Towards the Creation of a Holistic Video Analytics Platform for

Retail Environments

Christian Daase

, Daniel Staegemann

, Anastasija Nikiforova

, Victor Chang

Johannes Hintsch

, Matthias Volk

and Klaus Turowski

Institute of Technical and Business Information Systems, Otto-von-Guericke University, Magdeburg, Germany

Institute of Computer Science, University of Tartu, Tartu, Estonia

Operations and Information Management, Aston Business School, Aston University, U.K.

nikiforova.anastasija@gmail.com, victorchang.research@gmail.com

Keywords: Video Analytics, Retail, Design Science Research, Artificial Intelligence.

Abstract: Retail is expected to be one of the industries that will benefit most from advances in artificial intelligence

(AI) in the future. One branch of AI is video analytics, which is used to analyze the behavior, flow, and

interactions of customers in a store. Properly implementing features to optimize store operations, prevent theft,

or provide targeted advertising can increase a store's profitability and reduce shrinkage. This paper proposes

an adaptation and specification of an action design research approach that forms the basis for implementing a

holistic video analytics platform that could potentially incorporate a variety of identified beneficial features.

In addition, the challenges in this regard are explained and an outlook on the future realization of such a

platform is provided.

1 INTRODUCTION

When disruptive scientific or technological means

emerge, history has shown that their impact is not

limited to the application of that technology, but is

also reflected in social structures, moral codes, and

laws (Jiang et al., 2022). One of the ubiquitous

technologies today is artificial intelligence (AI), as it

penetrates almost every aspect of our daily lives,

including education, the economy, commerce,

healthcare, public administration and governments

(Kaplan and Haenlein, 2019). In synergy with

humans, AI-related technologies, connected sensor

systems as well as underlying processing capabilities

form the basis of a new societal concept called Society

5.0 also known as a super smart society or society of

imagination (Carayannis and Morawska-

Jancelewicz, 2022; Muslikhin et al., 2021; Nair et al.,

https://orcid.org/0000-0003-4662-7055

https://orcid.org/0000-0001-9957-1003

https://orcid.org/0000-0002-0532-3488

https://orcid.org/0000-0002-8012-5852

https://orcid.org/0000-0003-3394-4131

https://orcid.org/0000-0002-4835-919X

https://orcid.org/0000-0002-4388-8914

2021), where the above are expected to serve the

needs of this form of society.

A study by McKinsey & Company revealed that

across 19 industries AI is likely to have the biggest

impact on the retail sector (Guha et al., 2021). The

range of possible applications for AI in retail is

manifold, including trend analysis through social

media, targeted marketing, implementation of virtual

and augmented reality (VR and AR, respectively) so

that customers can experience products more vividly,

automating customer service via chatbots, or ensuring

e-commerce applications recommend products

(smarter recommendation systems) that

disadvantaged people can afford (Alexandrova and

Kochieva, 2021; Bellis and Johar, 2020; Chang et al.,

2023; Daase et al., 2023; Ferracuti et al., 2019; H.

Zhang et al., 2020). One of the prominent areas for AI

in retail is video analytics (VA), either in-store or by

216

Daase, C., Staegemann, D., Nikiforova, A., Chang, V., Hintsch, J., Volk, M. and Turowski, K.

Towards the Creation of a Holistic Video Analytics Platform for Retail Environments.

DOI: 10.5220/0012148600003552

In Proceedings of the 20th International Conference on Smart Business Technologies (ICSBT 2023), pages 216-225

ISBN: 978-989-758-667-5; ISSN: 2184-772X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

analyzing online video content. VA can help

understand customer behavior and intentions by

examining their movements (for in-store scenario) or

shopping cart contents (for online scenario) (Kaur et

al., 2020; Liciotti et al., 2017). However, almost all

tasks that typically require humans (workers) to

monitor in-store cameras are of interest for VA in

retail. For example, the prevention of loss due to

shoplifting is a key component of in-store camera

utilization and is now expected to be facilitated by

VA. Another use of knowledge gained through VA is

to present products to customers that are believed to

be of particular interest to them based on prior

behavior, which can be summarized under the

designation of targeted advertisement (S. Zhang et al.,

2021). All in all, VA in retail is seen as an enabler for

smart retail business (Chandramana, 2018).

This article presents a methodological approach

to constructing a holistic platform for in-store VA by

adapting the action design research (ADR) approach

proposed by Sein et al. (2011), which is a

methodology from the field of design science

research (DSR). DSR is a research discipline whose

methodologies can be used to create usable IT

artifacts to solve organizational problems (Hevner et

al., 2004). Holistic, in this regard, means that a

modular platform is envisioned that can be extended

by integrating any generic VA use case while relying

on the same input data mode in a consistent context

(i.e., visual data in a retail environment). According

to Sein et al. (2011), a distinctive / differentiating

aspect compared to other DSR methodologies is that

ADR specifically addresses the issue that the creation

of an artifact must be guided not only by the

intentions of the researchers, but also by the

interaction with the context of its application. Since

VA in retail is highly dependent on the individuals

involved (i.e., customers), and human behavior is a

factor of uncertainty for IT system developers, the

design of a VA platform should be informed by active

instantiation. However, the approach describes in the

original work on ADR proposes a rather generic

approach, which needs to be adapted to the specific

research endeavor. This paper aims to provide a

design and development proposal for a VA platform,

but the adapted methodology could also be used for

other business scenarios involving external

participants and multiple stages of internal and

external implementation. In the light of the above and

the existing body of knowledge on the topic, the

research question is:

RQ: How can the activities of the action design

research methodology be adapted and specified to

propose a suitable approach for creating a holistic in-

store VA platform in retail?

The article is structured as follows: the following

section presents the overall methodology for this

article and describes an exploratory systematic

literature review (SLR) that is conducted to identify

VA use cases for their further integration into the

platform in more detail. Section 3 focuses on

explaining the choice of implementable use cases.

Section 4 is dedicated to the assembly of an adapted

ADR approach to implement a holistic VA platform

in retail. Potential challenges for VA are explained in

more detail. Finally, section 5 summarizes and

provides suggestions for further research, as well as

the aspired realization of the VA platform.

2 METHODOLOGY

This section presents the research methodology. First,

a general approach to the research is presented, as

well as those parts that are planned to be implemented

in the future are presented. Second, the review

protocol for the explorative SLR is presented,

including the used databases, search queries, and

inclusion / exclusion criteria.

2.1 Overall Approach

This research can be divided into three stages. First,

the research is motivated and narrowed down by the

exploratory SLR to identify VA use cases in retail that

will constitute the knowledge base and serve as an

input for the later stages. Second, the ADR

methodology (Sein et al., 2011) is purposefully

adapted and extended to provide a foundation for a

scientifically sound, i.e., evidence-based, real-world /

practical implementation of a holistic VA platform in

retail. These two phases are addressed in this paper.

The third stage, envisioned for the future, is the stage

of the implementation of the respective artifact, at

which a prototype implementation of such a platform

is expected to be proposed and validated. The overall

research process is shown in Figure 1.

2.2 Exploratory SLR

To establish a knowledge base and examine how the

topic under question has been reflected in the

literature over the years, we studied all relevant

literature covering this topic. In order to identify

relevant literature, the SLR was carried out to form

the knowledge base. This was done by searching

for relevant studies covered by Scopus and Google

Towards the Creation of a Holistic Video Analytics Platform for Retail Environments

217

Figure 1: Overall research endeavor.

Scholar, which index most well-known publishers of

peer-reviewed literature, thereby allowing us to

ensure the knowledge base is as rich as possible. The

search query was defined as a combination of the

terms “video analytics” and “retail”. We limited the

scope of the search to the article title, keywords, and

abstract in order to limit the number of articles to

those where these topics were the primary object of

study, rather than mentioned in the body, for example,

as future work. By following this approach, Scopus

retrieved six articles, while Google Scholar yielded

23 papers.

Figure 2: Exploratory SLR search process.

From the initial body of literature of 29

publications, 19 were excluded at the first stage of

review when reading the abstracts. Eight out of 19

articles were found to be duplicates, and another

eleven articles were either not written in English or

were not available for further investigation. In the

second stage, the remaining articles were read to

retrieve potential use cases and applications of

intelligent in-store video systems. Of the ten

remaining articles, four were identified as either

technical reports or unavailable sources. Finally, six

articles serve as the basis for an exploratory review to

describe a selection of VA use cases. Figure 2 shows

the exploratory review process.

3 VIDEO ANALYTICS IN RETAIL

In order to harness the opportunities that video

analytics can provide for the retail sector, it is

necessary to be aware of them. Therefore, the

following elaborations are intended to give an idea of

what capabilities can be realized with a holistic VA

platform as it can be built using the methodology

described in the following sections. Since this paper

presents VA use cases mainly from a technical point

of view, the legal requirements have to be taken into

account depending on the local law under which a

platform such as the one proposed here is to be

operated. In addition, it should be noted that strict

privacy and data security regulations, as applicable in

the European Union, may cause that some of the use

cases cannot be realized without extensive

anonymization and data security measures.

3.1 Applications from the Literature

This section briefly describes three overarching terms

for potential VA application scenarios in retail. These

include the use of VA to optimize store operations,

safety and loss prevention, and merchandising, which

we will discuss in more detail in further subsections.

3.1.1 Store Operations

The first application area of VA can be summarized

under the designation of store operations. It

encompasses various strategies for optimizing the

physical store in ways that increase employee

productivity / performance and revenue. Cameras can

be used to monitor store traffic, queues, and shoppers

/ customer behavior in general (Connell et al., 2013;

Pletcher, 2023; Senior et al., 2007). Moreover,

assistance through video cameras can be used to

analyze not only customers, but also employees

working in the store, for example, in order to optimize

their positioning or adjust the number of active staff

in the store area (Musalem et al., 2015). By

monitoring the entrance and exit of a store, retailers

can estimate the conversation rate, which stands for

the percentage of customers who buy items divided

by all people entering the store (Connell et al., 2013;

Senior et al., 2007).

ICSBT 2023 - 20th International Conference on Smart Business Technologies

218

Unlike individual customer observations, an

alternative with VA is to detect shopping carts and

items based on specific patterns (Rai et al., 2011). In

this regard, when products are part of the analysis

intent, VA can also help optimize shelf layouts

(Pletcher, 2023). Especially when considering the

interplay of / interaction between customers and

products, the following application areas for VA

show potential.

3.1.2 Safety and Loss Prevention

The loss prevention refers to several scenarios in

which a retailer’s revenue is reduced due to illegal

actions such as shoplifting and employee theft,

returns fraud, and tag switching, but it also refers to

unintentional shrinkage such as accounting errors

(Senior et al., 2007; Singh, 2018). Connell et al.

(2013) categorize three types of loss prevention:

store-floor (i.e., detecting shoplifters in the public

area of a store), back-of-store (i.e., detecting theft and

unusual behavior in the warehouse area), and front-

of-store (i.e., detecting theft and criminal behavior at

the check-out area by employees or customers). VA

can help identify shoppers who do not scan all their

items properly before payment, switch the tags

between items, or when cashiers are involved in such

acts.

Another related application scenario for VA in

retail is to improve security and safety in case of

emergencies or criminal acts that are not directly

related to the retail store (Connell et al., 2013). VA

systems can be used to detect accidents, technical

misbehavior, dangers or hazards, and medical

emergencies.

3.1.3 Merchandising

A third area of physical retail where VA can be

purposefully applied is merchandising, also related to

targeted advertising and product promotion (Pletcher,

2023; Rai et al., 2011; Singh, 2018). In this regard,

the estimation of the locations of customers in the

store can be a component of such targeted

promotional offers. Spots where customers spend an

unusual amount of time, or places that attract a lot of

customers, can lead to hot zones and dwell times for

customers (Connell et al., 2013). Heat map

visualizations of in-store places with increased traffic

flow can also help retail managers to optimize

spontaneous advertising (Senior et al., 2007). Pricing

strategies can also play a role in promotional

campaigns when it comes to setting prices based on

previously identified trends. If an increased flow of

customers is expected based on previous seasonal

experience, a retailer may decide to adjust prices

while promoting a particular product or group of

goods.

In addition to observing physical movements for

location-based merchandising, the psychological

behavior of customers can also be assessed using VA.

One example is tracking eye movements using gaze

analysis to identify which products or information

attract the most attention from individual customers

(Connell et al., 2013). Knowing this, shelves can be

optimized, and promotional calls can be placed

nearby (Pletcher, 2023).

3.2 Further Potential Applications

In addition to the above discussed, we think that

further potential applications of VA could be

harnessed to provide value to retailers.

For example, like (Musalem et al., 2015), we

believe that VA could be useful for monitoring not

only customers / clients, but also staff / personnel.

However, we think that it can serve not only the

purpose of their allocation but also seek to evaluate

the quality of assistance they provide to consumers.

This would potentially allow the business owner to

find pain points, best and worst practices that could

be further converted / transformed into training and

policies, as well as identify problems in the respective

business process (or its specific activities).

Improvements based on the above would further

improve the image of the business (in the long term).

Similar to what was discussed in 3.1.3 and based

on a suggestion by Connell et al. (2013), gaze analysis

can be potentially applied to inspect the overall

planning / layout of the sales areas, identify areas that

are of greater interest to customers (due to their

physical location), where then promotions or other

types of products or services in whose popularity the

business owner is interested most should be located

avoiding spots that are typically overlooked by

customers, or triggering the redesign activities. Here,

however, not only gaze analysis, but also heat maps

could be useful, or, preferably, a combination of both,

thereby seeking an increased accuracy and validity of

the results.

Additionally, with today’s technologies,

predictive analytics and (near) real-time VA-based

predictions / forecasts, in particular, are seen as an

emerging trend that has the potential to further

contribute to the overall business success. For

example, reducing the waiting times between

analyses of “as-is” and “to-be” models based only on

historical data is possible, thereby allowing business

owners to make adjustments to their marketing

Towards the Creation of a Holistic Video Analytics Platform for Retail Environments

219

strategies faster. This is also in line with (Ghose et al.,

2022). This is of particular importance also for daily

activities such as the number of cashiers needed at

this moment and expected to be needed in one hour

based not only on the historical data, meaning the day

of the week or season, but considering the actual

situation of the given day. Of course, in this example,

historical data is used as well, but rather as

complementary source and not as primary input (also

in line with (Anderson, 2022).

4 IMPLEMENTATION

STRATEGY

A platform for VA in retail that integrates a suitable

set of the aforementioned capabilities and more

requires a tailored design methodology. In this

section, ADR is adapted to provide a starting point for

constructing such a platform. A multistep approach is

taken, starting from scientifically grounding the

design to implementing it in artificial scenarios and

later real-world environments.

4.1 Action Design Research (ADR)

The adoption of ADR with its focus on organizational

contexts for the set goal of creating an adaptive

reference architecture may seem partially

contradictory. The focus on organizational contexts

may indicate the existence of a rather precise

scenario, while the goal of providing a universal

reference architecture aims at an artifact for multiple

use cases. However, considering that the design part

in DSR is the “purposeful organization of resources

to accomplish a goal” (Hevner et al., 2004), and

assuming that the most positive arrangement of

integrable tools is revealed only when the context is

known, this discrepancy can be resolved by

specifying that different levels of detail can be

defined for the same reference architecture. Sein et al.

(2011) also recognize the disparity of inherent

challenges in ADR, noting that the artifact building

process must incorporate the influence of users and

contextual use in the same way that it must

incorporate theoretical precursors and researcher

intent. In the following, ADR is presented in a

specially adapted version that describes the individual

steps required to develop an applicable holistic VA

platform in a real business context.

4.2 Adapted ADR to Realize Video

Analytics Platform

The first stage of the research process, the problem

formulation, is founded on input derived from reliable

sources such as experienced practitioners or

researchers, the study of existing technologies, and

the review of prior research. In this article, this sub-

stage of selecting and investigating related references

is considered as an initial exploration phase, covered

in section 3. In order to position the envisioned VA

platform in a scientific context, the findings impact

the formulation of precise RQs that are in turn

decisive for the assembly of appropriate search

queries for an SLR. As a second sub-stage for the first

part of the adapted ADR approach, insights gained

from an extensive SLR in the future research process

represent the main source for design decision support.

Furthermore, a solid groundwork will be laid by

conducting a comprehensive SLR to justify the

research objective, examine comparable approaches

to the targeted artifact, and identify key technologies

with their potential interplay for a prototypical

concept.

Shaping the subsequent entire research process,

Sein et al. (2011) note that the critical elements in

conceptualizing the workflow are, firstly, that the

long-term commitment of participating

organization(s) must be secured and, secondly, that

the problem to be solved should be defined as an

instance of a broader class of problems. Regarding the

aspired VA platform with several capabilities

integrated, the involved organizations might be retail

companies, store owners, or malls in general.

Furthermore, the instance of a problem can be the

implementation of a subset of VA capabilities, while

the class of problems is the broader domain of VA in

general. By relying in the first iteration of the research

process on simulated placeholders whose behavior is

based on insights from the SLR, both critical elements

are addressed. On the one hand, the constancy of the

instance setup is ensured. On the other hand, the level

of detail of the organizations can be seamlessly

adjusted so that both an instantiated specific problem,

as well as the superordinate class of problems it

belongs to, can be described. The methodological

guidelines propose two principles for this stage. First,

practice- or problem-inspired research should be

pursued to solve a specific problem and generate

generally applicable knowledge for the problem class.

Second, the targeted artifact should be theory-

ingrained, meaning that the initial design is driven by

theory and later reshaped by organizational practice.

Both principles are met here by reviewing the

ICSBT 2023 - 20th International Conference on Smart Business Technologies

220

scientific literature and considering the contextual

implications for the relevant research activities. In an

elaborated version of the ADR process model,

Mullarkey and Hevner (2019) remark that problem

formulation further depends on which iteration of the

research is concerned, as the original stages of ADR

can be applied with different scopes, starting with a

phase of diagnosis of the research objective and

progressing to a phase of evolution (i.e., refinement

and further development of the final artifact).

The second stage is a potentially cyclical

combination of building the artifact, its intervention

in the organization, and the evaluation (BIE). The

result of this stage is usually an approved variant of

an artifact realization that has undergone a continuous

and partially repetitive process of designing,

evaluating, and redesigning. Sein et al. (2011) present

two perspectives from the edges of the ADR

spectrum: an IT-dominant BIE process and an

organization-dominant one. The former can be

summarized as striving for an innovative

technological design, creating early designs and alpha

versions that are light-weight interventions in an

organizational context with limited scope,

continuously instantiating the artifact and testing it

repeatedly, and finally applying it in a wider

organizational setting. The second perspective takes a

closer connection to the organizational context, as the

inherent design knowledge is mainly oriented to the

organizational intervention itself. This present

research adopts an IT-dominant BIE process. In the

early research stadium, theoretical contributions from

the literature are translated into a practical

manifestation of the holistic VA platform with

placeholders that is artificially tested in a controlled

environment. The results are subsequently used to

extract design knowledge and, if necessary, influence

the redesign for a repetition of the previous step. If

the design seems sufficient, it is then adapted for a

concrete use case scenario, leading to a specific

instantiation. In case this specific instance does not

comply with the organizational requirements, a new

iteration of the adaptation process can be initiated.

Once the specific instance has been successfully

applied in the realistic scenario, the design proposal

is returned to the research team to abstract the

approved design as a reference architecture. Sein et

al. (2011) state three principles for this stage. First,

the principle of reciprocal shaping emphasizes the

consideration of the influence the artifact has on the

organizational context and vice versa. The

methodology employed here recognizes these

interdependent relationships by integrating two

potential cycles between activities related to

theoretical research and practical applications of the

artifact. On the one hand, the original artificial

manifestation commissioned by practitioners is

evaluated from the research perspective and proven

design knowledge is extracted, whereupon the

artificial instantiation is either redesigned and

recommissioned or forwarded for adaptation to a

concrete application scenario. On the other hand,

once the concrete adaptation is taken into an end-user

context as a specific instantiation, it is either returned

for a refined adaptation if the design is not approved

by the end-user, or it is forwarded to the research team

to be abstracted into the final reference architecture

for different use cases, meaning the integration of a

wide range of VA capabilities. The second principle,

mutually influential roles, refers to the assignment of

individuals to different roles, such as theoretical

researchers, practitioners, and end-users, which is

related to the respective areas of experience. The third

and last principle for this stage, authentic and

concurrent evaluation, emphasizes the evaluation of

the artifact directly during the building process rather

than afterwards. However, since a new version must

be fully built before its value to the organization can

be assessed, short build-evaluate cycles are chosen in

this study instead of an actual concurrent evaluation

approach.

The third research stage, reflection and learning,

is conducted as a parallel process to the first two

stages. According to the ADR methodology, the

encompassed tasks relate to mental work such as

reflecting on the design, evaluating the adherence of

the artifact to certain principles, and analyzing the

intervention results compared to the goals. Therefore,

the only stated principle for this stage, guided

emergence, refers to the integration of acquired

knowledge during the conduction of activities of the

BIE process into further design, evolution, and

refinements.

The last ADR stage is the formalization of

learning, which targets the development of a

generalized solution based on the findings that could

be gained by addressing a specific problem. This

stage overlaps with the final activity of the BIE

process, which is the abstraction of a specific

instantiation into a reference architecture suitable for

different levels of detail. In the methodological

guidelines, the outcomes of this stage are termed as

design principles or, after refinements, theories. The

only principle here is to aim for generalized

outcomes. Sein et al. (2011) recognize that this poses

a challenge for the research endeavor since ADR

outcomes are usually organization context-specific.

However, the abstraction of the solution (i.e.,

Towards the Creation of a Holistic Video Analytics Platform for Retail Environments

221

Figure 3: Adapted ADR methodology for VA platform realization.

translating a specific instantiation into a reference

architecture) is possible if the problem is

simultaneously translated into a broader class of

problems. This procedure is considered in the

corresponding step in the overall research process.

In this paper, the problem formulation stage,

including an exploratory SLR is conducted. In the

next section, potential challenges for implementing

capabilities of VA are explained in more detail.

Figure 3 illustrates the entire research process

envisioned with respect to the design and construction

of a holistic VA platform in retail environments. The

activities are structured as in the original ADR

methodology by Sein et al. (2011) and connected to

the respective principles. The flow between the steps

and additional feedback loops are visually provided.

4.3 Potential Challenges

It should be noted that there are potential challenges

which can surface during implementation. The first

one, which applies to VA in a general case, is related

to data privacy in terms of two types of risks – legal

and ethical, since it is seen to fall into grey area. More

specifically, this is about unconsented video taking

and analytics, which is the case for VA in retail. The

customer who visits the store does not provide his or

her consent to be filmed with further analysis of this

information. Moreover, the customer not only does

not provide such consent but is not even asked about

it, and thereby does not have an “opt-out” option. A

question that arises here is also the ownership of the

recorded material, meaning whether the retailer

becomes the legal owner of the video footage or

whether the customer reserves the right to have access

to the recorded data and delete it upon request. While

recording the video in the retail area is something the

customer tends to be warned about when entering the

area through the respective signs on the doors, it is not

mentioned that these data are then used for analytic

purposes. This might be incompliant with the General

Data Protection Regulation (GDPR). Thus, although

technologies can be available and frameworks can be

developed, the risk of data being illegal to be used

remains. VA in retail on-site retail stocks can incur

costly legal expenses and risk damage to their brand

(Pletcher, 2023). However, while providing an

information about cameras and further VA of the

material being filmed (including face recognition,

biometric surveillance etc.) could be at least a partial

solution, it was found by (Garaus et al., 2021) that the

retailer most probably will not give a preference to

this option since the value of the data, when the

customer is aware of being filmed, reduces (i.e., the

behavior – movements, emotions etc. changes and are

not intuitive and natural anymore). While retailers

ICSBT 2023 - 20th International Conference on Smart Business Technologies

222

heavily rely on cameras for surveillance, it is

sometimes not appreciated when customers perform

so-called sousveillance, meaning the recording of

activities from their perspective (Mann, 2017). Due to

multidimensionality of this issue, it is currently a hot

topic, where the predominant body of literature just

admits this issue and rather makes a call to action

(Gregorczuk, 2022; Pletcher, 2023).

In addition, this technology acceptance by the

customers tends to be seen as low. For example,

Garaus et al. (2021) found that when the customers

were informed that they had been served up

advertisements based on an algorithm and their

pictures, the customers felt manipulated and

discouraged. Hence the need for further investigation

of how to find the balance between the benefits VA

brings to both the business and its customers, who

beforehand VA are targeted with products and

services aligned with their interests, and trust in both

VA and the result it brings through the technology

acceptance theories.

Another challenge comes from the technologies

that VA is associated with, namely, machine learning

(ML) models and computing power. The first – ML

models-related – requires the development of

appropriate models (including requirements such as

being sufficiently accurate, unbiased, preferably

transparent / white-box etc.) and their availability for

more mass use in order to ensure that each business

owner does not have to have its own ad-hoc ML

model in place with an employee with the relevant

knowledge and skills responsible for their

development, deployment and maintenance. This is

even more important considering that the VA in retail

depends heavily on the ability to analyze people

properly, however, today’s ML models often fail

when race and ethnicity serve as crucial parameters,

which, however, in the case of the retail should be

treated accurately. In addition, value-adding VA

should employ a variety of different techniques and

methods to get the most profit and benefit of VA,

including but not limited to deep neural networks,

facial recognition, gesture analytics, and motion

analytics, emotion mining that preferably should also

be combined with audio gathering and processing

along with video etc. The latter, in terms of

computing power, however, is more related to the

amounts of data a VA are expected to deal with. In

other words, the larger the amount of data being

processed and the closer the idea is to real-time

analytics, the more computing power may be needed.

Depending on the approach, the amount of data being

processed can increase from day to day as they

accumulate cumulatively. This, in turn, leads to the

need for constant monitoring of the available and

required for the near future computing power, as well

as the need to assess the feasibility of the selected

approach periodically.

An apparent challenge also comes from security

and privacy as follows. First, the platform will need

advanced technologies to block unauthorized access

and have a full access control and authentication to

ensure only authorized users can have access. Second,

network security will need to be enhanced to ensure

VA can be broadcast and disseminated on secure

network platforms and protect the network and users’

safety and privacy. Third, additional technologies and

functions will be developed to identify any

impersonators and deep fake technologies to prevent

identity theft and the use of deep fake to damage the

service and reputation of the VA platform. AI-

enabled security can be used to monitor the network

and user activities, detect any abnormalities and

provide feedback to the VA system, which can then

enable security functions, such as authentication,

access control, identify management, intrusion

detection, quarantine and blockage of any

unauthorized uses of VA system.

5 CONCLUSION AND FUTURE

RESEARCH

This article examined the current trend of

implementing AI technologies in the retail industry,

with a focus on video analytics. In the future, this area

is expected to help retailers increase store profitability

and reduce shrinkage. An adaptation and

specification of the action design research

methodology were proposed that could be used to

develop a holistic VA platform that integrates a

variety of the capabilities presented here. The next

methodological steps of the overarching research

endeavor are planned to first construct a prototype

version of such a platform in a controlled

environment and then extend its use to real-world

scenarios. We also discuss challenges in depth and

provide our views and recommendations to reduce

any impact. Finally, an instantiation with realistic

parameters is to be built and evaluated.

REFERENCES

Alexandrova, E., & Kochieva, A. (2021). Modern Aspects

of Digital Technologies Development in Retail

Networks. In T. Antipova (Ed.), Lecture Notes in

Towards the Creation of a Holistic Video Analytics Platform for Retail Environments

223

Networks and Systems. Comprehensible Science (Vol.

186, pp. 111–120). Springer International Publishing.

Anderson, L. (2022). Video Analytics Applications In

Retail - Beyond Security. https://www.securityi

nformed.com/insights/co-2603-ga-co-2214-ga-co-188

0-ga.16620.html

Bellis, E. de, & Johar, G. V. (2020). Autonomous

Shopping Systems: Identifying and Overcoming

Barriers to Consumer Adoption. Journal of Retailing,

96(1), 74–87.

Carayannis, E. G., & Morawska-Jancelewicz, J. (2022).

The Futures of Europe: Society 5.0 and Industry 5.0 as

Driving Forces of Future Universities. Journal of the

Knowledge Economy, 13(4), 3445–3471.

Chandramana, S. B. (2018). Overcoming the Challenges

and Realizing the Potential of Retail Analytics for Next

Generation Smart Retail Business. https://doi.org/

10.6084/m9.figshare.13323170

Chang, V., Marshall, R., Xu, Q. A., & Nikiforova, A.

(2023). E-commerce assistant application incorporating

machine learning image classification. International

Journal of Business and Systems Research, 17(1),

Article 127711, 1.

Connell, J., Fan, Q., Gabbur, P., Haas, N., Pankanti, S., &

Trinh, H. (2013). Retail video analytics: an overview

and survey. In R. P. Loce, E. Saber, & S. R. Vantaram

(Eds.), SPIE Proceedings, Video Surveillance and

Transportation Imaging Applications (86630X). SPIE.

https://doi.org/10.1117/12.2008899

Daase, C., Volk, M., Staegemann, D., & Turowski, K.

(2023). The Future of Commerce: Linking Modern

Retailing Characteristics with Cloud Computing

Capabilities. In Proceedings of the 25th International

Conference on Enterprise Information Systems

(pp. 418–430). SCITEPRESS - Science and

Technology Publications.

Ferracuti, N., Norscini, C., Frontoni, E [E.], Gabellini, P.,

Paolanti, M., & Placidi, V. (2019). A business

application of RTLS technology in Intelligent Retail

Environment: Defining the shopper's preferred path and

its segmentation. Journal of Retailing and Consumer

Services, 47, 184–194.

Garaus, M., Wagner, U., & Rainer, R. C. (2021).

Emotional targeting using digital signage systems and

facial recognition at the point-of-sale. Journal of

Business Research, 131, 747–762.

Ghose, A., Li, B., Li, R., & Xu, K. (2022). Real-Time

Purchase Prediction Using Retail Video Analytics. In

ICIS 2022 Proceedings.

Gregorczuk, H. (2022). Retail Analytics: Smart-Stores

Saving Bricks and Mortar Retail or a Privacy Problem?

Law, Technology and Humans, 4(1), 63–78.

Guha, A., Grewal, D., Kopalle, P. K., Haenlein, M.,

Schneider, M. J., Jung, H., Moustafa, R., Hegde, D. R.,

& Hawkins, G. (2021). How artificial intelligence will

affect the future of retailing. Journal of Retailing, 97(1),

28–41.

Hevner, March, Park, & Ram (2004). Design Science in

Information Systems Research. MIS Quarterly, 28(1),

75.

Jiang, Y., Li, X., Luo, H., Yin, S., & Kaynak, O. (2022).

Quo vadis artificial intelligence? Discover Artificial

Intelligence, 2(1), Article 4.

Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand:

Who’s the fairest in the land? On the interpretations,

illustrations, and implications of artificial intelligence.

Business Horizons, 62(1), 15–25.

Kaur, J., Arora, V., & Bali, S. (2020). Influence of

technological advances and change in marketing

strategies using analytics in retail industry.

International Journal of System Assurance Engineering

and Management, 11(5), 953–961.

Liciotti, D., Frontoni, E [Emanuele], Mancini, A., &

Zingaretti, P. (2017). Pervasive System for Consumer

Behaviour Analysis in Retail Environments. In K.

Nasrollahi, C. Distante, G. Hua, A. Cavallaro, T. B.

Moeslund, S. Battiato, & Q. Ji (Eds.), Lecture Notes in

Computer Science. Video Analytics. Face and Facial

Expression Recognition and Audience Measurement

(Vol. 10165, pp. 12–23). Springer International

Publishing.

Mann, S. (2017). Big Data is a big lie without little data:

Humanistic intelligence as a human right. Big Data &

Society, 4(1), 205395171769155.

Mullarkey, M. T., & Hevner, A. R. (2019). An elaborated

action design research process model. European Journal

of Information Systems, 28(1), 6–20.

Musalem, A., Olivares, M., & Schilkrut, A. (2015). Retail

in High Definition: Monitoring customer assistance

through video analytics. https://doi.org/10.13140/

RG.2.1.4343.7925

Muslikhin, M., Horng, J.‑R., Yang, S.‑Y., Wang, M.‑S., &

Awaluddin, B.‑A. (2021). An Artificial Intelligence of

Things-Based Picking Algorithm for Online Shop in the

Society 5.0's Context. Sensors (Basel, Switzerland),

21(8).

Nair, M. M., Tyagi, A. K., & Sreenath, N. (2021). The

Future with Industry 4.0 at the Core of Society 5.0:

Open Issues, Future Opportunities and Challenges. In

2021 International Conference on Computer

Communication and Informatics (ICCCI) (pp. 1–7).

IEEE.

Pletcher, S. N. (2023). Visual Privacy: Current and

Emerging Regulations Around Unconsented Video

Analytics in Retail. https://doi.org/10.31219/

osf.io/tfw96

Rai, H. G., Jonna, K., & Krishna, P. R. (2011). Video

analytics solution for tracking customer locations in

retail shopping malls. In C. Apte, J. Ghosh, & P. Smyth

(Eds.), Proceedings of the 17th ACM SIGKDD

international conference on Knowledge discovery and

data mining (pp. 773–776). ACM.

Sein, M. K., Henfridsson, O., Purao, S., Rossi, M., &

Lindgren, R. (2011). Action Design Research. MIS

Quarterly, 35(1), 37.

Senior, A. W., Brown, L., Hampapur, A., Shu, C.‑F.,

Zhai, Y., Feris, R. S., Tian, Y.‑L., Borger, S., &

Carlson, C. (2007). Video analytics for retail. In 2007

IEEE Conference on Advanced Video and Signal Based

Surveillance (pp. 423–428). IEEE.

ICSBT 2023 - 20th International Conference on Smart Business Technologies

224

Singh, H. (2018). Applications of Intelligent Video

Analytics in the Field of Retail Management. In J.

Wang, A. Kumar, & S. Saurav (Eds.), Advances in

Logistics, Operations, and Management Science.

Supply Chain Management Strategies and Risk

Assessment in Retail Environments (pp. 42–59). IGI

Global.

Zhang, H., Li, Y., Ai, Q., Luo, Y., Wen, Y., Jin, Y., &

Ta, N. B. D. (2020). Hysia: Serving DNN-Based

Video-to-Retail Applications in Cloud. In C. Wen

Chen, R. Cucchiara, X.-S. Hua, G.-J. Qi, E. Ricci, Z.

Zhang, & R. Zimmermann (Eds.), Proceedings of the

28th ACM International Conference on Multimedia

(pp. 4457–4460). ACM.

Zhang, S., Feng, Y., Bauer, L., Cranor, L. F., Das, A., &

Sadeh, N. (2021). “Did you know this camera tracks

your mood?”: Understanding Privacy Expectations and

Preferences in the Age of Video Analytics. Proceedings

on Privacy Enhancing Technologies, 2021(2), 282–

304.

Towards the Creation of a Holistic Video Analytics Platform for Retail Environments

225