Content uploaded by Vishvesh Soni
Author content
All content in this area was uploaded by Vishvesh Soni on Dec 18, 2023
Content may be subject to copyright.
RCBA-2023 Volume 6, Issue 1
P a g e | 133
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
Impact of Generative AI on Small and Medium
Enterprises' Revenue Growth: The Moderating Role
of Human, Technological, and Market Factors
Vishvesh Soni
Marketing specialist and market research analyst at Alliage Inc.
Abstract
Background: The generative artificial intelligence (AI) technologies have become strategic tools for small
businesses seeking maintain competitive advantage. The usefulness of of these technologies are well-
recognized in SMEs, yet it is becoming important to explore how other factors might affect the degree to
which these benefits are realized. The present study emerges from this necessity, aiming to provide a
quantitative assessment of how generative AI adoption affects SME revenue growth and to what degree
this effect relies on human capital, technological infrastructure, and market competition.
Objectives: The aim of this research is to empirically examine not only the direct effects generative AI on
revenue growth but also how this relationship is shaped by several moderating factors such as human,
technological, and market factors. These factors can either amplify or diminish the potential gains from
generative AI adoption.
Data and Methods: To understand the relationships, data from 331 SMEs were analyzed using 3
Regularization regression methods, namely, Ridge, Lesso, and Elastic Net Regression methods.
Findings: The results indicates that companies benefit from adopting generative AI technologies. The
moderating effects of human capital indicates that businesses not only benefit from adopting generative AI
but do so especially when they have highly educated employees. This implies that human capital can
enhance or is complementary to the advantages provided by the generative AI, possibly through more
effective utilization. The moderating effects of existing firm’s infrastructure also has a positive effect,
suggesting that the benefits of generative AI are amplified when a business has good existing technological
infrastructure. This means that businesses with modern or advanced tech facilities can leverage AI
technology more effectively than those with outdated or less capable infrastructure. The moderating effects
of market competition showed a negative result indicating that the advantage gained from generative AI
adoption may decrease as market competition intensifies. This suggests that in highly competitive markets,
the edge provided by AI is less distinct, perhaps because competitors are also likely to adopt similar
technologies, negating the competitive advantage.
Conclusion:
The findings suggest that simply deploying generative AI will not suffice; instead, it should be part of a
broader strategy that considers market dynamics, skilled human capital to operate, and improving existing
technological infrastructure.
Keywords:
1. Artificial Intelligence Adoption
2. Competitive Markets
3. Generative AI
4. Revenue Growth
5. Small and Medium Enterprises (SMEs)
6. Technological Infrastructure
Introduction
The surge in artificial intelligence (AI) tools has significantly impacted how business
organizations operate, marking a transition from traditional processes to more
technology-driven approaches. These AI-driven systems are primarily defined by their
capacity to automate routine tasks, enabling organizations to allocate human resources
RCBA-2023 Volume 6, Issue 1
P a g e | 134
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
to more strategic initiatives [1]–[3]. The automation spectrum ranges from simple tasks
like data entry to more complex processes like supply chain management, where
predictive algorithms can anticipate disruptions and suggest mitigations. AI's role in
knowledge extraction from large datasets has been particularly transformative, allowing
businesses to gain actionable insights that were previously obscured by the volume of
information. AI-powered analytics have transcended human analytical capabilities,
offering a level of precision and efficiency that can drastically enhance decision-making
processes. The integration of AI into business systems has, therefore, not just been a
matter of adopting new technologies, but has fundamentally altered how organizations
conceive and execute their operations. The consequences of AI's integration into
business models are extensive, leading to the emergence of novel business models and
consumer offerings. The incorporation of AI tools has enabled businesses to reimagine
their products and services, resulting in innovative solutions that were once thought
impossible.
Table 1. An Overview of Generative Modeling Frameworks in AI
Generative Model
Key Components
Training Process
Output
Quality
Speed of
Generation
Best Use Cases
GAN (Generative
Adversarial
Network)
Generator Network,
Discriminator
Network
Adversarial training until
the discriminator cannot
distinguish synthetic
content [4]
High
Fast
High-fidelity images,
domain-specific data
generation
Diffusion Models
(DDPMs)
Forward Diffusion,
Reverse Diffusion
Add noise to data, then
reverse to reconstruct
original data [5]
Very High
Slow due to
complex
training
High-quality image
and audio generation,
foundation models
VAE (Variational
Autoencoder)
Encoder, Decoder
Encode input into latent
space and decode to
reconstruct [6]
Moderate
Fast
Quick data generation
with less detail
required
GPT (Generative
Pre-trained
Transformer)
Transformer
Architecture
Pre-training on large
datasets, followed by fine-
tuning [7]
High (for
text)
Varies with
application
scale
Natural language
generation, text-based
AI applications
Generative artificial intelligence (AI) encompasses a range of technologies that can
create new content, such as text, images, or various forms of media, through the use of
advanced algorithms known as generative models. The inception of this technology
dates back to the 1960s with the creation of early chatbots, which represented the
primitive stages of AI's ability to produce human-like text. However, it was not until
the development of Generative Adversarial Networks (GANs) in 2014 that generative
AI truly began to demonstrate a remarkable capacity to synthesize highly realistic
images, videos, and audio recordings [8]. These GANs employ a unique machine
learning approach, involving dueling neural networks, to improve the quality of
artificial creations to the point where they are often indistinguishable from authentic
human-generated content.
Generative modeling artificial intelligence (GAI) represents a transformative subset of
machine learning that diverges from traditional supervised learning. These models,
operating either with minimal human supervision or without any at all, leverage
statistical methods and probabilistic frameworks to create new, artificial artifacts. By
RCBA-2023 Volume 6, Issue 1
P a g e | 135
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
analyzing vast amounts of digital content—from text and images to audio and video—
GAI identifies patterns, assimilates the distribution of the input data, and then generates
novel outputs that reflect the learned data characteristics. These outputs are not just
mimicries but often indistinguishable from their real-world counterparts, blurring the
lines between what is generated and what is organic. The development of deep learning
(DL) techniques has been pivotal to the evolution of GAI, allowing the creation of
complex and nuanced artificial content. The field has made significant strides,
particularly in the areas of content generation and enhancement, enabling applications
that span from artistic image synthesis to the augmentation of virtual realities.
Two architectures within GAI are the Generative Adversarial Network (GAN) and the
Generative Pre-trained Transformer (GPT), each with unique mechanisms and
applications [9]. GANs function through a dynamic training system involving two
competing neural networks: the generator, which creates synthetic data, and the
discriminator, which evaluates the authenticity of this data. This interplay progresses
iteratively; the generator learns to produce increasingly convincing artifacts, while the
discriminator becomes more adept at detecting nuances that differentiate the artificial
from the real. The training continues until the discriminator can no longer reliably
identify synthetic content, effectively accepting it as real. This adversarial process has
been instrumental in generating highly realistic images, videos, and voices, making
GANs the prevailing technique.
Diffusion models, especially denoising diffusion probabilistic models (DDPMs), add
another aspect to the GAI. Through a sophisticated two-phase training methodology—
forward diffusion that corrupts the training data with noise, and reverse diffusion that
seeks to recover the data by removing the noise—these models learn to generate new
samples from pure noise [10]. This iterative noise addition and subtraction process
endows diffusion models with the potential to train numerous layers, conferring upon
them the capacity to produce outputs of remarkable quality. Although their training is
more time-consuming compared to other models like variational autoencoders (VAEs),
the fidelity of the outputs often justifies the investment in computational resources and
time.
Diffusion models have also gained prominence as foundation models because they are
scalable, versatile, and produce high-fidelity results that are applicable to a variety of
generalized use cases. The significant computational demands of these models,
stemming from their reverse sampling process, mean they are not the quickest. The
quality of the end product is frequently superior to that of other generative models.
VAEs introduce another approach to generative modeling by focusing on encoding
input data into a compressed latent space before reconstructing it back to its original
form through a decoder. This encoder-decoder architecture ensures that only the most
relevant features of the data are captured and preserved, enabling the model to generate
new data that mirrors the original input. While VAEs excel in the rapid generation of
new data instances, their output often lacks the detailed precision found in diffusion
models. Nonetheless, their efficiency in generating content quickly and their relative
RCBA-2023 Volume 6, Issue 1
P a g e | 136
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
simplicity make them a valuable tool for tasks that require faster output generation
without the necessity for minute detail.
Table 2. Use cases of Generative AI in SMEs
Use case
Use Case Description
Content Creation
- Marketing Content: Generate text for ads, social media, and campaigns.
- Blogging and SEO: Create draft articles that are SEO-friendly.
Personalized Customer
Communications
- Generate personalized emails and messages based on customer data.
Product Design and Development
- Create new product designs or modify existing ones based on feedback and
trends.
Graphic Design
- Design logos, marketing materials, and other elements without a full-time
designer.
Prototyping and 3D Modeling
- Assist in creating prototypes for rapid iteration and testing.
Automation of Paperwork and
Reports
- Draft reports, generate invoices, and handle routine paperwork tasks.
Customer Support
- Provide first-level support through AI chatbots.
Language Translation
- Translate content into multiple languages to aid operation in different regions
without language barriers.
Generative AI offers substantial utility in enhancing the operational efficiencies and
marketing endeavors of Small and Medium-sized Enterprises (SMEs). AI-facilitated
text generation for advertising, social media, and campaigns enables these enterprises
to produce a consistent and engaging digital presence with minimal human capital
investment. Moreover, the capability of generative AI to compose search engine
optimization (SEO)-friendly blog posts augments an SME's visibility and searchability
online, which is critical for capturing and maintaining digital market share. Such
technological leverage in content creation allows SMEs to reallocate financial and
human resources to strategic growth areas, fostering an environment conducive to
revenue enhancement by broadening their digital footprint and consumer base.
Generative AI is instrumental crafts individualized emails and messages by intelligently
analyzing customer data, facilitating a level of bespoke engagement that often surpasses
that of larger corporations. This heightened personalization can translate into customer
loyalty and increased transaction frequency, both pivotal for an SME's revenue.
Concurrently, generative AI expedites the product design and development process. It
autonomously generates innovative product designs or iteratively refines existing ones,
informed by consumer feedback and emergent trends. This accelerates the product life
cycle, empowering SMEs to more swiftly adapt to market demands and expedite time-
to-market for new offerings, enhancing competitive advantage and revenue potential.
Furthermore, generative AI's application in graphic design and prototyping presents
cost-effective solutions for SMEs. By facilitating the creation of marketing materials
and logos, adopter SMEs avoids the substantial expenditures associated with
professional design services. In prototyping and 3D modeling, generative AI aids in the
rapid production and modification of prototypes, substantially reducing the costs and
time associated with traditional prototyping methods. The automation of routine
RCBA-2023 Volume 6, Issue 1
P a g e | 137
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
documentation and report generation similarly mitigates administrative burdens,
allowing SMEs to economize on operational costs and minimize human error. These
efficiencies not only optimize resource allocation but also maximizes the capacity for
business growth and profit maximization.
Conceptual framework
This study argues that the successful implementation and utilization of Generative AI
is heavily moderated by the human capital within these organizations. Human capital,
defined as the stock of expertise, knowledge, skills, and social attributes, is critical in
leveraging AI tools effectively. The proficiency of employees in understanding and
interfacing with advanced AI systems can significantly dictate the extent to which these
technologies enhance productivity and creative capacity. For SMEs, where resource
constraints are common, the quality of human capital assumes an even greater
significance, as it becomes a pivotal factor in ensuring that AI adoption leads to actual
economic and strategic benefits rather than merely adding to the technological
overhead.
The agility offered by a skilled workforce is crucial in adapting to the disruptive nature
of Generative AI. Such a workforce can pivot and align the organization's operational
strategies with the capabilities of AI, ensuring that technology acts as a complement to
human effort rather than a replacement [11]. This synergetic relationship is essential as
it can lead to the creation of new products, services, and business models that are
informed by AI-driven insights yet curated by human expertise and understanding of
market nuances. Furthermore, employees who are adept in utilizing AI can perform
higher-level tasks, delegate routine processes to AI systems, and focus on complex
decision-making and strategic planning. This not only enhances efficiency but also
fosters an environment of continuous learning and innovation, which is vital for SMEs
to maintain a competitive edge in their respective sectors. The capacity of human capital
to adapt to and integrate generative AI tools can also enhance customer experiences and
open up new revenue channels for SMEs. Service-oriented SMEs, in particular, can
benefit from the personalized and efficient customer interactions that AI enables.
Employees who can skillfully manage AI tools can offer swift and accurate responses
to customer inquiries, leading to higher satisfaction rates and repeat business, thereby
increasing customer lifetime value. Sales and customer service representatives
equipped with AI-generated insights can provide customized recommendations,
improving conversion rates and boosting sales figures. The emotional intelligence and
ethical decision-making capabilities of humans are irreplaceable assets in interpreting
AI-generated data and results, making strategic decisions that align with the core values
and mission of the SME, and maintaining the trust of customers and stakeholders in an
increasingly technology-driven business environment.
RCBA-2023 Volume 6, Issue 1
P a g e | 138
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
Figure 1. Conceptual framework
The existing technological infrastructure within small and medium-sized enterprises
(SMEs) plays a significant role in moderating the impact of generative AI adoption on
these organizations. Infrastructure readiness can either accelerate the benefits or
become a bottleneck in the effective deployment of AI technologies. On one hand,
SMEs with robust and scalable IT systems can seamlessly integrate generative AI into
their operations, allowing them to rapidly leverage the technology for enhanced data
processing, product development, and customer service. This seamless integration can
lead to immediate improvements in efficiency and productivity, which can translate into
cost savings and potentially increased revenue. Conversely, SMEs with outdated or
limited IT capabilities may struggle to support the advanced computational and data
storage needs of generative AI, potentially leading to disruptions and increased costs.
Furthermore, the current technological setup determines an SME's agility in adopting
new tools and solutions. A modular and interoperable IT architecture allows for plug-
and-play integration of AI systems, reducing the time and resources needed for
implementation [12]. This agility is critical for maintaining competitiveness, as it
enables SMEs to rapidly adapt to market changes and technological advancements. For
example, if an SME's infrastructure is already cloud-enabled, it can easily scale up its
use of AI services offered by cloud providers, benefiting from the latest developments
without the need for substantial upfront investment. In contrast, SMEs with rigid
systems may find themselves facing significant overhauls or custom development, both
of which can be costly and time-consuming, potentially delaying the realization of AI
benefits.
In markets where competition is intense, the rapid deployment of generative AI by
SMEs can quickly transform from a unique strategic asset to a universal requirement.
The initial competitive advantage offered by AI's adoption fades as more competitors
harness similar technologies. This rapid leveling of the playing field means that SMEs
must adopt AI not to lead the market but merely to keep pace with it. As such, the
technology's adoption becomes less about seeking a competitive edge and more about
not falling behind. Consequently, SMEs are pressured to continuously evolve their AI
RCBA-2023 Volume 6, Issue 1
P a g e | 139
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
capabilities to find new avenues for differentiation as the technology itself becomes a
common commodity in the industry.
With generative AI becoming a standard tool across competitors, the marginal benefits
for each enterprise begin to decrease. When all players in the market deploy AI to
optimize their operations and enhance product offerings, the unique value that AI could
provide when less widespread is significantly reduced. For SMEs, this saturation leads
to smaller gains from AI, as operational efficiencies and innovations become uniform
across the board. In this environment, the focus shifts from the adoption of generative
AI to maximizing its efficiency and finding innovative ways to apply its outputs to
create value that can distinguish an SME from its rivals.
Moreover, the financial strain of operating in a highly competitive market impacts
SMEs’ ability to invest in generative AI. With a relentless focus on reducing prices to
stay competitive, SMEs often operate on slim margins, which can restrict their capacity
for significant investments in new technologies. As a result, SMEs might opt for more
affordable, less advanced AI solutions that do not fully exploit the technology's
potential due to budgetary limitations. This cost-sensitive approach to AI investment
means that while SMEs may implement generative AI to some extent, their ability to
benefit from its full suite of capabilities is limited, potentially curbing the more
profound transformational effects that AI could have on revenue and growth.
Method
Building upon the discussion presented in the conceptual framework, three moderating
variables have been incorporated to represent proxies for human capital, technological
capacity, and market factors. Tables 3 and 4 display a comprehensive list of the
independent variables and the corresponding interaction terms, respectively. Detailed
descriptions of each variable are also provided within these tables. The following
equation present the basic model formulation of the study:
Where, βTIL is the coefficient for the Generative AI Integration Level (TIL).
βED is the coefficient for the Establishment Duration (ED).
βIA is the coefficient for Infrastructure Adjacency (IA).
βITS is the coefficient for International Trade Status (ITS).
βCO is the coefficient for Capital Origination (CO).
βGBI is the coefficient for Generational Business Indicator (GBI).
0 TIL ED IA ITS
CO GBI HCL PET
CTI
TIL ED IA ITS
CO GBI HCL+ PET
CTI+ MKT
YRI
MKT
= + + + + +
+ + +
+
RCBA-2023 Volume 6, Issue 1
P a g e | 140
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
βHCL is the coefficient for Human Capital Level (HCL).
βPET is the coefficient for Proprietor Expertise Tenure (PET).
βCTI is the coefficient for Current Technological Infrastructure (CTI).
ΒMKT is the coefficient for the Market Competitor Density (MKT).
With the interactions terms (moderating variables),
Where, βGen*HCL, βGen*CTI, βGen*MKT are the coefficients for the interaction terms.
1i
k
Z
i
Z
=
represents the sum of the products of the coefficients and the other
independent variables (denoted as Z) in the model. ϵ is the error term.
Table 3. the dependent and independent variables
S
L
Variable Name
Symbo
l
Type
Description
1
Yearly
Revenue
Increment
YRI
Continuous,
Dependent
This represents the percentage increase or decrease
in the revenue of a business on an annual basis.
2
Generative AI
Integration
Level
TIL
Categorical,
Independent
Indicates whether the business has implemented
Generative Artificial Intelligence technology, where
a value of 0 signifies no adoption, and a value of 1,
and 2 signify moderate and high adoption.
3
Establishment
Duration
ED
Continuous,
Independent
Denotes the number of years the business has been
operational.
4
Infrastructure
Adjacency
IA
Dummy,
Independent
Assess the business's location in relation to major
roadways, with 0 indicating no proximity to main
roads and 1 indicating close proximity.
5
International
Trade Status
ITS
Dummy,
Independent
Designates whether the business is engaged in
international trade of its products or services. The
value is 0 if it does not export, and 1 if it does.
6
Capital
Origination
CO
Dummy,
Independent
Describes the origin of the business's capital. A
value of 0 means the capital is sourced internally,
and a 1 means it is sourced from external credit
institutions.
7
Generational
Business
Indicator
GBI
Dummy,
Independent
Identifies whether a business is family-owned with
0 for non-family-owned and 1 for family-owned
enterprises.
8
Human Capital
Level
HCL
Continuous,
Independent
The average academic achievement level of the
employees, measured by the highest educational
qualification obtained.
9
Current
Technological
Infrastructure
CTI
Continuous,
Independent
The existing technological assets and systems
within the business, quantitatively measured by age,
capability, and capacity.
0 Gen*HCL Gen*CTI Gen*MKT 1
Gen*HCL Gen*CTI Gen*MKT i
k
Zi
i
YRI Z
=
= + + + + +
RCBA-2023 Volume 6, Issue 1
P a g e | 141
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
Table 4. interaction terms (moderating variables)
S
L
Variable
Name
Notation
Type
Description
10
Generative
AI and
Human
Capital
Interaction
Gen*HCL
Continuous,
Independent
This captures the combined effect of Generative
AI Adoption and the level of employee education
or expertise, hypothesizing that skilled
employees may utilize AI technologies more
proficiently.
11
Generative
AI and
Infrastructur
e Interaction
Gen*CTI
Continuous,
Independent
Represents the interaction between Generative
AI Adoption and the Current Technological
Infrastructure, investigating whether entities with
established technological resources benefit more
from Generative AI.
12
Generative
AI and
Market
Competition
Interaction
Gen*MK
T
Continuous,
Independent
Explores the potential differential impact of
Generative AI Adoption in relation to the
intensity of Market Competition, distinguishing
effects in various competitive environments.
Regularized regressions
Ridge Regression, also known as L2 Regularization, stabilizes the regression estimates
in such a way that it reduces the standard errors by imposing a penalty on the size of
coefficients. The Ridge Regression function to be minimized is [13]:
where Y is the response variable, X the matrix of predictors, β the vector of coefficients,
and λ the regularization parameter. The regularization term
2
1
p
j
j
=
penalizes the
magnitude of the coefficients and effectively shrinks them towards zero. However,
unlike Lasso Regression, the Ridge penalty tends to shrink the coefficients evenly and
does not set them to zero, thus, all variables are kept in the model. This technique is
particularly useful when there is a need to retain all features in the model but still
mitigate the problem of multicollinearity.
Lasso Regression, known as L1 Regularization, is another linear regression technique
that includes a penalty term to the loss function, but with a different approach to
constraint coefficient estimates. The Lasso's objective function can be written as [14]:
In the equation, Y represents the vector of observations, X is the predictor matrix, β
stands for the coefficient vector, and λ is the regularization parameter. The Lasso
technique differs from Ridge Regression by the type of penalty it applies; the L1
penalty,
1
||
p
j
j
=
, encourages sparsity in the model by allowing some of the coefficient
estimates to be exactly zero.
22
11
Ridge( , ; ) ( ) p
nT
i i j
ij
Y X y x
==
= − +
2
11
Lasso( , ; ) ( ) | |
p
nT
i i j
ij
Y X y x
==
= − +
RCBA-2023 Volume 6, Issue 1
P a g e | 142
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
Elastic Net Regression combines the strengths of Lasso and Ridge regression methods
into a single type of regularized linear regression. It is highly effective when dealing
with a scenario where there are more predictors than observations or when there is a
high degree of correlation among the predictors. The goal of the Elastic Net is to
minimize the total of the squared differences between observed and predicted values,
which is akin to what is done in ordinary least squares (OLS). However, it distinguishes
itself by adding two extra terms to the function it seeks to minimize, which are
specifically for the purpose of regularization. [15]. The objective function in Elastic Net
Regression is:
where α represents the regularization parameter, while ρ determines the equilibrium
between the Lasso and Ridge regularization methods. These parameters are crucial in
adjusting the model to achieve the desired level of regularization, with α controlling the
overall strength and ρ setting the proportion of the mix between the two techniques
The regularization component within the objective function of Elastic Net is composed
of an L1 penalty term (|θj|) and an L2 penalty term (θj²). The L1 penalty promotes
sparsity within the model, potentially shrinking some coefficients to zero, effectively
omitting those predictors. This is particularly advantageous in datasets with a large
number of dimensions, aiding in feature selection. Conversely, the L2 penalty shrinks
coefficients towards zero but usually not to zero, which is useful for dealing with
predictor variables that are highly correlated. The mixing parameter ρ provides the
means to find a middle ground between Lasso and Ridge regularization methods. At
ρ=1, the Elastic Net is equivalent to Lasso regression, and at ρ=0, it corresponds to
Ridge regression [16], [17]. Thus, Elastic Net allows for more nuanced model
adjustment.
In Elastic Net Regression, estimating parameters is commonly done using optimization
techniques like gradient descent or coordinate descent. The convexity of the objective
function guarantees that global minimization is attainable. Selecting the appropriate
values for the regularization parameter α and the mixing parameter ρ typically involves
methods such as cross-validation. A grid search may be conducted over a spectrum of
α and ρ values to find the pair that minimizes cross-validation error. This process
enables Elastic Net to be versatile and effective for different data sets and modeling
problems.
Results
The results from Ridge Regression are provided in Table 5 and 6. The R-squared value
of 0.81 shows that the model accounts for 81% of the variance in the dependent variable.
The Adjusted R-squared remains at 0.76, reaffirming that after adjusting for the number
( ) ( )
22
1 1 1
1
1()
22
n m m
T
i i j j
i j j
J y x
n
= = =
−
= − + +
RCBA-2023 Volume 6, Issue 1
P a g e | 143
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
of predictors, the model still retains a strong explanatory power. The Mean Squared
Error (MSE) slightly improved to 8.94, along with a minimal decrease in the Root Mean
Squared Error (RMSE) to 2.99 and a steady Mean Absolute Error (MAE) at 2.31,
suggesting a consistent average deviation of the predictions from the actual values. The
Cross-Validation Score is unchanged at 0.77, which aligns with the adjusted R-squared
value, indicating the model's performance is robust across different subsets of the
dataset. However, the Akaike Information Criterion (AIC) and Bayesian Information
Criterion (BIC) have increased significantly to 292.87 and 323.74, respectively. This
increase could indicate a model with higher complexity or a different model structure
altogether when compared to the previous ones.
Table 5. Model performance of Ridge Regression
Metric
Value
R-squared
0.81
Adjusted R-squared
0.76
Mean Squared Error (MSE)
8.94
Root Mean Squared Error (RMSE)
2.99
Mean Absolute Error (MAE)
2.31
Cross-Validation Scores
0.77
Akaike Information Criterion (AIC)
292.87
Bayesian Information Criterion (BIC)
323.74
Table 6. Feature importance in Ridge regression
Feature
Importance Score
Gen*CTI
3.13
ED
2.57
Gen
1.75
Gen*HCL
1.46
CTI
1.24
HCL
0.78
IA
0.68
ITS
0.27
GBI
-0.17
CO
-0.18
MKT
-0.74
IC
-0.77
Gen*MKT
-1.70
Gen*CTI still holds the highest positive importance score but has decreased slightly to
3.13. ED's importance has increased, suggesting a greater relevance in this model
iteration. Gen remains an important feature with a small decrement in its score, while
Gen*HCL has seen an increase, indicating a stronger impact on the model's output. CTI
continues to be a positive predictor but with a slightly increased score, while HCL and
IA both have positive importance, although IA's score has decreased. ITS shows a minor
positive importance. GBI and CO have small negative scores, reflecting a minimal but
negative relationship with the dependent variable. MKT, IC, and Gen*MKT hold
negative importance values, with Gen*MKT having the largest negative importance,
RCBA-2023 Volume 6, Issue 1
P a g e | 144
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
which has increased from the previous model. This suggests that these features are
inversely related to the target variable and that their role may be more significant in this
iteration of the model.
Interaction terms like Gen*CTI and Gen*HCL remain significant, emphasizing the
importance of these combined features in the model's predictive capability. The
negative importance of Gen*MKT has grown, which could imply a stronger inverse
relationship between this interaction term and the dependent variable in this model.
Figure 2. Evaluation plots for Ridge Regression
The Residual Plot in the top left quadrant in figure 2 demonstrates the relationship
between the predicted values of the dependent variable (YRI) and the residuals,
which are the differences between the observed and predicted values. The relatively
random dispersion of points around the horizontal axis suggests that the model's
errors are distributed fairly evenly for different levels of predicted values, without
any obvious pattern. This lack of systematic structure is indicative of a well-fitting
model. However, there's a slight trend in the data as indicated by the red line,
suggesting that the model may systematically over or underestimate the YRI across
the range of predictions.
RCBA-2023 Volume 6, Issue 1
P a g e | 145
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
The Prediction Error Plot in the top right quadrant is used to compare the actual and
predicted values. Ideally, the points should fall along the dashed line, where the
predicted values are equal to the actual values. The plot shows that the model predicts
YRI reasonably well, as most points are clustered around the line, but there are
deviations, particularly for higher values. The Learning Curve in the bottom left
quadrant shows the model's performance on the training and validation sets as the
training size increases. The convergence of the training and cross-validation scores
suggests that adding more data might not significantly improve the model's
performance. The Ridge Coefficients Path in the bottom right quadrant illustrates
how the model's coefficients change with different regularization strengths (alpha).
As alpha increases, the coefficients of certain features are driven towards zero, which
is characteristic of Ridge Regression's ability to reduce overfitting by penalizing large
coefficients. The vertical line indicates the alpha value chosen for the final model,
balancing the need to penalize large coefficients while retaining predictive power.
Table 7. Model performance of Lasso Regression
Metric
Value
Adjusted R-squared
0.76
Mean Squared Error (MSE)
8.98
Root Mean Squared Error (RMSE)
3.00
Mean Absolute Error (MAE)
2.31
Cross-Validation Scores
0.77
Akaike Information Criterion (AIC)
175.08
Bayesian Information Criterion (BIC)
205.94
Table 8. Feature importance in Lasso Regression
Feature
Importance
Gen*CTI
3.27
ED
2.49
Gen
1.79
CTI
1.10
Gen*HCL
1.01
HCL
0.84
IA
0.83
ITS
0.18
CO
-0.16
GBI
-0.25
IC
-0.70
MKT
-0.84
Gen*MKT
-1.39
The results from Lasso Regression are presented in tables 7 and 8. The model's adjusted
R-squared value stands at 0.76, indicating a strong explanatory power for the variance
of the dependent variable. The Mean Squared Error (MSE) is calculated to be 8.98, with
a Root Mean Squared Error (RMSE) of 3.00 and a Mean Absolute Error (MAE) of 2.31,
which all point to the model's predictions being relatively close to the actual values. The
Cross-Validation Score is similar to the adjusted R-squared at 0.77, suggesting the
model's predictive stability. The feature importance scores reveal that Gen*CTI is the
RCBA-2023 Volume 6, Issue 1
P a g e | 146
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
most positively influential feature in the model. Following this, ED, Gen, and CTI also
show substantial positive importance, indicating they are significant predictors within
the model. Features such as Gen*HCL, HCL, IA, and ITS are also positively valued but
have a smaller impact. In contrast, CO, GBI, IC, MKT, and Gen*MKT have negative
importance values, suggesting a decrease in the dependent variable as these increase or
a complex relationship where their presence may enhance the predictive quality of the
model due to interactions with other variables. The interaction terms such as Gen*CTI,
Gen*HCL, and Gen*MKT imply that the combined effect of these variables
significantly influences the model's output and their relationships with the dependent
variable are not simply additive but interactive.
Figure 3. Evaluation plots for Lasso Regression
The Residual Plot in the top left reveals the distribution of the residuals, which are
the differences between the observed and predicted values. The residuals are
scattered around the zero line without any clear pattern, suggesting that the Lasso
Regression model does not suffer from systematic errors or bias across the range of
predictions. The concentration of residuals around the zero line also indicates that
there are no extreme errors in prediction, which is desirable. Nevertheless, some
RCBA-2023 Volume 6, Issue 1
P a g e | 147
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
structure in the residuals could indicate potential improvements in model fit, such as
non-linearity in the data that the model does not currently capture.
The Prediction Error Plot in the top right quadrant is a visual representation of the
accuracy of the Lasso Regression model's predictions. The fact that most points are
clustered around the identity line (where actual values equal predicted values)
suggests that the model has a good level of accuracy. However, there are deviations
from the line, especially at the higher end of the scale, which implies that the model
is less accurate at predicting higher values of the dependent variable. The Learning
Curve on the bottom left indicates that the training and cross-validation scores are
converging, suggesting that the model is stabilizing and that additional training data
is unlikely to improve the model's performance significantly. The Elastic Net Path on
the bottom right shows the trajectory of the model's coefficients as the regularization
strength is varied. The blue lines represent individual feature coefficients, and their
convergence towards zero indicates the model's increasing preference for simplicity
and feature selection as regularization becomes more stringent. This path helps in
understanding the impact of regularization on the model complexity and feature
selection.
Table 9. Model performance of Elastic Net Regression
Metric
Value
Adjusted R-squared
0.76
Mean Squared Error (MSE)
9.1
Root Mean Squared Error (RMSE)
3.02
Mean Absolute Error (MAE)
2.32
Cross-Validation Scores
0.77
Akaike Information Criterion (AIC)
175.95
Bayesian Information Criterion (BIC)
206.82
Table 10. Feature importance in Elastic Net regression
Feature
Importance
Gen*CTI
3.00
ED
2.42
Gen
1.38
Gen*HCL
1.24
CTI
1.18
IA
0.84
HCL
0.77
ITS
0.18
CO
-0.18
GBI
-0.24
IC
-0.68
Gen*MKT
-0.95
MKT
-1.00
The adjusted R-squared value of 0.76 shows that the model explains a significant
proportion of the variance in the dependent variable. The Mean Squared Error (MSE)
has increased slightly to 9.1 from the previous set of results, as has the Root Mean
RCBA-2023 Volume 6, Issue 1
P a g e | 148
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
Squared Error (RMSE), now at 3.02, and the Mean Absolute Error (MAE) at 2.32,
indicating a marginal increase in the average errors of the model's predictions.
However, the Cross-Validation Score remains consistent at 0.77, suggesting that the
model's performance is stable across different data subsets. The small increments in the
Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to
175.95 and 206.82, respectively, might be inconsequential without a direct comparison
to other models, but they suggest a slight decrease in the model's relative quality.
Gen*CTI remains the most positively influential feature, albeit with a slightly reduced
score from the previous result. ED, Gen, and CTI are also key positive contributors to
the model's predictions, with Gen having a noticeably lower score than in the previous
set. Gen*HCL and CTI show a small increase in importance, indicating their more
significant role in the model. IA and HCL have positive impacts as well, although to a
lesser extent. On the other hand, CO, GBI, IC, Gen*MKT, and MKT are negatively
associated with the dependent variable. The feature MKT shows a larger negative
importance than before, while Gen*MKT's negative influence has decreased. This
could imply that the decrease in the target variable with respect to these features is more
pronounced in the current model iteration. The interaction terms continue to indicate
that the combined effect of certain features (Gen*CTI and Gen*HCL) is substantial,
reflecting the importance of considering how features interact with each other rather
than just their individual effects. These interactions can sometimes elucidate hidden
relationships in the predictive modeling process.
The Residual Plot for the Elastic Net regression, presented in the top left quadrant,
shows the distribution of residuals across different predicted values. The light blue
points are evenly scattered around the zero line, with the blue trend line remaining close
to zero across the range of predictions, indicating a uniform variance of residuals. This
uniformity suggests that the Elastic Net model is consistent in its predictive errors
across the range of values, without showing signs of heteroscedasticity (a condition
where the variance of the residuals is not constant across all levels of the explanatory
variables). The slight spread of residuals at higher predicted values may point to
potential model improvements, such as accounting for non-linear relationships that the
current model might not fully capture.
The Prediction Error Plot illustrates the relationship between the Elastic Net model's
predicted values and the actual values. The light blue points, which denote individual
predictions, are mostly clustered around the blue violet dashed line that represents
perfect prediction. This close clustering indicates a strong predictive accuracy of the
model, especially around the lower to mid-range of values, while deviations at the
higher end suggest less accuracy for larger values of the dependent variable. The
Learning Curve in the bottom left shows a good balance between training and validation
scores, with both lines plateauing as more data is used for training. This indicates that
the model is generalizing well and is neither underfitting nor overfitting. Finally, the
Elastic Net Path in the bottom right reveals the impact of regularization on model
complexity: as the regularization parameter increases, the model simplifies by reducing
the magnitude of coefficients, which can be seen in the transition of lines towards the
RCBA-2023 Volume 6, Issue 1
P a g e | 149
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
baseline. This illustrates the trade-off between model complexity and regularization
strength, where the Elastic Net model balances the inclusion of features and the
prevention of overfitting through its penalty terms
Figure 4. Evaluation plots for Elastic Net Regression
All three models present an adjusted R-squared of 0.76, signifying that, despite
differences in their constraining mechanisms, they equally account for a substantial
proportion of the variance in the dependent variable after adjustment for the number of
predictors. In terms of predictive accuracy, Ridge Regression shows a marginally better
performance with the lowest MSE and RMSE values, implying that it has the smallest
average prediction errors. The Cross-Validation Scores are consistent across all models
at 0.77, indicating similar robustness in predictive performance when generalized to
unseen data.
Gen*CTI consistently appears as the most positively influential feature across all
models, although its importance score varies slightly. The Lasso Regression appears to
assign more pronounced importance to the features, both positive and negative, which
aligns with its characteristic of possibly reducing the coefficients of less important
features to zero. In contrast, Elastic Net, which blends L1 and L2 regularization,
distributes feature importance more evenly, neither exaggerating nor minimizing them
to the extent seen in Lasso.The negative importance scores of features like CO, GBI,
RCBA-2023 Volume 6, Issue 1
P a g e | 150
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
IC, MKT, and Gen*MKT vary across the models, with the most substantial negative
weights seen in Ridge Regression, particularly for Gen*MKT. This could indicate a
stronger penalization for these features, which Lasso and Elastic Net may not emphasize
as much due to their propensity to eliminate certain predictors altogether. The use of
interaction terms such as Gen*CTI and Gen*HCL is validated in all models, although
the magnitude of their importance fluctuates. This underscores the relevance of
considering not only the individual contribution of each feature but also the synergistic
effects that occur when combining them.
Conclusion
The findings of this study indicate that when Generative AI adoption and Current
Technological Infrastructure interact, their combined effect on revenue growth in SMEs
seems to be considerably stronger than their individual impacts. This robust interaction
indicates that a well-established technological environment within a business
significantly enhances the utility and effectiveness of Generative AI. It is likely that
existing technological frameworks provide a fertile ground for AI applications to be
deployed and integrated into the business processes more seamlessly. For SMEs, this
can result in more effective data processing, automation of complex tasks, and an
overall more adaptive use of AI capabilities to meet business objectives. Consequently,
SMEs that already have modern technology infrastructure are in a better position to
extract more value from Generative AI, leading to potentially larger increments in
revenue.
For SMEs, the prominent role of technology infrastructure in maximizing the benefits
of Generative AI cannot be overstated. This means that prior investments in technology
can significantly dictate the scale and pace at which AI can be adopted and used to drive
revenue growth. If an SME’s existing infrastructure is dated, integrating AI may require
additional resources and could lead to less-than-optimal results. For these businesses, it
may be necessary to upgrade their technological framework to create a conducive
environment for AI. This does not necessarily mean that all legacy systems need to be
replaced; however, it does emphasize the importance of ensuring that the technology in
place can support and amplify the advantages that AI is expected to bring.
The substantial impact of the interaction between Generative AI and technological
infrastructure on business outcomes suggests that decision-making around AI should
not be isolated from the consideration of the business’s technological status quo. SMEs
might need to conduct an in-depth analysis of their existing systems to identify
compatibility and scalability in the context of AI integration. For those with modern
infrastructures, adopting AI could represent a strategic enhancement, driving innovation
and creating opportunities to outperform competitors. This dual focus on technology
and AI can become a critical aspect of strategic planning for SMEs aiming to achieve
significant revenue growth in the dynamic market landscape.
The moderate importance scores for the interaction between Generative AI adoption
and Human Capital Level suggest that while there is a clear relationship, it's not as
strong as the interaction with technological infrastructure. This finding can be
RCBA-2023 Volume 6, Issue 1
P a g e | 151
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
interpreted to mean that the education level and expertise of employees do enhance the
implementation and utilization of Generative AI, but these human factors may not be
as critical to revenue growth as having the right technological assets in place. SMEs
should take note of this as it indicates that simply having a well-educated workforce is
not enough to fully capitalize on the advantages of Generative AI. There might be other
elements at play that determine the success of AI integration, such as the nature of the
business, the specific industry sector, or the type of AI applications being adopted.
For SMEs, this could mean that while investing in employee education is important,
expecting it to be the sole driver of significant growth through Generative AI could be
unrealistic. Instead, the skills and knowledge of the workforce should be seen as
complementary to the technological infrastructure. Employees with higher academic
achievements may be better equipped to interact with advanced AI systems and could
contribute to more innovative approaches in AI application. However, the potential of
these contributions to translate into revenue growth is likely tempered by how well the
AI systems are integrated with the existing technology and how they're applied to the
business's operations and strategy.
This distinction is vital for SMEs in allocating resources and shaping strategies for AI
adoption. It suggests that companies may benefit more from a balanced investment in
both technological infrastructure and human capital development. By recognizing that
the impact of human capital on leveraging AI is significant but not paramount, SMEs
can aim to create a more harmonious interaction between their workforce’s capabilities
and their technological advancements. This balance is likely to be a key factor in
achieving optimal outcomes from AI investments, especially in terms of revenue
growth and business performance.
The negative importance scores associated with the interaction between Generative AI
adoption and Market Competition in all datasets are intriguing, as they suggest a
consistent adverse influence on the revenue growth of SMEs. It appears that the
presence of Generative AI in a highly competitive market does not necessarily translate
to better financial performance, as one might intuitively expect. This could imply that
the pressures and dynamics of a competitive market may diminish the effectiveness of
Generative AI technologies, or perhaps that the saturation of AI within an industry
makes it harder for any single SME to gain a distinct advantage. The introduction of AI
might lead to a race where all competitors rapidly adopt similar technologies, thus
nullifying the competitive edge that such a technology could otherwise provide.
This finding serves as a cautionary for SMEs that operate in very competitive markets,
suggesting that the deployment of Generative AI alone is not a panacea for revenue
growth. It could be that in such markets, the rapid diffusion of AI technologies levels
the playing field, making it more difficult for any individual firm to leverage AI for a
significant competitive advantage. Alternatively, it could reflect a misalignment
between the AI applications and the actual needs or strategic goals of SMEs within these
markets. It seems that in competitive sectors, the ability to harness AI effectively
requires a nuanced approach, considering factors such as the timing of adoption, the
RCBA-2023 Volume 6, Issue 1
P a g e | 152
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
uniqueness of the application, and the ability to integrate AI deeply into the value
proposition of the business.
SMEs may need to reconsider how they view the role of Generative AI within their
competitive strategy. Rather than expecting AI to be a straightforward driver of growth,
it may be more prudent to think of AI as a tool that needs to be carefully integrated into
a broader strategy that includes other competitive factors. For SMEs, this could mean a
greater focus on innovation in the application of generative AI, or perhaps a tailored
approach that focuses on niche market segments where generative AI can be used to
meet specific customer needs in ways that competitors have not yet exploited. It
indicates the importance of a strategic, rather than purely technological, approach to
generative AI adoption — one that is aware of the market dynamics and is designed to
create a competitive advantage rather than just keeping pace with competitors.
References
[1] P. Ainsley, B. Bailey, and P. Ainley, The business of learning. Continuum
International Publishing Group, 1997.
[2] T. Zwingmann, Ai-powered business intelligence. O’Reilly Media, 2022.
[3] S. Sinha and K. Al Huraimel, Reimagining businesses with AI. Nashville, TN: John
Wiley & Sons, 2020.
[4] A. Creswell, T. White, V. Dumoulin, K. Arulkumaran, B. Sengupta, and A. A.
Bharath, “Generative Adversarial Networks: An Overview,” IEEE Signal Process.
Mag., vol. 35, no. 1, pp. 53–65, Jan. 2018.
[5] Z. Lyu, X. U. Xudong, C. Yang, D. Lin, and B. Dai, “Accelerating Diffusion
Models via Early Stop of the Diffusion Process,” arXiv [cs.CV], 25-May-2022.
[6] S. Odaibo, “Tutorial: Deriving the Standard Variational Autoencoder (VAE) Loss
Function,” arXiv [cs.LG], 21-Jul-2019.
[7] D. M. Korngiebel and S. D. Mooney, “Considering the possibilities and pitfalls of
Generative Pre-trained Transformer 3 (GPT-3) in healthcare delivery,” NPJ Digit
Med, vol. 4, no. 1, p. 93, Jun. 2021.
[8] E. Brynjolfsson, D. Li, and L. R. Raymond, “Generative AI at Work,” National
Bureau of Economic Research, Apr-2023.
[9] I. Solaiman, “The Gradient of Generative AI Release: Methods and
Considerations,” in Proceedings of the 2023 ACM Conference on Fairness,
Accountability, and Transparency, Chicago, IL, USA, 2023, pp. 111–122.
[10] Z. Epstein et al., “Art and the science of generative AI,” Science, vol. 380, no.
6650, pp. 1110–1111, Jun. 2023.
[11] M. W. Eysenck and C. Eysenck, AI vs humans. London, England: Routledge, 2021.
[12] H. Rijhwani, Technology to business. Independently Published, 2020.
[13] A. K. Ehsanes Saleh, M. Arashi, and B. M. Golam Kibria, Theory of ridge
regression estimation with applications. Nashville, TN: John Wiley & Sons, 2019.
[14] J. A. Chan-Lau, Lasso regressions and forecasting models in applied stress testing.
Washington, D.C., DC: International Monetary Fund, 2017.
[15] H. Zou and T. Hastie, “Regularization and Variable Selection Via the Elastic Net,”
J. R. Stat. Soc. Series B Stat. Methodol., vol. 67, no. 2, pp. 301–320, Mar. 2005.
RCBA-2023 Volume 6, Issue 1
P a g e | 153
Impact of Generative AI on Small and Medium Enterprises' Revenue Growth: The Moderating Role of
Human, Technological, and Market Factors
[16] J. Liu, G. Liang, K. D. Siegmund, and J. P. Lewinger, “Data integration by multi-
tuning parameter elastic net regression,” BMC Bioinformatics, vol. 19, no. 1, p.
369, Oct. 2018.
[17] Z. Zhang, Z. Lai, Y. Xu, L. Shao, J. Wu, and G.-S. Xie, “Discriminative Elastic-
Net Regularized Linear Regression,” IEEE Trans. Image Process., vol. 26, no. 3,
pp. 1466–1481, Mar. 2017.
