Also, we experienced that the Microsoft configurator
stopped using automated reconfiguration in April or May 2018.</p>
        <p>The strategies that we identified can be further classified into
strategies avoiding all or at least some conflicts and those strategies
resolving conflicts. Default configurations, hiding invalid
combinations, compound options, and configuration steps help to avoid
conflicts. All other strategies let users make conflicting decisions and
later solve them automatically or interactively.
4</p>
      </sec>
    </sec>
    <sec id="sec-12">
      <title>Related Work</title>
      <p>
        There exists a number of previous studies comparing web-based
configurators [
        <xref ref-type="bibr" rid="ref1 ref12 ref14">1, 14, 12</xref>
        ]. Each of these studies considers different
aspects of these tools. For instance, Streichsbier et al. focuses on the
differing user interface of configurators for different domains [
        <xref ref-type="bibr" rid="ref14">14</xref>
        ].
They describe differences and similarities in the design of
configurators, such as navigation buttons, product images, and prices displays,
across multiple domains. Abbasi et al. compare 111 different
configuration tools regarding the process of configuration and the internal
behavior of configurators [
        <xref ref-type="bibr" rid="ref1">1</xref>
        ]. In their work, they consider the visual
representation of options and constraints and the corresponding
semantics behind it, constraint syntax and handling, and control flow
of the configuration process. Pil and Holweg studied configurators
of ten car manufacturers [
        <xref ref-type="bibr" rid="ref12">12</xref>
        ]. In addition to the external variety
provided by the configurator, they also studied the internal variety by
surveying the manufacturers. In contrast to these prior studies, we
explicitly focus on the underlying strategies for handling conflicts
and categorize these further.
      </p>
      <p>
        Other work, such as Herrmann et al., Haag et al., and Hubaux et al.
have looked at specific strategies that we cover in our paper [
        <xref ref-type="bibr" rid="ref6 ref7 ref9">6, 7, 9</xref>
        ].
In their paper, Herrmann et al. investigate the effects of starting the
configuration process with a default configuration, which can be
adapted by the user [
        <xref ref-type="bibr" rid="ref7">7</xref>
        ]. Although we do not investigate user
behavior when using a particular conflict handling strategies, we aim
to provide a broader view of currently used strategies on the web.
Haag et al. consider different techniques for finding explanations
in conflicting configurations [
        <xref ref-type="bibr" rid="ref6">6</xref>
        ]. This is corresponds to our
identified strategy of resolving conflicts interactively. Amongst others,
Hubaux et al. investigate which conflict handling strategies are being
used in the two product lines Linux and eCos [
        <xref ref-type="bibr" rid="ref9">9</xref>
        ]. Although they are
more focused on the user perspective, they find that the Linux’
configuration tool uses a mix of different strategies, such as automated
reconfiguration, hiding invalid combinations, and resolving conflicts
interactively and that eCos also allows to continue with invalid states.
      </p>
      <p>
        Product configuration is very similar to software configuration [
        <xref ref-type="bibr" rid="ref8">8</xref>
        ],
for which many tools exist [
        <xref ref-type="bibr" rid="ref11 ref2">2, 11</xref>
        ]. Berger et al. conducted a
survey among practitioners and researchers with industrial experience
on software product-line engineering and variability modeling [
        <xref ref-type="bibr" rid="ref2">2</xref>
        ].
Amongst others, they present data about the distribution of
modeling and configuration tools that are used in practice and the scale of
real-world variability models, regarding number of configuration
options and dependencies. While, in this paper, we are also interested
in dependencies that result from variability modeling, we foremost
consider the strategies to enforce these dependencies during the
configuration process.
5
      </p>
    </sec>
    <sec id="sec-13">
      <title>Conclusion and Future Work</title>
      <p>Product configurators are essential for the vision of mass
customization. Customers use configurators to identify options and their valid
combinations. A configurator is often the main source of knowledge
being queried in the decision making process. We studied a corpus of
seven configurators to understand how they handle dependencies
between configuration options. Those dependencies can easily lead to
decisions by a customer that are in conflict with each other. We argue
that how a configurator supports users in avoiding or fixing conflicts
is crucial for the success of mass customization.</p>
      <p>In our study, we identified eight strategies to handle dependencies.
Half of those strategies aim to prevent conflicting decisions whereas
the rest help customers to resolve conflicts automatically or with their
interaction. Most strategies are applied by several but not all
configurators, whereas one strategy is applied by all configurators and
all other strategies are applied by at least two configurators.
Similarly, each studied configurator applies at least two strategies and
we also found a configurator that applies all eight strategies. It seems
that smaller configurators with fewer options and fewer dependencies
tend to use fewer strategies and larger configurators apply numerous
strategies.</p>
      <p>While we discussed advantages and disadvantages of all strategies,
it is an open research question which combinations of those strategies
are best in which situations. Furthermore, it would be interesting to
investigate whether there are further strategies and how each strategy
can be supported with the state-of-the-art techniques for the
implementation of configurators. In any case, studying further real-world
configurators may give more insights.</p>
    </sec>
  </body>
  <back>
    <ref-list>
      <ref id="ref1">
        <mixed-citation>
          [1]
          <string-name>
            <given-names>Ebrahim</given-names>
            <surname>Khalil</surname>
          </string-name>
          <string-name>
            <surname>Abbasi</surname>
          </string-name>
          , Arnaud Hubaux, Mathieu Acher, Quentin Boucher, and Patrick Heymans, '
          <article-title>The Anatomy of a Sales Configurator: An Empirical Study of 111 Cases'</article-title>
          ,
          <source>in Proc. Int'l Conf. Advanced Information Systems Engineering (CAiSE)</source>
          , pp.
          <fpage>162</fpage>
          -
          <lpage>177</lpage>
          , Berlin, Heidelberg, (
          <year>2013</year>
          ). Springer.
        </mixed-citation>
      </ref>
      <ref id="ref2">
        <mixed-citation>
          [2]
          <string-name>
            <given-names>Thorsten</given-names>
            <surname>Berger</surname>
          </string-name>
          , Ralf Rublack, Divya Nair,
          <string-name>
            <surname>Joanne M. Atlee</surname>
          </string-name>
          , Martin Becker, Krzysztof Czarnecki, and Andrzej Wa˛sowski, '
          <article-title>A Survey of Variability Modeling in Industrial Practice'</article-title>
          ,
          <source>in Proc. Int'l Workshop Variability Modelling of Software-Intensive Systems (VaMoS)</source>
          , pp.
          <volume>7</volume>
          :
          <fpage>1</fpage>
          -
          <issue>7</issue>
          :
          <fpage>8</fpage>
          , New York, NY, USA, (
          <year>2013</year>
          ). ACM.
        </mixed-citation>
      </ref>
      <ref id="ref3">
        <mixed-citation>
          [3]
          <string-name>
            <surname>Stanley</surname>
            <given-names>M.</given-names>
          </string-name>
          <string-name>
            <surname>Davis</surname>
          </string-name>
          , Future Perfect,
          <year>1987</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref4">
        <mixed-citation>
          [4]
          <string-name>
            <given-names>Rebecca</given-names>
            <surname>Duray</surname>
          </string-name>
          , Peter T. Ward, Glenn W. Milligan, and William L. Berry, 'Approaches to Mass Customization:
          <article-title>Configurations and Empirical Validation'</article-title>
          ,
          <source>J. Operations Management (JOM)</source>
          ,
          <volume>18</volume>
          (
          <issue>6</issue>
          ),
          <fpage>605</fpage>
          -
          <lpage>625</lpage>
          , (
          <year>2000</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref5">
        <mixed-citation>
          [5]
          <string-name>
            <given-names>Cipriano</given-names>
            <surname>Forza</surname>
          </string-name>
          and
          <string-name>
            <given-names>Fabrizio</given-names>
            <surname>Salvador</surname>
          </string-name>
          ,
          <article-title>Product Information Management for Mass Customization: Connecting Customer, Front-Office and Back-Office for Fast and Efficient Customization</article-title>
          , Palgrave Macmillan, New York, NY, USA,
          <year>2006</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref6">
        <mixed-citation>
          [6]
          <string-name>
            <surname>Albert</surname>
            <given-names>Haag</given-names>
          </string-name>
          ,
          <string-name>
            <given-names>Ulrich</given-names>
            <surname>Junker</surname>
          </string-name>
          , and
          <string-name>
            <surname>Barry O'Sullivan</surname>
          </string-name>
          , '
          <article-title>A survey of explanation techniques for configurators'</article-title>
          ,
          <source>in Proceedings of ECAI-2006 Workshop on Configuration</source>
          , volume
          <volume>41</volume>
          , p.
          <fpage>44</fpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref7">
        <mixed-citation>
          [7]
          <string-name>
            <given-names>Andreas</given-names>
            <surname>Herrmann</surname>
          </string-name>
          , Daniel G. Goldstein, Rupert Stadler,
          <string-name>
            <surname>Jan R. Landwehr</surname>
          </string-name>
          , Mark Heitmann, Reto Hofstetter, and Frank Huber, '
          <article-title>The Effect of Default Options on Choice-Evidence from Online Product Configurators'</article-title>
          ,
          <source>J. Retailing and Consumer Services</source>
          ,
          <volume>18</volume>
          (
          <issue>6</issue>
          ),
          <fpage>483</fpage>
          -
          <lpage>491</lpage>
          , (
          <year>2011</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref8">
        <mixed-citation>
          [8]
          <string-name>
            <given-names>Arnaud</given-names>
            <surname>Hubaux</surname>
          </string-name>
          , Dietmar Jannach, Conrad Drescher, Leonardo Murta, Tomi Männistö, Krzysztof Czarnecki, Patrick Heymans,
          <string-name>
            <given-names>Tien N.</given-names>
            <surname>Nguyen</surname>
          </string-name>
          , and Markus Zanker, '
          <article-title>Unifying Software and Product Configuration: A Research Roadmap'</article-title>
          ,
          <source>in Proc. Configuration Workshop (ConfWS)</source>
          , pp.
          <fpage>31</fpage>
          -
          <lpage>35</lpage>
          , (
          <year>August 2012</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref9">
        <mixed-citation>
          [9]
          <string-name>
            <given-names>Arnaud</given-names>
            <surname>Hubaux</surname>
          </string-name>
          , Yingfei Xiong, and Krzysztof Czarnecki, '
          <article-title>A User Survey of Configuration Challenges in Linux and eCos'</article-title>
          ,
          <source>in Proc. Int'l Workshop Variability Modelling of Software-Intensive Systems (VaMoS)</source>
          , pp.
          <fpage>149</fpage>
          -
          <lpage>155</lpage>
          , New York, NY, USA, (
          <year>2012</year>
          ). ACM.
        </mixed-citation>
      </ref>
      <ref id="ref10">
        <mixed-citation>
          [10]
          <string-name>
            <given-names>Cynthia</given-names>
            <surname>Huffman</surname>
          </string-name>
          and Barbara E. Kahn, 'Variety for Sale: Mass Customization or Mass Confusion?',
          <string-name>
            <given-names>J.</given-names>
            <surname>Retailing</surname>
          </string-name>
          ,
          <volume>74</volume>
          (
          <issue>4</issue>
          ),
          <fpage>491</fpage>
          -
          <lpage>513</lpage>
          , (
          <year>1998</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref11">
        <mixed-citation>
          [11]
          <string-name>
            <surname>Jens</surname>
            <given-names>Meinicke</given-names>
          </string-name>
          , Thomas Thüm, Reimar Schröter, Fabian Benduhn, and Gunter Saake, '
          <article-title>An Overview on Analysis Tools for Software Product Lines'</article-title>
          ,
          <source>in Proc. Workshop Software Product Line Analysis Tools (SPLat)</source>
          , pp.
          <fpage>94</fpage>
          -
          <lpage>101</lpage>
          , New York, NY, USA, (
          <year>2014</year>
          ). ACM.
        </mixed-citation>
      </ref>
      <ref id="ref12">
        <mixed-citation>
          [12]
          <string-name>
            <surname>Frits</surname>
            <given-names>K.</given-names>
          </string-name>
          <string-name>
            <surname>Pil</surname>
          </string-name>
          and Matthias Holweg, 'Linking Product Variety to OrderFulfillment Strategies',
          <source>Interfaces</source>
          ,
          <volume>34</volume>
          (
          <issue>5</issue>
          ),
          <fpage>394</fpage>
          -
          <lpage>403</lpage>
          , (
          <year>October 2004</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref13">
        <mixed-citation>
          [13]
          <string-name>
            <given-names>Giovani</given-names>
            <surname>Da</surname>
          </string-name>
          <string-name>
            <given-names>Silveira</given-names>
            , Denis Borenstein, and Flávio S. Fogliatto, 'Mass Customization: Literature Review and Research Directions',
            <surname>Int'l J. Production</surname>
          </string-name>
          <string-name>
            <surname>Economics</surname>
          </string-name>
          ,
          <volume>72</volume>
          (
          <issue>1</issue>
          ),
          <fpage>1</fpage>
          -
          <lpage>13</lpage>
          , (
          <year>2001</year>
          ).
        </mixed-citation>
      </ref>
      <ref id="ref14">
        <mixed-citation>
          [14]
          <string-name>
            <given-names>Clarissa</given-names>
            <surname>Streichsbier</surname>
          </string-name>
          , Paul Blazek, Fabian Faltin, and Wolfgang Frühwirt, 'Are
          <string-name>
            <surname>De-Facto Standards</surname>
          </string-name>
          <article-title>a Useful Guide for Designing HumanComputer Interaction Processes? The Case of User Interface Design for Web Based B2C Product Configurators'</article-title>
          ,
          <source>in Proc. Hawaii Int'l Conf. System Sciences (HICSS)</source>
          , pp.
          <fpage>1</fpage>
          -
          <lpage>7</lpage>
          , Washington, DC, USA, (
          <year>2009</year>
          ). IEEE.
        </mixed-citation>
      </ref>
      <ref id="ref15">
        <mixed-citation>
          [15]
          <string-name>
            <given-names>Juha</given-names>
            <surname>Tiihonen</surname>
          </string-name>
          and Alexander Felfernig, '
          <article-title>An Introduction to Personalization and Mass Customization'</article-title>
          ,
          <string-name>
            <given-names>J.</given-names>
            <surname>Intell</surname>
          </string-name>
          . Inf. Syst.,
          <volume>49</volume>
          (
          <issue>1</issue>
          ),
          <fpage>1</fpage>
          -
          <lpage>7</lpage>
          , (
          <year>August 2017</year>
          ).
        </mixed-citation>
      </ref>
    </ref-list>
  </back>
</article>