The American Journal of Engineering and Technology
57
https://www.theamericanjournals.com/index.php/tajet
TYPE
Original Research
PAGE NO.
57-64
10.37547/tajet/Volume07Issue04-08
OPEN ACCESS
SUBMITED
23 February 2025
ACCEPTED
25 March 2025
PUBLISHED
08 April 2025
VOLUME
Vol.07 Issue04 2025
CITATION
Sergei Berezin. (2025). Explainable Ai In Customer Experience
Management: Personalization Algorithms in Crm Systems. The American
Journal of Engineering and Technology, 7(04), 57
–
64.
https://doi.org/10.37547/tajet/Volume07Issue04-08
COPYRIGHT
© 2025 Original content from this work may be used under the terms
of the creative commons attributes 4.0 License.
Explainable Ai In Customer
Experience Management:
Personalization Algorithms
in Crm Systems
Sergei Berezin
Student at Midwestern Career College in the Associate of Applied Science
in Information Technology program
Founder and project manager of CRM-system for restaurants with AI
integration. Chicago, USA
Abstract:
The article examines the features of
integrating artificial intelligence algorithms (Explainable
AI, XAI) into CRM systems aimed at enhancing customer
experience (Customer Experience, CX). Based on an
analysis of recent publications, the study explores the
principles of personalization as well as approaches to
the explainability of machine learning algorithms,
including chatbots and recommendation systems. It
demonstrates that transparency and interpretability of
model outputs positively influence customer trust and
loyalty while simultaneously improving the efficiency of
internal business processes. The article analyzes the
implementation experience of XAI in the banking sector,
insurance call centers, and online retail, which has led to
improvements in retention, conversion, and satisfaction
metrics. The information presented in the article is
intended for researchers and professionals in the field
of artificial intelligence focused on developing
interpretable machine learning algorithms, as well as for
analysts seeking to optimize CRM systems to enhance
customer experience management. In addition, the
material is useful for professionals in corporate
governance and marketing who aim to integrate
advanced Explainable AI methods into personalization
strategies and decision-making processes, ensuring the
transparency and adaptability of services under
dynamic market conditions.
Keywords:
artificial intelligence, Explainable AI (XAI),
Customer Experience (CX), personalization, CRM
systems, machine learning, chatbots, recommendation
systems
The American Journal of Engineering and Technology
58
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
Introduction:
Modern companies increasingly rely on
artificial intelligence (AI) tools to manage customer
relationships, including the personalization of services
and products [4]. It is important not only to enhance
the effectiveness of interactions but also to ensure the
transparency of the algorithmic decisions made, since
explainability (Explainable AI) becomes a decisive
factor in users' trust [3]. In the context of ensuring a
positive customer experience (Customer Experience,
CX), such algorithmic "transparency" is gaining
importance, as a lack of understanding of the
mechanics behind recommendations or data analysis
results can negatively affect consumer satisfaction and
loyalty [5].
The literature review demonstrates a variety of
approaches, ranging from conceptual models and
theoretical reviews to empirical studies focused on the
practical application of AI technologies across various
industries. In the theoretical realm, Peruchini M., da
Silva G. M., and Teixeira J. M. [1] review the
interrelation between AI and customer experience,
emphasizing the interdisciplinary nature of the issue,
which is confirmed by the findings of Ameen N. et al.
[3] and Dwivedi Y. K. et al. [4], who examine both the
challenges and opportunities that arise when
integrating AI into organizational processes. An
additional contribution to the theoretical foundation is
provided by the work of Robinson S. et al. [10], which
proposes an evolved model of service interaction, as
well as by the study of Phillips-Wren G., Daly M., and
Burstein F. [17], which aims to integrate analytics,
business intelligence, and decision support systems,
collectively creating a solid basis for further empirical
research in this area.
Empirical literature actively employs explainable AI
approaches to assess service quality and analyze
customers' emotional responses. In particular, Guo Y.
et al. [2] propose a method for measuring service
quality based on the analysis of customer emotions
using explainable algorithms, allowing not only the
prediction but also the interpretation of user reactions.
A similar approach is confirmed in the work of Ceccacci
S. et al. [12], where the use of facial expression analysis
within the framework of auditing impressions from
cultural events demonstrates the potential for
objectively assessing the emotional component of
customer experience.
Separate attention in the literature is devoted to the
application of interactive systems, such as chatbots
and voice assistants, which have a significant impact on
the personalization of service. Abdelkader O. A. [5]
examines the influence of ChatGPT on digital
marketing, revealing moderating effects on the
perception of customer experience, while the studies by
Abdo A. and Yusof S. M. [8] demonstrate how voice
chatbots contribute to improving the quality of
customer interactions. The research of McLean G. and
Osei-Frimpong K. [9] and Nguyen D. M., Chiu Y. T. H., and
Le H. D. [13] focuses on the determinants of adopting
such technologies in the banking sector, emphasizing
the importance of adaptive algorithms in establishing
sustainable customer relationships.
The banking and financial services sectors represent
another area of AI application, where personalization
and risk assessment play a key role. In the study by Ho
S. P. S. and Chow M. Y. C. [11], the influence of AI on
shaping customer preferences in retail banking is
examined, while Bhattacharya C. and Sinha M. [14]
underscore the strategic significance of AI for enhancing
competitiveness
through
improved
customer
experience. In addition, Zhou J. et al. [16] demonstrate
the application of integrated methods that combine
increased data volume and model refinement for
multistage credit risk assessment, an important aspect
of personalization in financial services.
Equally interesting are studies dedicated to the specifics
of AI application in the entertainment and events
industries. Neuhofer B., Magnus B., and Celuch K. [7],
using a scenario-based approach, analyze the influence
of AI on event perception, while Puntoni S. et al. [6] offer
an experiential perspective that reveals new facets of
consumer interaction with artificial intelligence in the
marketing environment. In turn, Wulff K. and
Finnestrand H. [15] address the issue of AI explainability
in the context of creating meaningful work, highlighting
the importance of algorithm interpretability not only for
customers but also for internal organizational
stakeholders.
Thus, the analysis of the presented studies reveals
contradictions between theoretical models proposing
universal approaches to integrating AI into customer
experience management and empirical works focused
on narrow industries and specific aspects, such as
emotion assessment or factors influencing the adoption
of interactive systems. At the same time, the problem of
comprehensive
integration
of
explainable
personalization algorithms into CRM systems remains
underexplored, indicating the need for further research
aimed at developing scalable and interpretable models
capable of providing a unified solution for a wide range
of customer scenarios.
The purpose of the research is to identify and analyze
the mechanisms by which explainable artificial
intelligence (Explainable AI) enhances customer trust
The American Journal of Engineering and Technology
59
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
and satisfaction when using personalization tools in
CRM systems.
The scientific novelty lies in the proposal of a method
that bridges the gap between understandable AI and
personalized CRM strategies. This is achieved through
a systematic review of current literature, a thorough
empirical analysis across various industries, and expert
evaluations that demonstrate how transparent
algorithmic decision-making fundamentally enhances
customer trust, engagement, and operational
efficiency.
The hypothesis is based on the assumption that
integrating Explainable AI into personalized CRM
algorithms contributes to increased customer trust,
which in turn positively affects Customer Experience
metrics.
The methodological basis for the study is the analysis
of existing research.
1. Theoretical foundations
Customer Experience (CX) is shaped at the intersection
of consumer perception of the brand, interaction
technology, and emotional customer engagement. In
contrast to User Experience (UX), where the primary
focus is on interface usability and functionality, CX
encompasses the complete cycle of contacts and
impressions formed during interactions with a
company. Research indicates that a high level of CX
leads
to
increased
goodwill
and
positive
recommendations, as well as serving as a competitive
advantage [6,13].
At the same time, technological innovation alone does
not guarantee a positive outcome; without sufficient
transparency in artificial intelligence, both customers
and managers may question the accuracy of the
recommendations provided [5,17]. The concept of
Explainable AI (XAI) emerges as a key element,
emphasizing the importance of model interpretability to
enhance trust and understanding of the underlying logic
of algorithms [4]. Within the context of CX, this implies
that transparency in the results of personalized
recommendations and chatbot decisions can influence
user satisfaction and subsequent behavior [1].
In recent years, chatbots and virtual assistants have
been actively explored in academic circles as tools for
improving service quality [8,14]. One advantage of these
technologies is their ability to collect large volumes of
textual or voice data, which facilitates a deeper
understanding of customer interests and emotional
responses [9]. However, the use of deep learning
algorithms or recommendation systems often leads to a
"black-box" effect, whereby neither users nor
developers can clearly interpret the rationale behind a
recommended action [10].
Table 1, presented below, illustrates the main
approaches to applying artificial intelligence in CX
identified in the literature, correlating them with
specific challenges (such as the "black-box" effect) and
potential benefits for both customers and companies.
Table 1. The main approaches in CX in terms of benefits and challenges [1, 4, 9, 11, 14]
Approach / Technology
Key Features and Benefits
Main Challenges
Machine
Learning
(recommendation
systems,
ML)
• Personalization of offers
• Analysis of large datasets
• Increased accuracy of
predictive models
• Lack of interpretability
(black-box)
• Potential algorithm errors
with incomplete data
Chatbots
and
Virtual
Assistants
• 24/7 support
• Reduced contact center costs
• Convenience and speed of
service
• Difficulties in humanizing
responses
• Risk of negative reactions
due to incorrect answers
Emotion Processing (Emotion
AI, sentiment analysis)
• Deeper understanding of
customer intentions
• Automatic identification of
dissatisfied customers
• Need to maintain privacy
• Technical challenges in
capturing precise emotional
data
The American Journal of Engineering and Technology
60
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
Approach / Technology
Key Features and Benefits
Main Challenges
Explainable
AI
(XAI
modules)
• Increased trust
• Facilitation of AI integration
in corporate environments
• Opportunities for staff
training
• Additional implementation
costs
• Potential trade-offs between
accuracy and interpretability
Thus, the application of AI in CRM systems enables the
analysis of increasingly complex aspects of customer
behavior. However, a lack of transparency and clear
explanations for users can adversely affect customer
perception and trust in a company. For this reason, the
study of Explainable AI is of great importance, as it
allows for balancing high recommendation accuracy
with interpretability.
2. Personalization algorithms and approaches to XAI
in CRM
In modern CRM systems, personalization is considered
one of the mechanisms for increasing conversion,
retention, and customer satisfaction [3,15]. Such a
system
includes
the
selection
of
relevant
recommendations, content, and individual interaction
scenarios based on the analysis of customer data. The
most common technological approaches are:
1.
Collaborative filtering (collaborative filtering).
Uses information about the preferences of similar
customers; the algorithm “recommends” products or
services that have appealed to other users with similar
characteristics [13,16].
2.
Content-based
filtering
(content-based
filtering). Focuses on the inherent attributes of
products or services; the system analyzes features
characteristic of the objects preferred by customers
and searches for similar elements [6,11].
3.
Hybrid models. A combination of collaborative
and content-based filtering to overcome typical
shortcomings (for exampl
e, the “cold start” problem,
when a new user has insufficient data on preferences)
[14].
4.
Deep learning models (deep learning).
Employed for more complex personalization scenarios
such as dynamic real-time recommendations and
multimodal analytics (text, voice, images) [1].
Recommendation systems that are trained on large
volumes of customer data (Big Data)
—
including
demographics, purchase history, behavioral patterns,
and even emotional reactions
—
are of particular interest
[12]. However, the increasing complexity of models
often gives rise to the “black
-
box” effect, where
interpreting the output becomes challenging. As a
result, the system may deliver highly accurate
recommendations without any means of understanding
the underlying decision-making logic.
Explainable AI (XAI) is designed to overcome the
problem of algorithmic opacity. Specific XAI methods
can be conditionally divided into three major groups:
1. Post-hoc explanation:
●
LIME (Local Interpretable Model-Agnostic
Explanations): builds local approximations of complex
models, showing the contribution of individual features
in a specific case [15].
●
SHAP
(SHapley
Additive
exPlanations):
calculates the marginal contribution of each feature,
combining them into a generalized interpretation akin
to game theory [4].
2. Interpretable models:
●
Decision trees and tree ensembles (Random
Forest, XGBoost): although not always obvious when
deep, they remain easier to understand than deep
neural networks [11].
●
Switch architectures (for example, transparent
two-way networks), where each part is responsible for
a specific set of features and follows an interpretable
logic [2].
3. Integrated interpretation:
●
Attention mechanisms in deep networks
(Transformer-based models) allow visualization of
which parts of the input data the model focuses on most
[10].
●
Explainable-by-design: the use of simplified
architectures or specialized layers that automatically
formulate the output logic in a more transparent format
[5,8].
In practice, the choice of an XAI approach is determined
by the balance between model accuracy and the
The American Journal of Engineering and Technology
61
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
required level of transparency. In CRM systems, where
the cost of an error is low (for example, recommending
a product of limited value), hybrid or “black
-and-
white” schemes may be applied, where
by part of the
model remains deeply trained (and less transparent)
while the “upper level” is responsible for generating
explanations understandable to both managers and
customers [14].
Below, Table 2 presents examples of practices for
integrating XAI into personalized CRM algorithms and
their expected effects on key metrics.
Table 2. Examples of XAI integration into personalized CRM projects [4, 5, 9, 11, 15]
Practice of XAI
integration
Description of mechanism
Expected effect
1. Visualization of
recommendation
“reasons”
(explanation panel)
• A supplementary window or panel that
describes
the
factors
influencing
the
recommendation, using simple language for
both the customer and the manager
•
Increased
trust;
reduced time required
to explain the product
2.
Feedback
on
feature weights
• Highlighting key parameters (click behavior,
purchase history, demographics) with threshold
visualization of the contribution of each feature
•
Enhanced
transparency;
optimization
of
marketing campaigns
3. “Dual approach”
(ENSEMBLE
+
XAI)
• The use of an ensemble combining a deep
model with an interpretable model, comparing
outputs
and
generating
understandable
explanations
•
A
compromise
between accuracy and
explainability;
reduction of errors
4. Risk scoring
• Providing a “risk rating” for the decision to the
customer or manager, indicating the probability
of model error
• Increased system
“honesty”; enabling
rapid response
Thus, personalized algorithms (recommendation
systems, chatbots, ML models) yield the greatest
benefits when complemented by explainability
mechanisms. This not only enhances customer
satisfaction but also facilitates system scalability within
a company by easing decision-making regarding model
adjustments and settings under changing business
conditions.
3. Implementation and evaluation of effectiveness
The successful implementation of Explainable AI in the
context of CRM personalization largely depends on
specific business scenarios and company capabilities. A
number of studies indicate that voice assistants and
chatbots supplemented with explainable modules are
capable of reducing service costs while simultaneously
increasing customer satisfaction [8,14]. Special
attention in such projects is given to service support: the
transparency of the chatbot algorithm helps managers
better understand the logic behind responses, and
customers receive explanations for recommendations in
an understandable form.
In call centers focused on mass service, methods of
voice analytics and automatic emotion recognition are
actively researched to improve service quality [2]. In
these
cases,
explainable
models
enable the
identification of negative communication episodes and
prompt responses to potentially dissatisfied customers.
For example, a system may signal an increased “stress
level” of the interlocutor and provide the contact center
employee with clear recommendations on the
necessary actions [4].
An interesting example is provided by Peruchini M., da
The American Journal of Engineering and Technology
62
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
Silva G. M., and Teixeira J. M. [1], who analyzed the
implementation of XAI in an online retail company. As
a result of integrating explainable recommendation
systems (combining deep learning with SHAP
interpretations) into the e-commerce platform, both
conversion rates and internal business processes
related to algorithm quality control improved.
Managers gained the ability to independently adjust
product selection priorities for customers based on
transparent metrics of feature contribution [1].
When implementing Explainable AI in CRM,
architectural and organizational characteristics must
be taken into account. According to Wulff K. and
Finnestrand H. [15], successful projects often follow a
phased approach, beginning with small-scale
experiments (proof of concept) before scaling the
solutions. Important factors include:
●
The availability of high-quality data: proper
model explainability requires a consistent set of
customer data [9].
●
Adaptive infrastructure: employing a modular
approach that allows the flexible integration of post-
hoc interpreters (LIME, SHAP) with existing ML
modules [11].
●
Staff training: the introduction of XAI modules
is accompanied by programs aimed at enhancing the
qualifications of managers and contact center operators
so that they can competently work with the
explanations [2].
●
Compliance with confidentiality standards:
explainable
models,
especially
those
handling
emotional or behavioral data, must address issues of
privacy and security [14].
An important stage in the integration of XAI is verifying
the quality and relevance of the generated explanations
for different user groups. The evaluation of systems
employing Explainable AI should consider both technical
and business metrics. Technical metrics typically include
accuracy, recall, F1-score, and other classification
metrics for algorithmic solutions. Business metrics often
comprise conversion, customer retention, growth in
average transaction value, and satisfaction (for
example, Net Promoter Score, NPS) [3]. In CX studies,
emotional indicators reflecting the level of stress or
satisfaction during interactions with a chatbot or
operator are also highlighted [12].
Below, Table 3 provides an example structure for
evaluating
the
effectiveness
of
implementing
Explainable AI in CRM, adapted from several projects
described in the literature.
Table 3. Example of the structure for evaluating the effectiveness of implementing Explainable AI in CRM [2,8]
Evaluation level
Key metrics
Tools and methods
of collection
Expected outcome
Technical accuracy of
the algorithm
• Accuracy, Precision,
Recall, F1-score
• Latency
• CRM system logs
• Reports on ML
module performance
• Improved model
stability
•
Reduction
of
technical errors
Explainability
and
transparency
•
Assessment
of
explanation
clarity
(survey-based)
• Number of requests
for manual review
• Customer/manager
surveys
•
LIME/SHAP
monitoring panels
•
Reduction
in
negative inquiries
• Increased trust
Impact on CX
•
Customer
satisfaction (CSAT)
• Net Promoter Score
(NPS)
• Repeat purchases
(retention)
• Surveys
• Transaction analytics
•
Customer
segmentation analysis
• Growth in loyalty
and recommendations
• Improved brand
perception
The American Journal of Engineering and Technology
63
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
Evaluation level
Key metrics
Tools and methods
of collection
Expected outcome
Financial
and
economic effects
• ROI, ROMI
•
Revenue
from
additional sales
• Cost savings in
contact centers
• Financial reports
•
BI
system
dashboards
• Demonstration of
company benefits
• Optimization of
operational costs
Thus, experience shows that the implementation of
XAI in CRM, particularly for personalization algorithms,
makes a measurable contribution to key business
metrics by increasing customer satisfaction, reducing
operational costs, and fostering growth in trust. At the
same time, a comprehensive evaluation approach is
necessary to assess both technological and marketing
aspects. The successful implementation of explainable
approaches in CRM personalization requires clear
planning at both the architectural and organizational
levels, as well as consistent monitoring of outcomes
based on diverse metrics. This approach not only
minimizes risks associated with insufficient AI
transparency but also enables the targeted
development of competitive advantages by focusing
on customer satisfaction, trust, and goodwill.
CONCLUSION
The conducted research confirmed the importance of
explainable artificial intelligence (XAI) algorithms for
creating and maintaining a high level of customer
experience (CX) in CRM systems. Standard
personalization technologies, whether chatbots,
recommendation engines, or emotion analysis
systems, demonstrate high efficiency; however, they
suffer from the “black
-
box” effect, which diminishes
user trust and complicates the interpretation of
outputs. Therefore, integrating XAI approaches
becomes a critical condition not only for increasing
conversion and retention but also for enhancing
customer satisfaction and trust.
The analysis of practical cases showed that transparent
algorithms enable managers to respond more
promptly to problematic situations, better understand
the logic behind suggestions and recommendations,
and allow end users to feel more secure when
interacting with digital services. This directly impacts
business performance indicators: CSAT and NPS
increase, contact center costs decrease, and the ROI
from implementing CRM solutions improves.
Thus, the comprehensive application of Explainable AI in
CRM systems opens up new opportunities for
developing
competitive
advantages.
The
methodological approaches developed, the metrics
proposed during the study, and the practical
recommendations can serve as a guide for companies
seeking to implement or refine AI technologies in
managing customer relationships. In the future, further
research may focus on the development of generative
XAI models as well as on the ethical and legal aspects of
ensuring algorithmic transparency and accountability.
REFERENCES
Peruchini M., da Silva G. M., Teixeira J. M. Between
artificial intelligence and customer experience: a
literature review on the intersection //Discover Artificial
Intelligence.
–
2024.
–
Vol. 4 (1).
–
pp. 4.
Guo Y. et al. Measuring service quality based on
customer emotion: An explainable AI approach
//Decision Support Systems.
–
2024.
–
Vol. 176.
–
pp. 1-
10.
Ameen N. et al. Customer experiences in the age of
artificial intelligence //Computers in human behavior.
–
2021.
–
Vol. 114.
–
pp. 1-9.
Dwivedi Y. K. et al. Artificial Intelligence (AI):
Multidisciplinary perspectives on emerging challenges,
opportunities, and agenda for research, practice and
policy
//International
journal
of
information
management.
–
2021.
–
Vol. 57.
–
pp. 1-12.
Abdelkader O. A. ChatGPT's influence on customer
experience in digital marketing: Investigating the
moderating roles //Heliyon.
–
2023.
–
Vol. 9 (8).
–
pp.1-
11
Puntoni S. et al. Consumers and artificial intelligence: An
experiential perspective //Journal of marketing.
–
2021.
–
Vol. 85 (1).
–
pp. 131-151.
Neuhofer B., Magnus B., Celuch K. The impact of
artificial intelligence on event experiences: a scenario
technique approach //Electronic Markets.
–
2021.
–
Vol.
The American Journal of Engineering and Technology
64
https://www.theamericanjournals.com/index.php/tajet
The American Journal of Engineering and Technology
31 (3).
–
pp. 601-617.
Abdo A., Yusof S. M. Exploring the impacts of using the
Artificial Intelligence voice-enabled chatbots on
customers interactions in the United Arab Emirates
//IAES Int J Artif Intell.
–
2023.
–
pp.1-12.
McLean G., Osei-
Frimpong K. Hey Alexa… examine the
variables influencing the use of artificial intelligent in-
home voice assistants //Computers in human
behavior.
–
2019.
–
Vol. 99.
–
pp. 28-37.
Robinson S. et al. Frontline encounters of the AI kind:
An evolved service encounter framework //Journal of
Business Research.
–
2020.
–
Vol. 116.
–
pp. 366-376.
Ho S. P. S., Chow M. Y. C. The role of artificial
intelligence in consumers’ brand preference for retail
banks in Hong Kong //Journal of Financial Services
Marketing.
–
2023.
–
pp. 1.
Ceccacci S. et al. Emotional valence from facial
expression as an experience audit tool: an empirical
study in the context of opera performance //Sensors.
–
2023.
–
Vol. 23 (5).
–
pp. 2-8.
Nguyen D. M., Chiu Y. T. H., Le H. D. Determinants of
continuance intention towards banks’ chatbot services
in Vietnam: A necessity for sustainable development
//Sustainability.
–
2021.
–
Vol. 13 (14).
–
pp. 1-7.
Bhattacharya C., Sinha M. The role of artificial
intelligence in banking for leveraging customer
experience //Australasian Accounting, Business and
Finance Journal.
–
2022.
–
Vol. 16 (5).
–
pp.1-17.
Wulff K., Finnestrand H. Creating meaningful work in
the age of AI: explainable AI, explainability, and why it
matters to organizational designers //AI & SOCIETY.
–
2024.
–
Vol. 39 (4).
–
pp. 1843-1856.
Zhou J. et al. Inferring multi-stage risk for online
consumer credit services: an integrated scheme using
data augmentation and model enhancement
//Decision Support Systems.
–
2021.
–
Vol. 149.
–
pp. 1-
10.
Phillips-Wren G., Daly M., Burstein F. Reconciling
business intelligence, analytics and decision support
systems: More data, deeper insight //Decision Support
Systems.
–
2021.
–
Vol. 146.
–
pp. 1-8.
