Amsterdam Business School
MSc. Business Administration
Track: Digital Business
Master Thesis
Student:
Samantha Pikero 11378638 Samantha_Pikero@yahoo.com1
stSupervisor:
Erik Dirksen MSc.2
ndSupervisor:
Merve Güvendik MSc. 17 August 2018Company performance after the
adoption of Artificial Intelligence
1
Statement of Originality
This document is written by Student Samantha Pikero who declares to take full responsibility
for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources
other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion
of the work, not for the contents.
2
Acknowledgement
This thesis could not have been completed without the support and encouragement of those
around me and I would like to take the time to thank them. First and foremost, I would like to
thank God for giving me the knowledge and strength throughout my academic career to succeed
in my goals.
To my thesis supervisor, Dhr. Erik Dirksen, thank you for you continuous support and being a
source of guidance and knowledge throughout this process. You made me feel at ease with what
seemed like a daunting task at first and I am forever grateful to have had you as a supervisor.
To my parents, Nacçia and Oswin Pikero, thank you for your love and support throughout this
period. Even though you might be 7,839 km away from me, you have always found ways to
encourage me and guide me to excel in my academic goals.
To all my friends and colleagues at the University of Amsterdam, thank you for making this an
unforgettable experience and making our master year one of the most fun and memorable parts
of my life. I can’t wait to see how we will shape the global markets in the future! Tune your ears to wisdom, and concentrate on understanding. Cry out
for insight, and ask for understanding. Search for them as you would for silver; seek them like hidden treasures. Then you will understand what it means to fear the Lord, and you will gain knowledge of God. Proverbs 2: 2-5
3
Abstract
The purpose of this study is to provide a reference point of what effects companies that have
adopted AI have experienced within the scope of their profit margin, employee productivity, the
firm profitability and the firm size. The paper contributes to the current literature by presenting a
case study of companies that are currently operational and have adopted AI in some capacity and
detailing what effects they have experienced, as well as providing an outline of factors that might
influence the employee’s acceptance of AI within the company. The study The findings for this
study proved to be insignificant in nature and further analysis of more years is necessary to be
able to truly measure if the implementation has any effect on the company. It is the first study of
its kind that aims to analyze the changes experienced by a group of companies who have adopted
4
Table of Contents
Acknowledgement ... 2 Abstract ... 3 1. Introduction ... 5 2. Literature review ... 7 2.1 Artificial Intelligence ... 7 2.1.1 AI major goals ... 8 2.2 Business application of AI ... 12 2.3 Company performance ... 15 2.4 Employee acceptance ... 17 2.5 Gap ... 203. Research design & Methodology ... 22
3.1 Research design ... 22
3.2 Methodology ... 24
3.2.1 Data collection and analysis ... 25
3.2.2 Pretest ... 26 4 Results ... 29 4.1 Profit margin ... 29 4.2 Employee productivity ... 30 4.3 Firm profitability ... 31 4.4 Firm size... 32
5 Conclusion & Discussion ... 33
5.1 Results ... 33
5.2 Managerial implications ... 35
5.3 Theoretical contribution ... 37
5.4 Limitations and reference for future work ... 37
References ... 39
Appendix ... 42
Profit Margin data ... 42
Employee Productivity data ... 44
Firm Profitability data ... 46
5
1. Introduction
From self-driving cars to voice controlled home pods, Artificial Intelligence (AI) has made
its way away from science fiction right into our homes within the last decade. What was once
thought of as a goal far beyond reach has now become a reality and it’s just getting started. Many consumers don’t even realize that the items which they use daily like their GPS systems or even searching on Google, all employ some form of AI. AI can be defined as machines or computers
exhibiting intelligent behavior (Wooldridge & Jennings, 1995). Keeping this in mind however, the
definition of what qualifies as truly intelligent is highly subjective to the perception of the user at
a certain point in time according to Marijn Markus, Data Scientist at Capgemini (Capgemini,
2018). Back in the ‘90s, the scientific community would have considered a machine that could
play and beat a human at chess to be intelligent however the same could not be said for today. We
currently have more processing power available for machines to use which ultimately leads to
better computational results and goals that supersede the goal of only beating a human in chess.
Programs at Google’s Deepmind have taught themselves how to move forward through obstacles
without any human direction as to what walking is or should look like (Heess et al., 2017). Another
impressive feat set by Deepmind researchers with the help of Machine Learning is that they were
able to cut the energy bills for one of Google’s data centers by 40% in 2014 (Evans & Gao, 2016). They were able to do this by training neural networks based off different operating scenarios that
arise and were able to pinpoint cues where the energy used could be reduced or allocated to other
parts of the facility, thus streamlining the total use of electricity. These examples of only one AI
system already show how vastly different and advanced the AI goals have changed over the years
and that the level of intelligence displayed by systems has seen a definite advancement over the
6 Management journals such as MIT Sloan Review and Harvard Business Review, cite AI as the
next big thing for businesses (Brynjolfsson & Mcafee, 2017). This is mainly due to AI allowing
businesses to create new products and services for customers, to streamline their business practices
and to remain ahead of other competitors that would otherwise jump ahead by adopting the systems
themselves. In their article Brynjolfsson & Mcafee (2017) state that in the fields of perception and
cognition are the fields that have made the most advancement up until now. In the field of
perception, applications such as speech recognition used by most personal assistant applications
such as Apple’s Siri or Amazon’s Alexa, have become so advanced that users can now speak in crowded rooms or even speak with different accents and still be understood, which is a true feat
accomplished by these systems. Speech recognition has also proven to be on average three times
faster than when typing out text messages in a recent study by Ruan et al., dropping from an 8.5%
error rate to 4.9% which took place within a year since the summer of 2016 (Ruan, Wobbrock,
Liou, Ng, & Landay, 2017).Businesses are adopting AI technologies and creating new products
for the market, yet little has actually been researched as to the impact of employing AI for the
businesses.
With this thesis I propose to research the impact on the businesses that AI implementation has
caused. I therefore propose the following research question:
What impact does the adoption of A.I. have on company performance and employee acceptance?
In the following chapters a literature review will be shown, followed by the research gap,
7
2. Literature review
This chapter provides a comprehensive overview of the existing literature on A.I. and its
business adaptations. First off, the concept of A.I. is introduced and broken down into the main
areas of studies where research has been focused in. Secondly, the business application of A.I.
technologies will be explored. Finally, the effect on the introduction of automation within a
business will be explored namely through the willingness of employees to adopt the new
technology and the costs the business endures of implementing new technological advances.
2.1 Artificial Intelligence
Wooldridge and Jennings (1995: 116) define Artificial Intelligence as “the subfield of
computer science which aims to construct agents that exhibit aspects of intelligent behavior”. In
this definition agents can be regarded as machines e.g. computers. The term was first mentioned
by John McCarthy in 1956 at the Dartmouth Summer Research Project on Artificial Intelligence
where a group of professors, scientists and mathematicians came together during an eight-week
period. The sole intention of this summer research project was to brainstorm on how to take on the
challenge of creating machines that can think and compute general calculations independently
(Moor, 2006). It was at this workshop that the field of AI research was born. Although great
projects came in the years after this workshop, the progress anticipated was slower than expected
and by 1974 the funding and research projects in the field had slowed down to almost a complete stop in a period which is now referred to as “the first AI winter” (Bohannon, 2015). The first AI winter ended in 1980 but soon after there was a second winter from 1987 to 1993. The cause of
these winters can be attributed to the fact that during these times, the costs of building tehse
8 was required to make the projects feasible. However, Gordon Moore theorized that the number of
transistors of a chip will double every two years, leading to the availability of more processing
power for computers which is now referred to as Moore’s law. This phenomenon, along with the
doubling of system speeds and the rate at which the public adopts technologies, has aided in the exponential growth of computing power over the years to the point that in the late 1990’s AI became a feasible field again for research and business (Anthes, 2017; Denning & Lewis, 2017).
2.1.1 AI major goals
Within the field of AI there are several different areas of focus. These areas represent what
the researcher perceives to be a goal that the machine has to be able to perform in order to be
deemed intelligent (Stuart & Norvig, 2009) (Luger, 2008).
The 7 major AI goals are:
1. Artificial general intelligence/ Automated reasoning
2. Computer vision
3. Knowledge representation and reasoning
4. Machine learning
5. Natural language processing
6. Automated Planning and scheduling
7. Robotics
The focus in artificial general intelligence is that a machine can successfully complete an
9 completes tasks based on reasoning done through complex theoretical models, whereas other forms
of AI, also referred to as weak-, narrow- or applied-AI, simply solve a limited number of problems
it was designed to do, not indicating any human level of reasoning beyond their programming
(Goertzel, 2014).
Computer vision refers to the machines ability to perceive and correctly identify objects. This
could be in the form of still images or moving videography. This form of AI can be applied to help
sort products that meet company criteria at a faster pace than it would take a human counterpart to
achieve. This form of AI has seen a significant improvement in recent years and has gone from an
error rate higher than 30% back in 2010 to around 4% in 2016 (Brynjolfsson & Mcafee, 2017).
Even with this great improvement, the systems still have a long way to go to surpass human level
of image recognition. In an experiment performed by the website called Topbots where
Metamaven’s Chief Technology Officer and Head of Product Mariya Yao, compared images of
chihuahuas and blueberry muffins, something that is very easy for a human to accomplish, the bots
misidentified some dogs as pastry items or even at times thought a muffin was a stuffed animal
(Yao, 2017). These systems are eternally learning as they get access to more image data that are
recorded into their databases and the error rate will only continue to lower as time goes by.
Knowledge representation and reasoning (KR&R) focuses on the ability to store aggregate data
and interpreting/transforming the data into knowledge that can be utilized to solve complex
problems. In order to accomplish this, discoveries from psychology and logic are used in building
the systems by attempting to automate how humans solve problems and create reason of what they
are presented. An example of KR&R is that of systems that analyze medical data and suggest
possible diseases that can be present using deductive reasoning, also referred to as “Observer
10 the what the health issue at hand is (Nieves, Lindgren, & Cortes, 2014). These latter systems are
also referred to as “Validating agents”.
Machine learning, the focus that currently is being reported on the most, is regarding creating
systems that can learn about subjects without having been programmed beforehand on how to do
so through the use of statistical methods. The NELL system for example, uses text input of over
500 million sentences to analyze and identify relationships between various concepts. Although
often confused with each other because they use similar techniques and overlap significantly,
machine learning is not the same as data mining. Machine learning focuses on systems that can
generate knowledge independently while data mining is concerned with discovering information
that was previously unknown. What researchers ultimately are interested in developing are systems
that are able to understand complex information on various subjects from independent learning
done by the systems themselves.
Natural language processing (NLP) focusses on the computer’s ability to process and interpret
large quantities of user generated natural language data. This might include recognizing speech,
understanding natural language and creating natural language output. One of the most difficult
tasks in recognizing and understanding human generated natural language for machines is that they
are unable to see the person who is speaking and without that visual information, the text becomes
pure abstract which is harder to interpret seeing that humans themselves rely on visual processing
30% of the time when communicating (Wiriyathammabhum, Summers-Stay, Fermuller, &
Aloimonos, 2016). However, Wiriyathammabhum et al. state that researchers are attempting to
circumvent this issue by combining computer vision and natural language programming in trying
to create systems that recognize small changes in facial expression and body language. The general
11 introducing the technique in their personal assistant application called “Siri” Amazon’s personal assistant called “Alexa”. Another form of NLP that has become popular in recent years is the use of a chatbot. A chatbot is a program on a website that interacts with customers as a customer
representative agent would. These programs are usually trained to be able to answer most
frequently asked questions with a predetermined response and redirecting the chat to their human
counterparts if the question being asked by the customer cannot be answered by themselves. This
frees up the time of the human customer representative to be able to help the customers with more
intricate questions in further detail than they previously would.
Automated Planning and Scheduling is the goal concerned with the computer’s ability to
interpret the optimal time and path to execute a task and plans/schedules this accordingly. This
goal of AI remains one with the least true progress achieved over the years due to the greatness of
planning in general. When met with an unknown problem, computers still need human input to
create reasoning of the problem and create a model for the system to analyze the situation and
create a proposed plan (Gesit, 2017). Researchers are continuously working on finding ways that
systems can create their own planning models so that in the future the plans they produce can be
used for military planning or strategic planning for business looking to expand beyond their current
borders.
Finally, as the name suggests, robotics is the goal that focuses on the system ability to
manipulate and move a machine. Although the field of creating robots and machines has long
existed as a form of engineering, researchers have been studying new ways to incorporate AI in
the way the machine operates. With AI a robot that malfunctions, or has a part that is hindered,
can assess the situation and derive an alternative motion to either relieve the hinderance or
12 Natural Language Processing has also been used in the field of robotics to create robots that can
carry a conversation with humans.
2.2 Business application of AI
When mentioning AI to the general public nowadays, the first thought that comes to most
consumers’ minds is the self-driving cars of Tesla and Google. These companies have adopted a series of connected sensors that actively calculate the speed, distance and elements around the
vehicle and controls the driving experience based on what they predict will happen next. But other
than self-driving cars, AI can also be seen in search and optimization algorithms, for example the
sentence predicter feature when typing out a search on Google, or robotic algorithms, visible in
robotic artificial limbs for example. In their research, Davenport and Ronanki (2018) found that
the majority of companies in their sample made use of Robotic Process Automation (RPA),
Cognitive insight, and Cognitive engagement. RPA automates processes such as sifting through
email communications for data and sorting them accordingly through use of coding on servers.
Cognitive insight refers to systems that detect patterns in the data and provide predictions on the
future based on that. Cognitive engagement are programs that use natural language processing
capabilities to engage with customers and employees such as chatbots. In order to successfully use
AI companies must possess valuable data that no other company has, continuously analyze their
business to find data adjacencies and format AI towards their targeted customer experience (Hardy,
2017).
Companies of all different industries are already using AI to leverage existing expertise
and to create new business opportunities with new products and services. In the healthcare industry
13 using deep learning techniques (De Fauw, et al., 2018). This AI was able to correctly assess
patients that are predisposed to retinal disease through analyzing the patients’ retinal scans with
an error rate of 5.5%, while practicing medical professionals were able to do so with an error rate
between 6.7 and 24.1%. The error rate of the medical professionals is also able to further go down
if they have additional medical history information on the patient to between 5.5 and 13.1%. This
tech can benefit hospitals and medical practices as a tool that they can use in addition to their own expertise to be sure that they don’t miss any possible illnesses.
A company within the healthcare industry that has been utilizing AI for the past 20 years
has been Siemens Healthineers, which operates in over 70 countries worldwide. With their
software system named Syngo, radiologists have a backup in diagnosing patients when assessing
CT and MRI scans. In fact, in the EU more than 28% of healthcare leaders expect AI to be used
in medical decision support within the next five years, 31% in Asia and over 50% in Middle East
and South Africa (The Economist, 2017). Siemens Healthineers have also been using deep
(machine) learning since 2012 in developing applications that enable the diagnosis process and
treatment support to be automated. The multinational Dutch technology company Philips has
launched its HealthSuite Insights recently also. The Healthsuite has been created to give
researchers and healthcare professionals a place where they can collaborate on projects that use
artificial intelligence solutions. This was developed as they believe it is important for there to be a
support platform for the advancement of adoption of analytics and AI in healthcare (Koninklijke
Philips NV, 2018). Philips’ own healthcare subdivision, aptly named Philips Healthcare, is active
in a wide variety of AI fields such as Natural Language Processing, Applied Machine Learning
14 Within the agricultural field, AI is also being utilized. CNH Industrial, headquartered in
the United Kingdom, introduced a concept called ‘Precision Farming’ in 1996 which in their own
words “sparked a great agricultural revolution” (CNH Industrial NV, 2017). What this tool allowed their farmers to do is analyze their fields with their tractor system to better assess which parts were
producing more crops and getting better nutrients, and which in comparison were performing
poorly, and this all, as close as the nearest square inch. It has been further developed over the years
to now include systems that can also operate the tractors themselves while following the guidance
of the system of field locations that need to be treated. These new systems CNH Industrial are
introducing are thus fully autonomous and can initiate turn sequences when nearing the end of a
field or operate individual parts of a tractor such as a plow or irrigation system, all independently.
The goal for the near future for these systems is to allow them to use real time big data information
from weather satellites to make better farming decisions during any time of the day without any
human input.
Within the insurance industry there is also a company that has already implemented AI
technologies within their operations in a variety of ways. Due to privacy concerns the specific
name of the company will remain anonymous and they will be referred to as company 14 as they
are one of the companies that form part of the research further in this thesis. When it comes to
Natural Language Processing systems, company 14 has implemented chatbots to interpret client
questions and give them the appropriate responses to them. They also have an algorithm that
analyzes and detects patterns in every damage claim they receive to signal possible cases of fraud.
This enables the company to catch these fraud cases at a higher rate than they were able to do so
before. They also utilize a model that scans through documents that customers hand in to the
15 if they need to signal the application as having issues. Machine Learning is also being used to route
the emails they receive, and it is so successful that it is able to sort and route emails 3 times better
than using traditional methods. Predictive models have also been implemented that use historical
data to calculate the chances of an application that they receive being accepted. Process mining
techniques are also in use where the processes of the organization are being monitored and
analyzed continuously in order for them to signal processes where there are deviations or
congestion in the workflow.
These are all examples of companies that have already embraced AI systems and
applications in their businesses that enable them to not only offer a better service to their clients,
but also to have better control of their own operations internally.
2.3 Company performance
To validate business decisions, executives rely on a plethora of measures that indicate
company performance referred to as performance measurement systems. Neely et al. (1995) define
a performance measurement system as “the set of metrics used to quantify both the efficiency and effectiveness of actions”. That is to say, in order for businesses to know how the decisions they have made are impacting the company’s performance, both in efficiency and effectiveness, they must have a set of metrics in place that can measure these. Whereas before businesses measured
performance mainly in accounting terms of profit and loss, recently they have an increased interest
in measuring it in other non-accounting ways (Choong, 2014). This is due to the everchanging
environment that businesses perform in and the rate at which globalization is happening within the
16 competitive with them, but also the international competition that is suddenly closer to them.
Kueng, Meier, & Wettstein (2001) categorized the performance measures used in a performance
measurement system into 5 categories: people, procedures, data, software and hardware. This is
largely due to the fact that companies should be keeping an eye on the changes in their own organization’s environment and having a system in place that covers all these different categories ensures that they can pinpoint where improvements should be made. Benitez, Chen, Teo, &
Ajamieh (2018) researched the impact of e-business technology within firms and defined the
following constructs for performance measurement: Profit margin, employee productivity,
operational excellence, firm profitability, firm size, industry and advertising spending. These
variables were selected for this study based on the work of Tatikonda et al. (2013) since they relate
to productivity management, gross margin and service excellence and are a good example of what
competences a firm needs to have in order to be successful and thrive in the future. If companies
are to adopt new technologies within their firms and want to be sure that these new additions are
beneficial for the company, they need to thus pay attention to these key company performance
measures to ensure their success. For the purpose of this study the constructs profit margin,
employee productivity, firm profitability and firm size were selected to be studied as these are
readily available information provided by all companies in their annual reports and fit within the
studies of Tatikonda and Benitez et al.
As with other analytical researches, a null hypothesis is always set where the assumption
is that there is no difference in the tested groups. Therefore we propose:
H0: The implementation of AI has no significant impact on company performance
Considering the aforementioned success companies faced after implementing AI within
17 performance. Since we chose to test the variables profit margin, employee productivity, firm
profitability and firm size to construct what the current business performance of the company is
we then present the following 4 alternative hypotheses:
H1: The implementation of AI has a significant impact on the profit margin
H2: The implementation of AI has a significant impact on the employee productivity
H3: The implementation of AI has a significant impact on the firm profitability
H4: The implementation of AI has a significant impact on the firm size
2.4 Employee acceptance
As with the introduction of any unfamiliar component to the workplace, the adoption of
new technologies is also met with varying levels of pushback from employees. The adoption of
new technologies by companies, which in many cases reflect a slight automation to certain tasks
performed by employees such as automated phone clients or quality control robots, are often met
with hostility by employees who see them as potential replacement for their jobs.
Acceptance is defined as “the action of consenting to receive or undertake something offered”, “the process or fact of being received as adequate, valid, or suitable.” or “agreement with or belief in an idea or explanation” by the Oxford Dictionary (2018). Adell et al. (2014) classify five different ways to define acceptance. Firstly, acceptance is defined using the word accept itself,
in other words, acceptance is the extent to which a concept is accepted. Secondly, the fulfillment
of the needs and demands of users is considered when defining acceptance as the extent to which
a user feels that the tool used fulfills their needs. Thirdly, emotions are considered when defining
18 the use of the tool based on their personal feeling. Fourthly, the intent of the user is used in defining
acceptance as the extent which a user intends the continuous use of a tool having accepted it as
being a positive addition to completing their goal. Lastly, the actual use of the tool is considered,
and acceptance is defined as extent which a user utilizes a tool for the purpose it is designed. In
summary these definitions thus move from a user assessing the helpfulness of a tool, to the extent
which a tool is actually used.
The rate at which employees accept new technologies allows productivity goals to be
reached at a faster pace (Anton, Camarero, & San Jose, 2014). The perception employees have on
the new technology also positively impacts the satisfaction levels experienced while working with
them (Au, Ngai, & Cheng, 2008). In their article titled User Acceptance of Computer Technology:
A comparison of two theoretical models, Davis et al. introduced a framework to assess the user
acceptance of technology called TAM which is short for the Technology Acceptance Model
(Davis, Bagozzi, & Warshaw, 1989)1. This framework put emphasis on external variables being
determinant to a user’s perceived ease of use and perceived usefulness of the technology. These factors in turn influence the user’s attitude toward using the technology, how their behaviour towards the intention of use is and finally how they actually use the system. This framework was
further expanded on by Ghazizadeh, Lee & Ng Boyle (2012) who added elements of Cognitive
Engineering (CE), which takes into consideration the engineering of work between people and
systems, to the framework2 and theorized that user compatibility and trust are not only also
influenced by external factors but in turn also influence the rest of the previous frameworks factors.
This framework provides the theoretical backing to test the perception of A.I. within the companies
1 See Appendix 1 2 See Appendix 2
19 to measure the user acceptance within the company. Brandon-Jones and Kauppi (2018) also
expanded on the TAM framework and stated that employees form their opinions on technology
from interactions and opinions of other employees also.
Another framework that has been referenced frequently in regards to technology adoption
is that of Moore and Benbasat (1991) called PCI or Perceived Characteristics of Innovating. They
identify 8 PCI factors that influence wether or not a user will accept using an innovation, namely:
the ease of use, compatibility, visibility, voluntariness, relative advantage, result demonstrability,
observability, and image.
As frequently as the TAM and PCI frameworks have been referred in literature as the
means of measuring technology acceptance, these frameworks have also been criticized.
Schwarz and Chin (2007) point out the fact that these frameworks’ means of measuring the
acceptance of technology is too dependent on usage. The fact that they measure acceptance by
how frequently a person uses the technology instead of considering the whole lifecycle a person
experiences while using technology and the different stages of acceptance that will occur in that
lifecycle, does not create an accurate depiction of acceptance as it only focuses on one factor
which is usage. Using etymology, they propose 5 dimensions that should be taken into consideration when assessing the IT acceptance of people, namely: the person’s ability to receive, grasp, assess, be given and submit. In this concept IT acceptance would then also
consider the person’s openness to accept the technology without doubting it (receive), the
person’s ability to wholly understand the concept of the technology at hand (grasp), the person’s
capacity to evaluate the value and desirability if the technology (assess), the person’s readiness to alter their routines to those required to use the technology (be given), and finally the person’s capacity to fully yielding to the purpose of the technology.
20 The aforementioned characteristics of employee acceptance are really important for
companies to consider according to Michel M. from an IT service corporation active globally
(2018). Employees who are unfamiliar with the new technology form strong opinions before
using the technology sometimes out of fear that it is taking away their jobs and that presents a
challenge for companies as they become unwilling to adopt the new systems. This pushback
from employees results in systems not being used for long periods of time, even though they are
ready, and the company to lose on that investment. Companies should take time to showcase to
these workers that AI usually does the tedious and repetitive work that takes up a lot of their
time, which then gives them the ability to spend their time on more enriching and engaging
work. In addition, being transparent about what the system actually does and building up trust
with those who will be exposed to it is also vital for success of the AI system within the
company (Fernando, 2018).
Because AI is still relatively new for the workforce outside high tech companies, the weight
then falls on management to consider the dimensions of acceptance that their employees face when
introduced with the technology.
2.5 Gap
Within the current scope of literature, the realm of A.I. is still mostly written on regarding
computer science advances and ideas and hypothesizing what the potential future would be like
when advanced A.I. starts being implemented in a manner readily accessible to the average
21 seems to be ignored. The gap identified in current literature therefore is that of the effect that A.I.
has had thus far on companies that have adopted it in the form of company performance and
acceptance.
22
3. Research design & Methodology
3.1 Research design
The purpose of this study is to evaluate what impact the introduction of A.I. within a
company has had on the company itself. To be able to do this a case study design has been used. A case study is defined as “an empirical inquiry that investigates a contemporary phenomenon within its real-life context; when the boundaries between phenomenon and context are not clearly
evident; and in which multiple sources of evidence are used.” by Yin (Yin, 1984) (Zainal, 2007). This design was thus chosen because it delivers an understanding of an intricate subject by means
of a contextual analysis of a limited number of examples or conditions and the relationships they
experience. Additionally, multiple sources of evidence are used to describe the company
performance within the selected time frame of the study, as is the case for other case studies. The
methodology selected for this research has been a qualitative study keeping in mind that the data
gathered would not have been analyzed in a quantitative manner but rather a descriptive manner.
This is due to the fact the variables that were analyzed will be descriptive of the state the companies
were in after the introduction of AI, and also this research could not say in any definitive capacity
that these results were a direct consequence of only the AI adoption within the company itself.
That is why even though numerical variables of the companies were analyzed, these variables were
used to be descriptive of what impact the introduction of AI had on company performance instead
of attempting to measure the exact numerical impact this introduction had on the companies. The
data source for the study will be 48 companies that have changed their practice with implementing
AI and have at least 1 year of subsequent data of the performance. The research aims to analyze
changes occurred after the implementation of AI and additionally we also wanted to see if there
23 technology had not been introduced yet. Knowing that companies have started implementing AI
within the last 3 years, and not wanting to limit the amount of companies that would fall within
the sample, the decision was thus made that the companies selected would need at least 1 years’
worth of data before they implemented AI, data of the year they implemented AI and 1 years’
worth of data after they implemented AI. This had ensured that companies that implemented AI
up until 2016 could still be analyzed within this research.
The database that was used to retrieve the company information necessary is Orbis from
Bureau van Dijk. From this database, a list of the top 600 companies with the highest generated
revenues from the Netherlands and Belgium who also have a minimum of 10 years of available
financial data has been made, and they were all contacted to inquire if they have ever implemented
any AI technology and what the year of implementation was if they answered positively to the first
question. Given that there are several companies that do not wish to disclose the technologies that
they use for a myriad of reasons from either staying ahead of the competition to not wanting to
showcase how primitive the technology they use is, it would be impossible to limit the scope of
the research to only one form of AI, for example natural language processing or robotics, and still
having a sizable sample size. When considering sample size, the goal of this research was to be
representative of 50 companies. This amount was chosen as there was a time limit within which
the research had to be completed and this amount was attainable within this timeframe. In
accordance with Krejcie and Morgan, in order for a research to be representative of a population
of 50, the sample size has to be 48 (Krejcie & Morgan, 1970). That is why a sample size of 48
companies has been selected for this research. For the employee acceptance section, a literature
review has been done of literature that focuses on what influences employee’s acceptance towards
24
3.2 Methodology
The main variable to be tested is company performance. Under company performance the
following were observed to construct the variable: Profit margin, employee productivity, firm
profitability, and firm size (see figure 1).
Figure 1
These variables were chosen in accordance with prior research conducted by Benitez et al.
researching a similar effect on a different technology type (Benitez, Chen, Teo, & Ajamieh, 2018).
In their research they aimed to measure the effect of e-business practices on companies in Spain.
Because this research similarly wanted to see the impact that an introduction of a technology has
had on companies, it was chosen as the reference point for selecting the variables. Table 1
25
Table 1
The data for the company performance variables will be collected through the database
Orbis from Bureau van Dijk over a period of 2 months. The same database will be used for all
variables to ensure that the measurements are completed in the same manner. The data gathered
will be for 1 year prior and 1 year post the implementation of the AI technology.
The strength of the research lies in that such a research has not been completed yet and most
literature currently implies AI implementation is beneficial for companies but have not backed up
these claims with research on companies that actually have already implemented the technology.
The limitations are in acquiring the information of which companies have adopted AI in the past
and in what specific year they did so.
3.2.1 Data collection and analysis
In total 588 companies were contacted and asked if they have employed AI before and
from what year this has happened. The response rate of the inquiry was at 19.39% with 114
responding, of which 55.26%, 63 companies, refused to disclose this information or did not employ
AI and 44.74%, 51 companies did employ AI and were willing to disclose the start year. From
these companies 3 however utilized AI in a period where the financial and company data was too
Variable name, measure definitions, and data sources
Variable name Measure definition Source
Profit Margin Profit Margin (%) = (Earnings before taxes / Operating Revenue) x 100
Orbis
Employee Productivity Operating Revenue per employee (1000 Euros) = Operating Revenue / Number of Employees
Orbis
Firm Profitability Return on assets (%) = (Earnings before taxes / Total assets) x 100
Orbis
26 far away or too recent to obtain thus they were removed and the final amount of participating
companies was the required 48. The descriptive statistics of the companies analyzed are in Table
2.
Table 2
Descriptive statistics of sample companies
Year of AI implementation 2014-2016 33 companies 68.75% Before 2014 15 companies 31.25% Firm size 0- 1000 employees 7 companies 14.58% >1000 employees 41 companies 85.42% Annual turnover
0 – 1 million Euros 7 companies 14.58%
1-10 million Euros 12 companies 25%
10 – 100 million Euros 29 companies 60.42%
3.2.2 Pretest
Due to the fact that we want to see if there was a significant change for each variable over the
course of 3 years in this analysis, we would have to compare the group means to measure this. A
Paired sample t-test would be recommended for this analysis. A paired sample t-test is a statistical
test that analyzes mean differences between two groups of observations. The resulting outputs
would then identify if the mean difference between the groups is significant or not. The paired
sample t-test has 2 requirements:
27 2. The dependent variables are either continuous or interval level
A normality check is thus performed to verify if the data indeed meets the first criteria. The
difference between year 1 and 2, 2 and 3 and 1 and 3 are all analyzed for skewness and kurtosis
and the results are visible in table 3.
Table 3
Normality check
Difference Skewness Kurtosis
Profit Margin year 1 and 2 4.883 28.634
Profit Margin year 2 and 3 -2.245 8.846
Profit Margin year 1 and 3 3.902 25.207
Employee productivity year 1 and 2 5.963 38.863
Employee productivity year 2 and 3 4.398 22.034
Employee productivity year 1 and 3 3.954 16.884
Firm profitability year 1 and 2 1.174 3.534
Firm profitability year 2 and 3 -2.949 10.637
Firm profitability year 1 and 3 -2.557 9.964
Firm Size year 1 and 2 -1.741 15.853
Firm Size year 2 and 3 2.107 10.891
Firm Size year 1 and 3 1.386 7.040
Keeping in mind the rule of thumb, that normal distributions have skewness and kurtosis
close to 0, we can conclude from these findings, which are all above 1 point away from 0, that
the differences between the pairs are not normally distributed and thus a Paired samples t-test
cannot be performed.
3.2.2.1 Wilcoxon Matched Pairs test
For cases where the difference between pairs is not normally distributed there is another test
that could be performed where normal distribution is not required called the Wilcoxon
Matched-Pairs test. Similarly to the Paired-Sample t-test, the Wilcoxon Matched-Matched-Pairs test is a statistical
test that measures the mean difference between two observations of a variable group. However,
28 distribution of the sample to be normally distributed. The Wilcoxon minimizes the effect of
outliers or skewed distribution on the results by ranking each measurement score by lowest
(receiving 1), the next lowest score then receives a rank of 2 and so on until all measurements are
ranked. A limitation of this test is that if there is no change between the two measurements of an observation, it’s result will thus be zero and it would be discarded from the calculation. This however is not the case for any of the observations in this research and thus it does not apply
here. The null hypothesis for this test is that there is no significant difference between the two
observed samples and the alternative hypothesis is that there is a significant difference between
the two observed samples. A difference is deemed significant if the observed two tailed
asymptotic significance is 0.05 or lower. The asymptotic significance indicates the probability of
the having the results and the null hypothesis, that there is no significant difference between the
two samples, remaining true. As a rule of thumb, if the probability is 5% or less (a significance
level of 0.05 or less), we then reject the null hypothesis and accept the alternative hypothesis that
there indeed is a difference between the two means.
The assumptions for this test are:
1. The data is of repeat measurement of the same population
2. Each pair of measurements is performed independently and is randomly chosen
3. The data is measured at ordinal, interval or ratio level and is not required to be normally
distributed
The data for this analysis is indeed repeated measurement of the same companies within the
population, has been performed independently and is measured on a ratio level. This means the
29
4 Results
In this chapter the results of the Wilcoxon Matched-Pairs test analysis will be reported. Due to
the fact that we are interested to see if there are any significant changes to each variable between
the year before implementation and the year of implementation, the year of implementation and
the year after implementation, and finally the year before implementation and the year after
implementation, we will be comparing the mean groups of the companies for each variable
against each other in this matter. In the following SPSS outputs of the different tests the Z score
and asymptotic significance will be indicated. The Z score is used in statistics to calculate the
significance level of a reported statistic. A Z score between smaller than -1.96 and larger than
1.96 is indicative of a reported statistic being significant of 0.05 in a two-tailed test. SPSS also
shows the significance level in the output which makes it faster to verify whether or not a
measurement is indeed significantly different or not.
4.1 Profit margin
A Wilcoxon Matched-pairs test had been run in SPSS where the profit margin means of
years 1 and 2, 2 and 3 and 1 and 3 were compared. The SPSS output for this Matched-pairs test
is in Table 4:
Table 4
From this output we can see that none of the pairs has a Z score less than -1.96 or greater
than 1.96 which would signify that they are indeed significantly different. Additionally, none of Wilcoxon Matched-Pairs test results – Profit Margin
Year 2 – year 1 Year 3 – year 2 Year 3 – year 1
Z -0.359 -1.133 -0.954
30 the pairs passed the 5% (0.05) confidence interval criteria for the asymptotic significance test
and are all above this. Therefore, we can conclude that there is not enough statistical evidence to
reject the null hypothesis and so we accept the null hypothesis that there is no significant
difference in the profit margins between the means of 1 year prior to AI adoption and the year of
AI adoption, the year of AI adoption and 1 year after AI adoption, and 1 year prior to AI
adoption and 1 year after AI adoption. We therefore also reject H1 which states that the
implementation of AI has a significant impact on the profit margin.
4.2 Employee productivity
A Wilcoxon Matched-pairs test had been run in SPSS where the employee productivity
means of years 1 and 2, 2 and 3 and 1 and 3 were compared. The SPSS output for this
Matched-pairs test is in Table 5:
Table 5
From this output we can see that none of the pairs has a Z score less than -1.96 or greater
than 1.96 which would signify that they are indeed significantly different. Additionally, none of
the pairs passed the 5% (0.05) confidence interval criteria for the asymptotic significance test
and are all above this. Therefore, we can conclude that there is not enough statistical evidence to
reject the null hypothesis and so we accept the null hypothesis that there is no significant
difference in the employee productivity between the means of 1 year prior to AI adoption and the Wilcoxon Matched-Pairs test results – Employee productivity
Year 2 – year 1 Year 3 – year 2 Year 3 – year 1
Z -0.852 -0.918 -0.476
31 year of AI adoption, the year of AI adoption and 1 year after AI adoption, and 1 year prior to AI
adoption and 1 year after AI adoption. We therefore also reject H2 which states that the
implementation of AI has a significant impact on the employee productivity.
4.3 Firm profitability
A Wilcoxon Matched-pairs test had been run in SPSS where the firm profitability means
of years 1 and 2, 2 and 3 and 1 and 3 were compared. The SPSS output for this Matched-pairs
test is in Table 6:
Table 6
From this output we can see that none of the pairs has a Z score less than -1.96 or greater
than 1.96 which would signify that they are indeed significantly different. Additionally, none of
the pairs passed the 5% (0.05) confidence interval criteria for the asymptotic significance test
and are all above this. Therefore, we can conclude that there is not enough statistical evidence to
reject the null hypothesis and so we accept the null hypothesis that there is no significant
difference in the firm profitability between the means of 1 year prior to AI adoption and the year
of AI adoption, the year of AI adoption and 1 year after AI adoption, and 1 year prior to AI
adoption and 1 year after AI adoption. We therefore also reject H3 which states that the
implementation of AI has a significant impact on the firm profitability.
Wilcoxon Matched-Pairs test results – Firm profitability
Year 2 – year 1 Year 3 – year 2 Year 3 – year 1
Z -0.118 -0.390 -0.123
32
4.4 Firm size
A Wilcoxon Matched-pairs test had been run in SPSS where the firm size means of years
1 and 2, 2 and 3 and 1 and 3 were compared. The SPSS output for this Matched-pairs test is as
follows:
Table 7
From this output we can see that none of the pairs has a Z score less than -1.96 or greater
than 1.96 which would signify that they are indeed significantly different. Additionally, none of
the pairs passed the 5% (0.05) confidence interval criteria for the asymptotic significance test
and are all above this. Therefore, we can conclude that there is not enough statistical evidence to
reject the null hypothesis and so we accept the null hypothesis that there is no significant
difference in the firm size between the means of 1 year prior to AI adoption and the year of AI
adoption, the year of AI adoption and 1 year after AI adoption, and 1 year prior to AI adoption
and 1 year after AI adoption. We therefore also reject H4 which states that the implementation of
AI has a significant impact on the firm size. Wilcoxon Matched-Pairs test results – Firm size
Year 2 – year 1 Year 3 – year 2 Year 3 – year 1
Z -0.472 -1.228 -1.226
33
5 Conclusion & Discussion
In this final chapter, the results of the analyses will be discussed in further detail. Thereafter, the
managerial implications of the findings will be given. Lastly, the limitations of the study and
references for future work will be given.
5.1 Results
From the results of the tests we can conclude that there was no significant change
between the year of AI implementation, the year before and the year after AI implementation
with regards to their profit margin, employee productivity, firm profitability and firm size.
Because of this, the null hypothesis was never rejected and was kept. This means that the
addition of AI within those practices did not alter the business performance in such a way that it
altered the workflow of the company for the subsequent year.
When considering the results for Profit Margin, we can see that difference between the
year of implementation and the year after implementation has had the biggest notable
change, even though this is not significant enough for us to make any conclusions. Due to the
fact that it also has the smallest Z score out of the 3 pairs we can also conclude that the
difference between the profit margin of the year of implementation and the year after
implementation has the biggest negative difference. This is to be expected as companies do incur
an initial loss in production levels and sales when implementing a new technology and that
would then have an effect on the overall profit margin.
Regarding the employee productivity, we can note again that the difference between the
34 it is not significant enough for us to make any conclusions and in this case, it is rather close to
the difference between the year before implementation and the year of implementation. What is
notable is that the difference between the year before implementation and the year after
implementation was the least significant change of the 3 pairs, signaling an averaging out of the
productivity levels to that of when there was no implementation yet within the companies.
When looking at the firm profitability, we can observe that the difference between the
year of implementation and the year after implementation had the largest change, although this is
not to such an extent that we can consider it significant and thus derive any conclusions from.
The significance level of the means of the year before implementation and the year after
implementation is also something to contemplate a little further since it indicates to a very high
degree that the means are similar. When considering that the mean difference of the profit
margin between the year before implementation and the year after implementation was of a
higher significance level than that of the other two pairs, indicating that this pair had a bigger
difference in the mean although not significant enough to be considered unequivocally different,
it is remarkable that the firm profitability is still similar to a greater extent between the pair. Due
to the fact that firm profitability takes the current assets of the company into consideration, while
profit margin is only concerned with the earnings in that specific year, we can also say that even
though the profit margin did experience a downward insignificant change, the firm profitability
of the observed firms were able to maintain a very similar level in the year after implementation
compared to the year before implementation due to the stability of the current assets within these
companies.
Lastly, when considering the firm size, we can see that both the pair of the year of
35 implementation and the year after implementation, had a similar significance level of 0.220
making these measurements the ones with the biggest shift in the means, although not to an
extent that they would be considered significant. In other words, the firms experienced a change
when comparing the year before implementation and the year after implementation, or when
comparing the year of implementation and the year after implementation, even though this
change was not significant and thus no definitive conclusions can be derived from it. The
negative Z score of the two pairs indicate that the mean difference between the two observations
was negative, which is the result when the second observation’s mean score is lower than the
first seeing that the way this score is calculated is the year after implementation minus the year
of implementation for one pair, and the year after implementation minus the year before
implementation for the other one. This means that for the observed companies there indeed was a
negative change in the mean sizes when comparing the year after implementation to both the
year before implementation and the year of implementation although these observations are not
of a significant size and no direct conclusion could be drawn definitively on them.
5.2 Managerial implications
Management of companies that are considering implementing AI within their company,
whether it be through automation of business processes that are already present or the
introduction of a new feature like image recognition, might find this paper as a useful example of
what companies have experienced with regards to their company performance that have
implemented the technology in the past. Although the results of the tests were not significant,
companies can still tentatively take some observations into consideration when contemplating
the implementation of AI within their own companies. Firstly, because there was no significant
36 the company is in a state where they want to see a significant improvement directly after
implementation or if they are willing to wait a few years to start noticing this effect. Secondly,
management must also consider if the goal of implementing the AI technology within the
company is to increase the employee productivity and, if this is the case, whether or not they are
willing to wait more than 1 year to notice this effect seeing that there was no significant
improvement within 1 year of implementing AI. Thirdly, management must also consider if the firm’s current assets would remain in a stable position over the first years after they implement AI. Although no definitive conclusions could be drawn from the test results, the stable firm
profitability mean changes measured comparing the year before implementation and the year of
implementation, and comparing the year before implementation and the year after
implementation, signal that the means remained similar to a great degree and this is likely due to
a stable current asset as the earnings before taxes were also considered for calculating the profit
margin and these measurements did not stay similar to that high of a degree. Having a stable
current asset may thus guarantee the firms profitability to stay the same even if the profit margin
does not stay the same to a similar high degree. Finally, with regards to the firm size, these
changes were not significant but there was a small downward change in firm size and
management must also ask themselves if the firm is open to decreasing in size in the years
following the implementation. This research is a good reference point to what changes
companies observed as they have adopted the technology and gives management the chance to
consider whether or not their own company would be willing to experience these changes for
37
5.3 Theoretical contribution
As is the world of technology, the current literature is always trailing behind the developments that
are actively happening in the world. This study adds to the current literature by providing a
reference point of what companies who have adopted AI have experienced in terms of finances
and growth in employee numbers. This thesis will contribute to existing literature by providing an
analysis of how the current adaptation of A.I. technologies are affecting businesses and provide
insights into which methods are proving to be more lucrative than others. This thesis also
summarizes on possible challenges companies might face when implementing these technologies
which will provide companies with an outline of what to expect when adopting A.I. and what to
look out for in terms of employee acceptance.
5.4 Limitations and reference for future work
Aside from the insights gained through this thesis, this research also experienced its own set
of limitations. Firstly, this research did not take into consideration that companies that are
already implementing AI are most likely companies that are highly profitable, and that the
addition of the technology did not influence their bottom line. When a company is returning high
profits every year they are more likely to invest in new technologies. Secondly, the timespan
after taken into consideration for the research after the companies has implemented the
technology was only 1 fiscal year. This might be too short of a time to truly be able to detect
serious changes in the company performance as these changes could have also taken place due to
other circumstances such as a particularly bad year for the market, the company writing off an
investment or the effects of the implementation of the AI technology are simply not visible yet
after 1 year. Thirdly, the research does not take into consideration the level of acceptance
38 level of acceptance an employee has for the technology influences the success of the technology
within the company. It could be that the employees at these specific companies used for this
analysis all showcased above average acceptance for the use of AI. This than would tilt the
outcome into a perspective that all companies would not experience if they do not have
employees with the same level of acceptance.
Future works should test the effect of AI on companies over a period of 5 fiscal years. This will
ensure that the effect of the implementation can truly be measured within the scope of the study.
In addition, another research area that should be explored is whether the level of innovation
within a company has any influence on the implementation of AI within companies. An effort to
39
References
Adell, E., Varhelyi, A., & Nilsson, L. (2014). The Definition of Acceptance and Acceptability. In M. Regan, T. Horberry, & A. Stevens, Driver Acceptance of New Technology: Theory, Measurement and
Optimisation (Human Factors in Road and Rail Transport) (pp. 11-21). CRC Press.
Anthes, G. (2017). Artificial Intelligence Poised to Ride a New Wave: Flush with recent successes.
Communications of the ACM, 60(7), 19-21. doi:10.1145/3088342
Anton, C., Camarero, C., & San Jose, R. (2014). Public employee acceptance of new technological processes: The case of an internal call centre. Public Management Review, 16(6), 852-875. doi:10.1080/14719037.2012.758308
Au, N., Ngai, E., & Cheng, T. (2008). Extending the Understanding of End User Information Systems Satisfaction Formation: An Equitable Needs Fulfillment Model Approach. MIS Quarterly, 32(1), 43-66.
Benitez, J., Chen, Y., Teo, T. S., & Ajamieh, A. (2018). Evolution of the impact of e-business technology on operational competence and firm profitability: A panel data investigation. Information &
Management, 55, 120-130.
Bohannon, J. (2015, July 17). Fears of an AI pioneer. Science, 349(6245), 252. doi:10.1126/science.349.6245.252
Brandon-Jones, A., & Kauppi, K. (2018). Examining the antecedents of the technology acceptance model within e-procurement. International Journal of Operations & Production Management, 38(1), 22-42.
Brynjolfsson, E., & Mcafee, A. (2017). The Business of Artificial Intelligence. Harvard Business Review. Capgemini (Director). (2018). WHAT'S NEXT. | Artificial Intelligence [Motion Picture]. Retrieved from
https://youtu.be/2br8yji-rcM
Choong, K. K. (2014). Has this large number of performance measurement publications contributed to its better understanding? A systematic review for research and applications. International Journal
of Production Research, 52(14), 4174-4197. doi:10.1080/00207543.2013.866285
CNH Industrial NV. (2017, January 13). CNH Industrial Newsroom. Retrieved from CNH Industrial web site: https://media.cnhindustrial.com/EMEA/CASE-IH/case-ih-celebrates-175-years-at-the- cutting-edge-of-agricultural-equipment-production-in-2017/s/7014604c-26f4-4f80-a574-a8120da4fa1f
Davenport, T. H., & Ronanki, R. (2018). Artificial Intelligence for the real world: Don't start with moon shots. Harvard Business Review, 96(1), 108-116.
Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1989, August). User Acceptance of Computer Technology: A comparison of two theoretical models. Management Science, 35, 982.
De Fauw, J., Ledsam, J. R., Romera-Paredes, B., Nikolov, S., Tomasev, N., Blackwell, S., . . . Meyer, C. (2018). Clinically applicable deep learning for diagnosis and referral in retinal disease. Nature
40 Denning, P. J., & Lewis, T. (2017, January). Exponential Laws of Computing Growth. Communications of
the ACM, 60(1), 54-65. doi:10.1145/2976758
Evans, R., & Gao, J. (2016, July 20). DeepMind AI Reduces Google Data Centre Cooling Bill by 40%. Retrieved from Deepmind Web site: https://deepmind.com/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-40/
Fernando, I. (2018, June). Transparency and Trust Vital for AI to Succeed. NZ Business + Management, pp. 10-11.
Gesit, E. M. (2017). (Automated) planning for tomorrow: Will artificial intelligence get smarter? Bulletin
of the Atomic Scientists, 73(2), 80-85.
Ghazizadeh, M., Lee, J. D., & Ng Boyle, L. (2012). Extending the Technology Acceptance Model to assess automation. Cognition, technology & work, 14, 39-49. doi:DOI 10.1007/s10111-011-0194-3 Goertzel, T. (2014). The path to more general artificial intelligence. Journal of Experimental & Theoretical
Artificial Intelligence, 26(3), 343-354. doi:10.1080/0952813X.2014.895106
Hardy, Q. (2017, August 23). 3 Ways companies are building a business around AI. 2-6. Boston, Massachusetts, United States of America: Harvard Business School Publication Corporation. Retrieved from https://hbr.org/2017/08/3-ways-companies-are-building-a-business-around-ai Heess, N., TB, D., Sriram, S., Lemmon, J., Merel, J., Wayne, G., . . . Silver, D. (2017). Emergence of
locomotion behaviours in rich environments. arXiv preprint arXiv:1707.02286.
Koninklijke Philips NV. (2018). AI research at Philips Research North America. Retrieved August 10, 2018, from Royal Philips web site: https://www.philips.com/a-w/research/research-programs/ai-research-at-philips-research-north-america.html
Koninklijke Philips NV. (2018, March 1). Philips News center. Retrieved from Royal Philips web site: https://www.philips.com/a-w/about/news/archive/standard/news/press/2018/20180301-philips-launches-ai-platform-for-healthcare.html
Krejcie, R., & Morgan, D. W. (1970). Determining sample size for research activities. Educational and
psychological measurement(30), 607-610.
Kueng, P., Meier, A., & Wettstein, T. (2001). Performance Measurement Systems Must Be Engineerd.
Communications of the Association for Information Systems, 7(3), 1-27.
Luger, G. F. (2008). Artificial Intelligence: Structures and strategies for complex problem solving (6th ed.). Pearson.
M, M. (2018, June 25). AI adoption. (S. Pikero, Interviewer)
Martinez, M., Fernandez, F., & Borrajo, D. (2016, September). Planning and execution through variable resolution planning. Robotics and Autonomous Systems, 83, 214-230.
Moor, J. (2006). The Dartmouth College Artificial Intelligence Conference: The Next Fifty Years. AI