Enhancing Returns Management in Fashion E-Commerce: Industry
Insights on AI-Based Prediction and Recommendation Systems
Soeren Gry
a
, Marie Niederlaender
b
and Dirk Werth
c
August-Wilhelm Scheer Institut, Uni Campus D 5 1, Saarbr
¨
ucken, Germany
fi
Keywords:
Returns Prediction, Returns Prevention, Survey Results, Expert Interviews, Fashion E-Commerce,
Recommendation System, Machine Learning, Sustainable Return Management, Sustainable Supply Chain.
Abstract:
The fashion industry is one of the most problematic sectors in terms of sustainability. The fashion e-commerce
sector is experiencing a surge in sales, which is leading to a significant increase in returns. This, in turn, is
placing a considerable burden on the environment. High transport volumes or even the destruction of garments
through returns pose major environmental and also economic problems. This study is based on a survey and
expert interviews with decision-makers from the fashion industry. It provides indications of how an AI-based
prediction and recommendation system could be used to avoid returns and manage them in an ecologically
and economically sensible way. On the one hand, use cases are discussed that can be applied in the webshop
system before the customer places an order, and on the other hand, ways are shown how returns predictions can
support planning in the reverse logistics network.
1 INTRODUCTION
E-commerce grew rapidly during the coronavirus pan-
demic. The share of e-commerce in total retail sales
increased from 15% in 2019 to 22% in 2022 (Mor-
gan Stanley, 2022). This growth will continue after
the coronavirus pandemic. Annual growth of 9.91%
is expected for the years 2024 (C3,334.00 billion) to
2028 ( C4,865.00 billion) (Statista Market Insights,
2024). Growing e-commerce sales are accompanied
by increasing transport volumes and higher volumes of
returns. As a result, e-commerce is focusing more and
more on reducing returns and managing them more sus-
tainably. This can be done, for example, by using size
finders which help customers find a garment that fits
their individual body shape. In fashion e-commerce,
the fitting room is often moved from the store to the
consumer’s home to overcome a major drawback of
e-commerce compared to bricks-and-mortar retail: the
customer wants to see, touch and try on the product.
It is therefore difficult to eliminate returns completely
(Asdecker and Karl, 2018; Lohmeier, 2024).
The fashion industry accounts for most of the re-
turns in the e-commerce sector. Of the 1.3 billion items
a
https://orcid.org/0000-0002-4441-0517
b
https://orcid.org/0009-0008-1935-821X
c
https://orcid.org/0000-0003-2115-6955
returned in Germany in 2021, 91% were clothing and
footwear. Comparable patterns of consumer behaviour
can be observed at the European level (Forschungs-
gruppe Retourenmanagement, 2022). The issue of
increasing product returns is adding to the pressure on
an industry that has long been criticised for its poor
environmental performance. This is the starting point
for this research project. To tackle the problem of high
returns, manufacturers and retailers have the opportu-
nity to work on preventing returns through preventive
measures such as sizefinders, or to improve the en-
vironmental and economic impact of returns through
reactive measures such as adjustments to returns lo-
gistics or a suitable second life plan for the returned
garment (Deges, 2021; Gry et al., 2023). As studies
have already demonstrated, AI-based returns predic-
tions at the shopping basket level afford manufacturers
and retailers a multitude of options: These include
both 1) preventive measures to avoid returns and 2)
reactive measures in the form of adjustments to the
reverse logistics network or the second life cycle of the
garment (Gry et al., 2023). The corresponding poten-
tial applications of an AI-based prediction and recom-
mendation system are discussed in expert interviews
with decision makers from the fashion e-commerce
sector. The findings of these expert interviews will be
integrated into the design of an AI-based prediction
and recommendation system. This will provide valu-
66
Gry, S., Niederlaender, M. and Werth, D.
Enhancing Returns Management in Fashion E-Commerce: Industry Insights on AI-Based Prediction and Recommendation Systems.
DOI: 10.5220/0012759900003764
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 21st International Conference on Smart Business Technologies (ICSBT 2024), pages 66-73
ISBN: 978-989-758-710-8; ISSN: 2184-772X
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.

able insights into the optimal points in the ordering
or returns process, and thus in which system (e.g. in
the webshop system or in the ERP or PDM system)
AI-based predictions can be most beneficial. Previ-
ous research has shown that AI-based predictions have
advantages over mathematical models or simple data
mining models when it comes to predicting returns, for
example (Gry et al., 2023; Niederlaender et al., 2024).
In particular, Niederlaender et al. (2024) have shown
that machine learning approaches deliver promising
results in predicting returns in fashion e-commerce. In
light of the aforementioned considerations, it is simi-
larly assumed in this paper that AI-based predictions
will be employed.
The structure of the paper is as follows. First, a
brief literature review presents the factors known to
influence the likelihood of returns when customers
place orders. Secondly, the results of a survey on
returns management among fashion retailers and man-
ufacturers with e-commerce activities are presented.
Then, key findings from expert interviews with fash-
ion e-commerce decision makers regarding the use of
AI-based predictions to avoid and manage returns are
provided. Finally, the implications of the survey and
expert interviews for the development and possible sys-
tem integration of AI-based prediction and recommen-
dation systems in the context of returns (avoidance)
are discussed. The paper concludes with a summary of
the findings, identified research gaps and an outlook.
2 METHODOLOGY
In the first part a literature review reflects the current
state of research on the factors that influence consumer
return behaviour. This knowledge can be used to sup-
port the selection of relevant variables and the cre-
ation of features in the context of machine learning
for AI-based returns predictions and to create a basic
understanding of the relationships between returns and
consumer behaviour (Niederlaender et al., 2024). This
knowledge also serves as a basis for discussion in the
following expert interviews.
To this end, a literature review was conducted, fo-
cusing on a period between 2018 and 2023, and was
carried out in November and December 2023. Springer
Link, ScienceDirect, Wiley Online Library and Google
Scholar were used, supplemented by snowballing tech-
niques to include relevant older sources outside the pri-
mary time window. The following keywords were used
in combination with the keywords Fashion, Apparel
and E-Commerce: Consumer Return Behaviour, Prod-
uct Return Prediction, Consumer Return Behaviour,
Prediction of Consumer Returns.
In order to subsequently contribute to answering
the question of how AI-based predictions can be used
in practice by manufacturers and retailers in fashion
e-commerce, a survey was conducted among man-
ufacturers and retailers on the one hand and expert
interviews with decision-makers on the other. The re-
sults and implications of which were analysed with
regard to an AI-based recommendation system based
on returns predictions.
3 LITERATURE REVIEW
The following brief literature review serves to high-
light the relationships between possible independent
variables, such as customer or product-related data
or payment methods, and the dependent variable of
the probability of returns. This was done in order to
build up a sound knowledge base on the occurrence
of returns and to be able to use this knowledge for
the expert interviews and for the development of the
AI-based recommendation system.
Product and Customer Related Data. Asdecker
et al. (2017) found that the most important factor in de-
termining the likelihood of returns is historical returns
data at the item and customer level. If a particular item
has a history of frequent returns, this, together with
historical returns behaviour on an individual customer
basis, is the most important variable in determining
the likelihood of returns.
Shipping Costs. Studies have shown that eliminat-
ing shipping costs as part of promotions encourages
consumers to order more goods for which they feel
uncertain about the purchase decision (Saarij
¨
arvi et al.,
2017). This in turn leads to a higher rate of returns
(Shehu et al., 2020). Many online retailers waive re-
turn fees above a certain turnover threshold. This has
also been shown to increase return rates, as Lepthien
and Clement (2019) found in a study conducted with a
streetwear and sportswear retailer.
Payment Methods, Price and Promotions. In their
study, Yan and Cao (2017) found that payment meth-
ods that exclude prepayment lead to higher return rates
than payment methods that require prepayment for
goods. For example, using data from an online retailer
of shoes, clothing and accessories, they showed that
credit card payment leads to a ”buy-now-pay-later”
attitude, which encourages impulsive buying and in-
creases the return rate. Customers also perceive it as a
Enhancing Returns Management in Fashion E-Commerce: Industry Insights on AI-Based Prediction and Recommendation Systems
67

lower risk if they do not have to pay in advance (Su-
tinen et al., 2022). Paying in advance is more likely
to be associated with a deliberate purchase decision,
which is reflected in lower return rates. Similar results
were found by Asdecker et al. (2017) and Makkonen
et al. (2021). Sahoo et al. (2018) show that the price
of the garments sold also influences the likelihood of
returns. They found that more expensive garments
are less likely to be returned than cheaper ones. One
reason for this could be that more mental effort is in-
vested in the purchase decision, which is reflected in
lower return rates. On the other hand, coupons lead to
customers being urged to make what they consider to
be riskier purchasing decisions, which in turn result in
higher return rates (Asdecker et al., 2017).
4 SURVEY
In the context of the returns problem described above,
a survey was created which was sent to contacts from
the fashion industry in Germany via LinkedIn and
newsletters. The aim of the survey is to find out the
status quo of the companies with regard to their online
business and their returns volume and to find out about
strategies that have already been applied to improve
the handling of returns.
4.1 Methodology and Data Set
The survey was conducted in German language and
consisted of 30 questions in total. The survey con-
sisted of a total of 20 checkbox questions with pre-
defined answers and the option of manually entering
additional answers. Cocurring manual responses were
summarized as part of the analysis. The 20 checkbox
questions included:
1. The perspective of the respondent: retailer or
manufacturer/brand?
2. − 4. Are the products offered via a marketplace
such as Zalando, Amazon or About You? Are the prod-
ucts exclusively offered via a marketplace? Which
marketplaces are used for selling the products? Op-
tions include Amazon, Zalando, AboutYou, Otto and
manual input fields
5. Which fulfillment components are utilized via
the marketplaces used? Options include storage, pick-
ing, shipping, processing of returns and manual input
options
6. The product categories offered for sale at the
market place or own store
7. − 10. Does the shipment via the shop or market-
place include a returns label? Does the shop or mar-
ketplace collect data on the specific return reasons?
Which specific return reasons are collected? Are re-
turns reused by the shop or the marketplaces that serve
as a sales platform (e.g. in the sense of reshipment)?
11. What specific reuse options are employed by
the shop or the marketplaces?
12. Are customers charged for the return shipment
of their returns?
13. − 15 . Does the store or marketplace used de-
termine return rates by product category? Are the
costs incurred in the context of a return (shipping, pro-
cessing, refurbishment, etc.) being calculated? Is a
shop-internal or an existing analytics solution within
the framework of the marketplace used in the returns
context?
16. − 18. Are return probabilities determined de-
pending on the contents of the shopping cart? Is a
returns history kept on a customer basis with items
returned in the past? Are returns probabilities deter-
mined on a customer basis?
19. Which of the following evaluation options are
used in the context of returns management? Correla-
tion between fit forms, correlation between customer
group and probability of returns, correlation between
product group and probability of returns, or no further
evaluations
20. Are there plans to address the issue of re-
turns processes through initiatives, projects or process
changes in your company?
In addition, 4 questions were asked to estimate the
returns rate and the reuse rate of various clothing cat-
egories, the customer friendliness of the particular
returns process currently in use, and the share of on-
line business in total sales. Remaining 6 questions
were posed with an input field to enter missing prod-
uct categories or to enter values like the number of
employees and year of foundation, time frame of the
right of return, the return fees the customer is being
charged, the costs of a single return for the company,
or to enter specific plans to improve the returns process.
The answers were analysed using histogram plots, and
both the percentage distribution of responses and the
number of participants were taken into account. In the
case of estimation questions on the return rate and re-
use rate of various product categories, the aggregated
percentage of all participant responses was evaluated
for each category.
4.2 Main Results
60 people from companies in the fashion industry took
part in the survey. The response rate varies from ques-
tion to question and is therefore stated for each of the
mentioned results in terms of the number n of par-
ticipants on that particular question. Although n is
ICSBT 2024 - 21st International Conference on Smart Business Technologies
68

too low in some cases to make a statistically signif-
icant statement, the authors decided to present these
subjective estimations by experts from the industry as
a qualitative insight and as preliminary findings that
draw attention to possibly emerging trends.
The participants in the survey are working in fash-
ion companies with a large share of online business
in total sales ( n = 11 ): 33% of respondents state the
share of online business to be above 90%. The aver-
age share of online business over all respondents is
80%. The average number of employees is around 200
( n = 10 ).The participating companies range from long-
established companies to young companies, founded
between 1916 and 2015, with an average company age
of 50 years.
62% of respondents belong to a manufacturer or
brand, while 38% are retailers ( n = 34 ). 76% offer
their products via a marketplace ( n = 34 ), while only
4% are exclusively using marketplaces to sell their
products ( n = 26 ). The most frequently named market-
places are given by Amazon (24%), Zalando (19%),
Otto (18%), AboutYou (10%), and Breuninger (4%)
( n = 20 ). All of the respondents working with a mar-
ketplace utilize storage and shipping as part of the
fulfillment offered by the marketplace, while 80% of
them also utilize picking and processing of returns
( n = 10 ). The mentioned product categories sold via
the online shops and marketplaces include, but are not
limited to the categories mentioned in Figure 1 a) and
b). Even though including a return label in the pack-
age may be a restriction on the further processing of
the return, 60% of shops or marketplaces decided for
this option ( n = 20 ), which might be due to improved
convenience for the customer in the returns process or
because options to route the package flexibly is pos-
sible even with a given returns address on the label.
This is in accordance with the fact that 25% of respon-
dents ( n = 20 ) rate their returns process as extremely
customer-friendly. The average rating of customer
friendliness is at 3.5 out of 5, where 5 marks an ex-
tremely customer friendly process. The large majority
(80%, n = 20 ) of shops and marketplaces also collect
the reason for return. Return reasons collected and
their share over all answers are the following ( n = 15 ):
Product too large or too small: 24%; product is defec-
tive: 23%; product is different than described: 16%;
bad purchase: 16%; poor quality: 13%; better offer
discovered: 5%; price: 2%; wrong item received: 2%.
11% of respondents state that returns are not being
reused by the shop or the marketplaces that serve as
a sales platform in the sense of reshipment ( n = 18 ).
Types of reuse and processing of returned items and
their share over all answers are the following ( n = 16 ):
Reshipment: 55%; Recycling of goods: 17%; Disposal
of goods: 14%; Repair: 3%; Second life in own stores:
3%; Secondary marketing: 3%; B2B warehouse mar-
keting: 3%.
The aggregated estimates of reuse rates for dif-
ferent clothing categories are shown in Figure 1 a)
( n = 10 ). According to these results the estimated
amount of non-reused returns can have a high impact
on the resource efficiency of companies, depending
on the amount of products sold, which should not be
ignored in the future strategy of the companies. Also,
it becomes clear that the estimates fall below the mea-
sured values of Forschungsgruppe Retourenmanage-
ment (2022). The time frame for the right of return
is between 14 and 60 days for 79% of respondents
( n = 14 ), where the average time frame is at about 60
days, which may be due to the comparably long time
windows offered by marketplaces, exceeding the 100
day mark. In Germany, it is common practice for fash-
ion e-commerce companies, especially marketplaces,
not to charge return fees. The findings are consistent
with this: 94 % of respondents ( n = 16 ) do not charge
return fees. If return fees are being charged, it ranges
around the standard shipping costs of a parcel, which
at the point of writing this paper spans around C4-5.
The survey respondent charging return fees has quoted
C4.20 as the return fee.
Based on question 13, the vast majority of respon-
dents (69% ( n = 16 )) do not collect return rates by
product category. The aggregated estimates of the re-
turn rates for different clothing categories are shown
in Figure 1 b) ( n = 3 ). However, the return rate is
highly dependent on the specific company and their
unique operation setting. Half of respondents do not
calculate the costs incurred in the context of a return
in the sense of shipping, processing, refurbishment
and other processing steps ( n = 16 ), while 12% do
so exclusively via the marketplace. The remaining
38% also calculate the costs of a return in the scope
of their own online business. The stated costs of a
single return span from C6 to C20 with an average
value of C10, which stresses the point that for many
companies, it may not be economically feasible to
process and reship returns. Most respondents (67%)
are using a shop internal analytics solution for returns,
33% use an existing analytics tool provided by the mar-
ketplace ( n = 10 ). Nevertheless, only 14% ( n = 14 )
state that they determine return probabilities depend-
ing on the contents of the shopping cart, which is an
important indicator for bracketing behaviour, where
customers order a selection of items with the inten-
tion of only keeping a subset of them (Bimschleger
et al., 2019). Nevertheless, 64% ( n = 14 ) collect a
customer-dependent returns history, which is an impor-
tant factor for estimating future returns behavior and
Enhancing Returns Management in Fashion E-Commerce: Industry Insights on AI-Based Prediction and Recommendation Systems
69

for determining a returns probability (Niederlaender
et al., 2024). An overview of which kind of evaluation
is being performed is given by the responses on ques-
tion 19 (n = 8):
Correlation between fit forms: 7%; Correlation be-
tween customer group and probability of returns: 7%;
correlation between product group and probability of
returns: 33%; no further evaluations: 53%. Based
on the answers in this paragraph on analytics meth-
ods currently used, we can see a trend towards partial
data aggregation. The answers also suggest that there
is no large focus on further evaluation, the results of
which could potentially be incorporated into current
strategies for actions in the context of avoiding or
processing returns. Based on the responses on ques-
tion 20, 64% ( n = 14 ) plan to address the issue of
returns processes through initiatives, projects or pro-
cess changes, while 29% do not have any plans doing
so. The remaining 7% employ strategies exclusively
via the marketplace. Some of the initiatives planned by
respondents to improve the returns process or decrease
returns are (n = 8):
Better fit guide; flyer in the parcel for a more conscious
online shopping; introduce a return management sys-
tem; more repairs in retail; automated product sales
channel selection controlled by excessive returns; im-
provement of shipping process: Speed, better package
material, including benefits.
5 EXPERT INTERVIEWS
To complement the mainly quantitative results from
the manufacturer and retailer survey, expert interviews
were conducted to provide qualitative insights that can
be used to inform the development of an AI-based
prediction and recommendation system, and as an im-
portant basis for where this system could be integrated
(ERP or PDM system, webshop system).
5.1 Methodology and Companies
Included
The expert interviews were mainly conducted in Jan-
uary and February 2024 with decision makers from
seven fashion retailers with a strong connection to
fashion e-commerce. Four of the seven retailers also
have bricks and mortar stores. This is particularly rele-
vant in the context of adjusting returns logistics based
on AI-based predictions, for example, when a return
should be sent directly from a customer to a particular
store rather than back to the central warehouse. Each
interviewee was presented with the basic idea (Figure
1 c)) of how returns predictions are determined and
what they can be used for, for example.
Using an interview technique, the main principles
of which can be traced back to the so-called Mom Test
by Fitzpatrick (2013), the interviewees were asked
to reflect on the use of AI-based return forecasts in
their respective companies. This open-ended interview
technique was designed to minimise priming. The in-
terviews were all scheduled for a period of 45 minutes.
Further details on the companies and decision makers
interviewed can be found in Table 1. In order to cover
as broad a spectrum as possible, the expert interviews
were conducted with companies of different sizes and
with different product ranges. In the following expla-
nations, the key findings from the expert interviews
regarding the use of AI-based returns prediction in
fashion e-commerce are considered from the perspec-
tive of returns avoidance and the adaptation of reverse
logistics in terms of a sustainable supply chain. The ex-
perts’ comments are also used to illustrate the systems
into which the return forecasts can be integrated.
5.2 Main Results
Avoiding Returns Through Transparency. In the
expert interviews, making the likelihood of returns
transparent to customers during the ordering process
was seen as an interesting application area for AI-
based returns. As five of the seven retailers surveyed
stated that they use a size finder in their webshops
to help customers choose the right garment, they see
a combination of returns prediction and size finder
as particularly attractive. For example, if the system
determines from the contents of the shopping basket
that a particular item has a return probability above a
threshold of a certain percentage, a pop-up in the order
process could recommend the use of the size finder in
the webshop system. Another possible use case from
a practical point of view is the charging of a return
fee if a certain return probability is determined for a
shopping basket. It must be taken into account that cus-
tomers may switch to other sales channels that do not
charge returns fees. As the expert interviews showed,
depending on the estimated likelihood of returns, the
returns fee is particularly suitable for companies that
are exclusively active in online retailing and do not use
other platforms or marketplaces.
Selection of Sales Channels. All companies sur-
veyed consider it extremely important to select distribu-
tion channels for newly launched products on the basis
of return forecasts based on certain product character-
istics. Different sales channels are subject to different
business calculations due to their fee systems, which
erode manufacturers’ margins. This means that selling
ICSBT 2024 - 21st International Conference on Smart Business Technologies
70

Figure 1: a) Aggregated estimate of the average reuse rates of returns for each product category and the mean over all averages.
b) Aggregated estimate of the average return rates for each product category and the mean over all averages. c) Central
document shown to the interview partners.
Table 1: Key data about the field of expertise of the interviewees and the corresponding companies.
Company
Role of the Ex-
pert
Sales of the
Company in
2022
E-commerce
Return Rate
# bricks-and-
mortar stores in
DACH Region
Return Fee
Retailer Data Science C1 billion unknown 13 No
Manufacturer Data Analytics ∼ C100 million unknown 0 No
Retailer E-Commerce ∼ C200 million unknown 200 No
Manufacturer
Managing
Director
unknown 5% 0 Yes
Manufacturer Sustainability ∼ C30 million unknown 0 Yes
Manufacturer E-Commerce ∼ C1.8 billion 30% 119 No
Manufacturer
Managing
Director
∼ C40 million 80% 0 No
through platforms and marketplaces only makes sense
for manufacturers up to a certain return rate. If a high
return rate means that some products can no longer be
sold profitably through certain sales channels, these
products could be prioritised for sale in the manufac-
turer’s own online store or in bricks-and-mortar stores,
where positive margins can still be achieved. On the
system side, returns forecasts would need to be inte-
grated into the ERP or PDM system to inform sales
planning and channel selection.
Reverse Logistics Network. Many manufacturers
and retailers in the fashion industry operate both on-
line and bricks-and-mortar stores. Returns prediction
offers the opportunity to link both worlds in a mean-
ingful way in terms of the reverse logistics network.
For example, if it is determined that a customer is
likely to return an item that is selling well in one of
the bricks-and-mortar stores, a returns label can be
sent to the customer with their order that includes the
address of a suitable bricks-and-mortar store. This
means that the garment does not have to be sent to a
central warehouse: If the garment is sent directly from
the returning customer to the store, logistics costs can
be significantly reduced. This results in environmen-
tal and economic benefits for the business. This ap-
plication of returns prediction is particularly suitable
for manufacturers and retailers who operate their own
stores. The transferability of this approach to chain
stores that divert returns from their online business to
bricks-and-mortar stores seems promising.
6 IMPLICATIONS FOR AI-BASED
RECOMMENDATION SYSTEMS
The following main implications for the development
and implementation of an AI-based prediction and rec-
ommendation system for returns (avoidance) emerged
from the expert interviews. With regard to avoidance,
Enhancing Returns Management in Fashion E-Commerce: Industry Insights on AI-Based Prediction and Recommendation Systems
71

the system to be developed can make the likelihood
of returns transparent to the customer by integrating it
into the webshop system and, for example, directing
the customer to the size finder when a certain likeli-
hood of returns is identified. If the customer then uses
the recommended size finder, an incentive could be to
waive the return fee for that order. The design options
in this context are diverse and depend heavily on the
existing ordering and returns modalities of the retailer
or manufacturer.
In addition, the expert interviews revealed that,
from a practical point of view, the returns prediction
system is also particularly suitable for selecting appro-
priate distribution channels, which is also a preventive
approach. To this end, it is necessary to check whether
the probability of returns can already be determined
with sufficient accuracy in relation to product char-
acteristics and sales channels. In this case, it makes
sense to integrate the AI-based prediction and recom-
mendation system into the sales planning area of an
ERP or PDM system.
A reactive approach where returns prediction is
used in the reverse logistics network is, for example,
to deliver parcels with a calculated probability of re-
turns for certain items of clothing to customers with a
returns label containing the address of a store where
the item is selling well. In this case, too, it makes
sense to integrate the AI-based prediction and recom-
mendation system with the ERP or PDM system to
improve reverse logistics planning. This assumes that
other variables, such as sales figures from individual
stores, are also available in the ERP system or that the
AI-based prediction and recommendation system has
access to them via interfaces.
The qualitative survey results reveal that the dis-
posal of returned items is still a common practice in the
fashion e-commerce sector. Return rate estimates ex-
ceed those found in recent studies (Forschungsgruppe
Retourenmanagement, 2022), which may be due to
the high dependency of return rates on the specific
practices of companies that participated in the survey
- return policies, the product range, online sales chan-
nels and realistic presentation and description of items
play a large role. The return process for the surveyed
companies is mostly focused on customer friendliness,
which may be explained by the strong influence of
marketplaces on the overall behaviour of the market in
this direction. A recommendation system would need
to be able to act taking the market dynamics in this
direction into account and be able to avoid migration
of customers to marketplaces if the own online store
charges return fees. The survey also reveals that the
collection of data on possible return indicators is rather
sporadic in a lot of cases and not aimed at a future im-
provement of the process through the evaluation of
the generated insights. An AI-based recommendation
system would be suitable to generate insights on return
drivers and inhibitors on the basis of sales and returns
data, which must be available to a certain extent in
every e-commerce company in order to operate. A
dashboard view of return rates for different product
categories or properties like fit, style, color or size may
help with targeting the specific drivers of returns and
initiate processes to avoid or improve handling.
7 CONCLUSION AND FUTURE
WORK
Growth rates for e-commerce, and fashion e-commerce
in particular, will be high in the coming years. This
will increase the need for action in terms of avoiding
returns and managing returns in the most environmen-
tally and economically sensible way. The survey and
expert interviews highlighted the current handling with
returns, the relevance of returns avoidance and returns
management in fashion e-commerce and provided im-
petus for the development and system integration of
an AI-based prediction and recommendation system.
In particular, the expert interviews showed that eco-
nomic and ecological aspects must go hand in hand
when considering the use of an AI-based prediction
and recommendation system. As avoiding returns can
also lead to a reduction in sales, this requirement is
not trivial. Use cases in which return fees are charged
on a per-customer, per-basket basis, depending on the
return probabilities determined by the system, appear
unattractive from a practical perspective in this con-
text, as there is a fear of customers migrating to sales
channels where there are no return fees. However, this
study has shown that there are ways in which returns
prediction can be used both preventively and reactively,
without fear of economic disadvantage: On the con-
trary, the selection of appropriate distribution channels
on the basis of returns predictions and sales figures
offers far-reaching potential for reducing costs, both
economically and ecologically. In this context, inte-
gration into the webshop system or the ERP or PDM
system can be seen as promising, depending on the
described application. In order to prove its practical
suitability, the next step should be to analyse a spe-
cific use case in which an AI-based prediction and
recommendation system is used in live operation in
the webshop system or in the ERP or PDM system. It
was also not the aim of this study to provide a com-
prehensive literature review of consumer behaviour
in relation to returns, so further research in this area
seems appropriate. Furthermore, the sample of manu-
ICSBT 2024 - 21st International Conference on Smart Business Technologies
72

facturers and retailers included in the survey and the
seven expert interviews was not representative. Al-
though survey and the interviews were conducted with
companies of different sizes and focal points, it would
be beneficial to validate them within a larger sample
in order to achieve more meaningful results for the
development of the AI-based prediction and recom-
mendation system. A field test with customers who are
confronted with the concrete ideas of the present study,
such as the introduction of return fees depending on
the probability of returns determined by the system,
could also be recommended. This would allow the
effectiveness of the proposed measures to be tested.
Nevertheless, AI-based prediction and recommenda-
tion systems are not sufficient to address the issue of
returns alone. Consequently, it is essential that future
research also concentrates on topics such as process
optimisation in the context of returns processes and
reverse logistics.
ACKNOWLEDGEMENTS
This research was funded in part by the Ger-
man Federal Ministry of Education and Research
(BMBF) under the project OptiRetouren (grant num-
ber 01IS22046B). It is a joint project of the August-
Wilhelm Scheer Institut, INTEX, HAIX and h+p.
August-Wilhelm Scheer Institut is mainly entrusted
with conducting research in AI for predicting returns
volume and for recommendations based on AI.
REFERENCES
Asdecker, B. and Karl, D. (2018). Big data analytics in
returns management-are complex techniques necessary
to forecast consumer returns properly? In 2nd Interna-
tional Conference on Advanced Research Methods and
Analytics. Proceedings, pages 39–46.
Asdecker, B., Karl, D., and Sucky, E. (2017). Examining
drivers of consumer returns in e-tailing with real shop
data. In Hawaii International Conference on System
Sciences, pages 4192–4201.
Bimschleger, C., Patel, K., and Leddy, M. (2019). Bringing
it back: Retailers need a synchronized reverse logistics
strategy. Technical report, Deloitte Development LLC.
Deges, F. (2021). Retourencontrolling im online-handel.
Controlling – Zeitschrift f
¨
ur erfolgsorientierte Un-
ternehmenssteuerung, 2/2021:61–68.
Fitzpatrick, R. (2013). The Mom Test: How to talk to cus-
tomers & learn if your business is a good idea when
everyone is lying to you. Robfitz Ltd.
Forschungsgruppe Retourenmanagement (2022).
Ergebnisse des europ
¨
aischen retourentachos
ver
¨
offentlicht. https://www.retourenforschung.de/info-
ergebnisse-des-europaeischen-retourentachos-
veroeffentlicht.html. Online; accessed 2024-03-05.
Gry, S., Niederlaender, M., Lodi, A., Mutz, M., and Werth,
D. (2023). Advances in ai-based garment returns pre-
diction and processing: A conceptual approach for an
ai-based recommender system. In Proceedings of the
20th International Conference on Smart Business Tech-
nologies - ICSBT, pages 15–25. INSTICC, SciTePress.
Lepthien, A. and Clement, M. (2019). Shipping fee sched-
ules and return behavior. Marketing Letters, 30(2):151–
165.
Lohmeier (2024). E-commerce in deutschland:
Daten und fakten zum boomenden onli-
negesch
¨
aft. https://de.statista.com/themen/247/e-
commerce/ ♯ topicOverview. Online; accessed
2024-02-29.
Makkonen, M., Frank, L., and Kemppainen, T. (2021).
The effects of consumer demographics and payment
method preference on product return frequency and rea-
sons in online shopping. In Bled eConference, pages
567–580. University of Maribor.
Morgan Stanley (2022). Here’s why e-commerce growth can
stay stronger for longer. https://www.morganstanley.
com/ideas/global-ecommerce-growth-forecast-2022/.
Online; accessed 2024-02-29.
Niederlaender, M., Lodi, A., Gry, S., Biswas, R., and Werth,
D. (2024). Garment returns prediction for ai-based
processing and waste reduction in e-commerce. In
Proceedings of the 16th International Conference on
Agents and Artificial Intelligence - Volume 2: ICAART,
pages 156–164. INSTICC, SciTePress.
Saarij
¨
arvi, H., Sutinen, U.-M., and Harris, L. C. (2017). Un-
covering consumers’ returning behaviour: a study of
fashion e-commerce. The International Review of Re-
tail, Distribution and Consumer Research, 27(3):284–
299.
Sahoo, N., Dellarocas, C., and Srinivasan, S. (2018). The
impact of online product reviews on product returns.
Information Systems Research, 29(3):723–738.
Shehu, E., Papies, D., and Neslin, S. A. (2020). Free shipping
promotions and product returns. Journal of Marketing
Research, 57(4):640–658.
Statista Market Insights (2024). ecommerce - weltweit.
https://de.statista.com/outlook/emo/ecommerce/
weltweit? currency=eur. Online; accessed 2024-02-29.
Sutinen, U.-M., Saarij
¨
arvi, H., and Yrj
¨
ol
¨
a, M. (2022). Shop
at your own risk? consumer activities in fashion e-
commerce. International Journal of Consumer Studies,
46(4):1299–1318.
Yan, R. and Cao, Z. (2017). Product returns, asymmetric
information, and firm performance. International Jour-
nal of Production Economics, 185:211–222.
Enhancing Returns Management in Fashion E-Commerce: Industry Insights on AI-Based Prediction and Recommendation Systems
73