A Work Project, presented as part of the requirements for the Award of a Master’s Degree in

Business Analytics from NOVA School of Business and Economics

The Impact of Artificial Intelligence in Banking

José Maria Themudo – 28872

A Field Lab regarding Artificial Intelligence in Digital Business, under the supervision of:

Professor João Castro

Lisbon, December 2021

Abstract

The following thesis will focus on the general topic of Artificial Intelligence (AI). The main

purpose of this work is to investigate how generally AI is being implemented and developed in

modern times. As banks process almost all individuals’ purchases, these can analyse those and

know their customers better to deliver them a powerful experience. This research aims to access

the current impact of AI in banking and the consequences of it, and how banks benefit now and

then. Banks will use AI to deliver fully personalized experiences and to perform their operations

with more efficiency.

Keywords

Artificial Intelligence, Banking, Automation, FinTech

1. Introduction

1.1 Background
The Covid-19 pandemic crisis has supercharged the adoption of Artificial Intelligence (AI).

According to a PWC study, 52% of companies accelerated their AI adoption plan due to the

pandemic. Of this 52%, 86% have claimed that AI will become a “mainstream technology” in

their company in 2021 (PWC 2020). The trend does not seem to be just a trend of the pandemic

but a trend that will continue through the 2020s. A survey by The AI Journal has demonstrated

that leaders are confident that AI will play a significant role in the future. 74% predict that AI

will bring more efficiency to business processes, create new business models (55%), and help

create new products and services (The AI Journal 2020). The increasing adoption of AI means

it is being used in very distinct ways by the different industries to perform various tasks and

achieve very distinct goals. As such, to continue to move toward its crucial to continue to

revolutionize and update its processes. An excellent way to do so is by gaining insight into

different industries that have personalized and created processes to maximize AI and learn from

it. Usually, these techniques imply gains in productivity in production processes and daily-

routine tasks since these can be automated by machines or fully digitalized. However, it is vital

to provide a definition and an understanding of what AI is, what it might achieve and the main

risks and benefits it brings across several industries and sectors clear to the general population.

Usually, when people think about AI, a significant portion still thinks that it is a tool to eradicate

jobs, increase revenues and collect data to breach privacy. According to a study performed by

the Oxford Commission on AI & Good Governance, 47% of North Americans and 43% of

Europeans think that including AI in our lives will mostly be harmful. If we explore the answers

by profession groups, construction and manufacturing workers are the most worried – 42%

think it will be mostly harmful –, while agricultural workers are the least worried ones – 28%

think it will be harmful, while 38% believe it will mostly help (Oxforf Internet Institute 2020).
1.2 Motivation and Purpose
The main goal of this of how the different industries could learn from each other and what

patterns are visible throughout the diverse industries.

The rest of the paper is organized as follows: In chapter two, the authors analyse the concept of

Artificial Intelligence, its history, and its acceptance. The third chapter comprises an overview

of the different fields of artificial intelligence. In the next chapter, the authors examine the

different benefits and risks of AI, both in a general overview and a sector-specific analysis. The

fifth chapter is an assessment of investments in AI in an overall and sector-specific evaluation.

The sixth chapter will cover a more in-depth dive into specific sectors. Finally, conclusions and

comparisons are in the seventh chapter.

2. Literature Review

2.1 The concept of Artificial Intelligence

The definition of AI has been a topic subject to great discussion due to the lack of consensus in

defining it. So much so that no singular definition of the field is universally accepted. While

numerous definitions of AI have emerged over the last few decades, John McCarthy provided

the following in 2004, “It is the science and engineering of making intelligent machines,

especially intelligent computer programs. It is related to the similar task of using computers to

understand human intelligence. Still, AI does not have to confine itself to methods that are

biologically observable” (McCarthy 2007). However, a distinct and more comprehensive

definition was proposed by Nils Nilsson in 2010, “Artificial intelligence is that activity devoted

to making machines intelligent, and intelligence is that quality that enables an entity to function

appropriately and with foresight in its environment” (Standford 2016). The main limitation in

defining AI as merely making machines intelligent is that it does not clarify what AI is and

what exactly is an intelligent machine. Thus, Britannica brought forth another definition along

these lines, “The ability of a digital computer or computer-controlled robot to perform tasks
commonly associated with intelligent beings. The term is frequently applied to the project of

developing systems endowed with the intellectual processes characteristic of humans, such as

the ability to reason, discover meaning, generalize, or learn from past experience” (Britannica

2021). Simply put, AI is the intelligence that is manifested by machines. These are programmed

to mimic human actions in order to, later on, be able to execute activities that are commonly

correlated with human minds, including problem-solving, learning, and performing physical

tasks (Advani 2021).

Strong artificial intelligence (AI), also referred to as general AI or artificial general intelligence

(AGI), is a theoretical kind of AI that describes a particular approach to AI development. Strong

AI aspires to mimic human functions including reasoning, planning, and problem-solving.

Simply said, Strong AI strives to develop intelligent machines that are indistinguishable from

human mind (IBM 2020). Weak AI, also known as narrow AI, focuses on a single activity, such

as answering questions or playing chess depending on user input. It can only perform one sort

of activity at a time, while Strong AI can handle a wide range of tasks. To ensure accuracy,

Narrow AI relies on human intervention to specify the parameters of its learning algorithms and

to provide appropriate training data and eventually educate itself to tackle new problems. On

the other hand, while it accelerates it is the growth phase, strong AI does not require human

input, eventually, it will teach itself how to solve new issues (IBM 2020).

2.2 The history of AI – Main marks in history

The idea of a “Machine that thinks” goes way back, being first mentioned in ancient Greece,

but gained particular importance after the birth of computing. Specifically, when Alan Turing

published Computing Machinery and Intelligence and posed the question “Can machines

think?”. Six years later, John McCarthy coined the term “Artificial Intelligence” for the first

time. Since then, AI has come a long way, from different programs like Deep Blue, Watson and

AlphaGo defeating champions in Chess, Jeopardy and more recently GO, a Chinese game, with
great complexity. Neural networks have also significantly evolved, starting in 1967 with Frank

Rosenblatt and its Mark 1 Perceptron, the first computer-based neural network learned through

trial and error. More recently, Baidu’s Minwa supercomputer identifies and categorizes images

with much higher efficiency than humans. AI has also come far in terms of practicality; in the

past, it was pure fiction. Nowadays, it is embedded in daily life, with most people carrying an

artificial assistant in their pocket. Likewise, AI is present in most, if not every, industry, being

used as a tool to reach multiple goals in a wide variety of scenarios, including control

management, personalization, customer interaction, decision making, and much more.

2.3 AI Acceptance
The importance of AI is rising in all parts of society. It is regarded as a source of competition

and innovation as it proposes targeted solutions in different areas. However, despite many

people using AI every day, it is evident that not everyone accepts or agrees with it (Arnold

2021).

On the one hand, recent studies show that the support towards AI is more significant among the

wealthy, educated, and those who have more experience dealing with technology. On the other

hand, an analysis by the OECD reveals that subgroups are more vulnerable and less enthusiastic

towards AI and workplace automation. These include people from developed countries with

lower levels of education, low incomes, and individuals whose jobs could easily become

automated. Hence, people who struggle to pay their bills regularly are more hostile towards AI

and robots than those who never experienced difficulties. Moreover, other analyses further

mention that men are, in general terms, more accepting of AI than women, which could be due

to the fact that women have shown greater distrust in technology than men (Zhang and Dafoe

2019).

On a different note, Pegasystems, a leader in customer engagement software, carried a global

study to assess consumer views on AI. The data unveiled that consumers have mixed feelings
towards AI. While most are eager to welcome AI and recognize a promising future ahead, others

fear AI and still favour human communication over a machine when given the option. Another

aspect that is open to discussion is that consumers believe AI falls short of fulfilling their

expectations, mainly due to the lack of understanding of the concept. Thus, there is room for

companies to take advantage of this uncertainty and align their approach with their consumers’

preferences (Pega 2019).

The acceptance of artificial intelligence also differs from country to country. In Asia, the

perception of AI is usually favourable, around two-thirds or more in most Asian countries. For

example, Singapore (72 percent), South Korea (69 percent), India (67 percent), Taiwan (66

percent), and Japan (65 percent) believe AI has benefited society (World Economic Forum

2020). However, most of the other continents and regions polled do not agree that AI has

benefited society. Countries such as France, the UK, and the US are predominantly negative

towards AI’s impact on society. On the other hand, Sweden and Spain are two of the few

countries outside of Asia-Pacific where a majority of people (60 percent) think AI is a good

thing (World Economic Forum 2020).

Consumers and corporate leaders have different concerns about AI. In the next five years, AI

will significantly impact the way companies do business, according to 85% of CEOs. However,

there are differing viewpoints on how much AI may be trusted. Over three-quarters of CEOs

believe AI is “positive for society,” but even more say that AI-based choices must be

explainable to be trusted (84 percent) (World Economic Forum 2019). A survey from the

Economist assessed whether executives thought AI could live up to its hype or not. A fifth of

the respondents claimed that AI was” just hype and no substance”, whereas 36% stated they

thought it helpful but that there is hype in this technology. On the other hand, 36% thought AI

would live up to its promises (The Economist 2020). The pandemic changed a little bit this

perception. Since the pandemic, 50% of respondents stated they have a more favourable view
of AI. Of these respondents, the most confident industries that AI would help them during the

pandemic were mining, manufacturing, and technology. The least confident ones were travel

and tourism, consumer, and retail (The Economist 2020). The adoption rate is also a good way

of measuring how accepting AI is in varied industries. A study from O’Reilly shows that

Computers, electronics, and technology, unsurprisingly, topped the list with 17% of

respondents. Financial services (15%), healthcare (9%), and education (8%) are the industries

that are using AI the most. In the pharmaceutical and chemical industries, we find minimal AI

utilization (2%). Similarly, only 2% of responders are from the automotive industry, even

though AI is critical to emerging goods such as autonomous vehicles. Finally, the energy

industry accounted for 3% of the respondents, while public utilities accounted for 1% (O'Reilly

2021).

3. The main fields of Artificial Intelligence

The objective of this section is to present and explain the main fields that exist related to AI.

These are the main tools used to deliver tasks once performed by human intelligence such as

learning, reasoning, solving problems, identifying, and understanding languages and perceive

specific situations or environments (Future Today Institute 2021). The fields found are based

on the interest put on by several studies and books, such as (Nilsson, The Quest for Artificial

Intelligence 2009), (Minsky 1960), and (Russell e Norvig 1995), and include Machine Learning

(ML), Robotics, Natural Language Processing (NLP), Computer Vision and Expert Systems.

3.1 Machine Learning (ML)

ML is a field of AI where algorithms learn from huge amounts of data without human

intervention to improve the accuracy in making predictions and pattern detection (Nilsson

1998). According to (Russell e Norvig 1995), inside these ML algorithms we can still make the

distinction between three: supervised, unsupervised and reinforcement learning. Supervised

learning is a subset of ML that uses labelled datasets (e.g. if we are trying to predict the age of
someone, age is our label), meaning that the label is the variable we want to predict – if both

inputs and outputs can be seen, that is supervised learning. The most common models are

Regressions (to predict outputs that are continuous variables, such as the height of someone)

and Classification (the outputs are discrete variables, and sometimes “yes” or “no”, like when

we’re trying to prevent the churn rate of a website). Unsupervised learning uses unlabelled

datasets, meaning that we have no output variable to predict. The most common algorithms are

Clustering (group data based on similarity) and Dimensionality Reduction (simplify data by

reducing the number of features). Finally, reinforcement learning algorithms deal with no data

and must solve problems. Instead of dealing with data, it deals with an environment, where

correct decisions give rewards and wrong decisions give punishments. One known use case is

chess – the algorithm will try random moves and receive rewards or punishments for them and

then will learn how to reach the terminal state: check! (Russell e Norvig 1995).

3.2 Robotics

Robotics develop artificial agents that are designed to perform human tasks and interact in the

physical world environment (Russell e Norvig 1995). It literally studies robots (designs,

production, and operation), machines that can be used to perform human tasks alone or

supervised by someone. They can be used in manufacturing automation, exploration of the sea,

hazardous waste inspection, surgeries, environment monitoring (e.g., drones), home robotics

(e.g., autonomous vacuum cleaners), among others. According to Medium, regarding the main

fields inside this subject, the focus is on Operator Interfaces, Mobility or Locomotion,

Manipulators and Effectors, Programming, and Sensing and Perception. As for the first one, an

operator interface is a vehicle through which the user of the robot and the robot itself

communicate, meaning that it is the platform through which a human gives instruction to a

robot. Robots need to move from a place to another to complete their job, and that’s why

mobility and locomotion are a field of great interest (for instance, drones use propellers and
other systems to move, and those need to be developed). These machines also need to grab,

transform, and move objects like they had a human hand (one of the most common applications

is in the auto-industry), and that’s why researchers are putting effort in manipulators and

effectors. Programming is important since it is the language used by the user to communicate

with the robot (and this is where ML might help, since robots can learn to avoid mistakes).

Finally, sensors are fundamental to collect data to inform the robot about the environment

around it (Medium 2020).

3.3 Natural Language Processing (NLP)

NLP gives computers the ability to extract data from written and spoken words, allowing for

tasks such as translation and speech recognition (Nilsson 2009). Combined, computational

linguistics, statistics and deep learning models can perform NLP tasks. It translates pieces of

text from a language to another, allows machines to answer to voice commands (e.g., Hey Siri),

serves as the motor to chatbots as they identify the text written, process and understand it and

powers GPS systems to talk, for example. Concluding, the main fields of interest are text

processing, speech recognition and speech synthesis – and most of them rely on ML algorithms

to perform (IBM 2020).

3.4 Computer Vision

Computer Vision deals with allowing machines to collect information through vision, analysing

images or videos to make predictions and pattern recognition (Nilsson 2009). We need to

distinguish this concept from image processing – the last one aims to create an image from an

existing one, while computer vision aims at understanding what is happening (Machine

Learning Mastery 2019). When there is a surveillance system, it is not uncommon to have two

cameras capturing some common zones – that is actually a good way to interpret distances

between objects and help in making decisions. The main tasks and fields computer vision is

exploring are optimal character recognition (e.g., when someone gets a speed ticket, the camera
automatically reads the plate number), machine inspection (e.g., scan the status of the outside

of a plane to check if it is ready to fly again), retail (recognition of products for an automatic

checkout, like the one used in the Amazon Go shops), warehouse logistics (development of

package deliveries in an autonomous way), medical imaging, self-driving vehicles (by

analysing the environment, the car adapts the speed and the turns), 3D model building (from

information collected by drones and planes), motion capture, surveillance (analysing the traffic,

drones in the sea, etc) and fingerprint recognition and biometrics (Szeliski 2021).

3.5 Expert Systems

Expert Systems are computer systems simulating complex human decision-making by

explaining the reasoning behind it. These are being integrated with databases to do recognition

and decision-making like humans, with the final objective of creating knowledge discovery

with the help of data-mining processes and end up with an intelligent database. It brings visible

advantages, such as an increased availability (since these systems are not specific to a single

computer), reduced costs per user (as a fixed cost, if the number of users increases, the average

cost decreases), consistency (if a human is tired will treat some problems with greater difficulty,

while an expert system is always ready), multiple expertise converted into one machine,

explanation of the reasoning behind the decisions done, quick responses and emotion-free

opinions (Jackson 1998).

Figure 1: : An Expert System, based on (Jackson 1998)

Most of the times these fields are not exclusive and require connections to achieve better results.

For example, Deep Learning (DL), a subset from ML that trains techniques such as Artificial

Neural Networks (ANN), techniques are used in NLP algorithms, since these are built to tackle

and understand the temporal nature of language (Jurafsky e Martin 2000). ANN are a set of

networks composed by non-linear elements (Nilsson 1998) and are compared to real neural

networks since they adjust the weights of connections with new inputs just like the human brain

connections between neurons change with more information, and both learn to make more

accurate decisions (Russell e Norvig 1995). These DL algorithms are used in Computer Vision

(Goodfellow, Bengio e Courville 2016) like the recognition of sound waves from the vibrations

they reproduce in objects seen in video (Davis et al. 2014) Also, DL algorithms are used for

Robotics. (Punjani e Abbeel 2016) used a DL algorithm to try to represent the dynamics from

a helicopter through a network model, and the model outperformed the baselines by large.

From the relations stated above and the studies observed, we can check that specially DL is

used to complement several other fields of AI. One of the best-known cases of how these Neural

Networks are present in our lives is to look at how easy it is to communicate with an iPhone

just by saying “Hey Siri” – a practical application of DL in speech recognition and NLP. The

holder of the iPhone says “Hey Siri”, there is a Deep Neural Network (DNN) that converts the

acoustic pattern of the voice into a probability distribution, and then temporal integration to

calculate a confidence score to check whether “Hey Siri” was said or not (Apple Machine

Learning Research 2017). If instead of talking to Siri, the holder just wants to unlock its iPhone

through face detection, another DNN – and here we have DL helping Computer Vision.

4. The benefits and risks of AI

4.1 General / Ethics of AI

The use of AI involves several benefits and is accompanied by numerous risks at the same time.

Davenport and Ronanki (2018) are approaching AI through the business lenses and describe it
as a cognitive technology which benefits three major business needs: automating business

processes, gaining insights through data analysis, and engaging with customers and employees.

The most common type is the automation of digital and physical processes, which are generally

easy and cost-effective to implement and usually bring a quick and high return on investment.

It is especially useful for automating back-office work like transferring data and updating

customer records or extracting information from multiple document types using natural

language processing. These kinds of business-processes are often outsourced offshore and can

be automated, which results in reduced costs without a loss of employees (Davenport and

Ronanki 2018). Routine operational activities, such as maintenance systems, accounting and

information inquiry tasks are performed much better and faster by AI systems than by humans

(Lee and Yoon 2021).

The second most frequent use of AI is in the field of data analytics with the help of algorithms

for pattern detection and interpretation in order to gain business relevant insights. With the help

of data analytics, companies can predict what a customer is likely to buy, identify credit fraud

in real time or automate personalized targeting of digital ads. Such tasks are often beyond

human ability and therefore do not pose a significant threat to human jobs (Davenport and

Ronanki 2018).

Cognitive engagement represents another main benefit of AI, which is used less frequently by

companies compared to automating business processes and gaining insight through data

analysis. This category includes the deployment of intelligent agents that offer customer service

at any time. It is also used within companies on internal sites for answering employee’s

questions regarding HR or IT related topics. In addition, it comes to use in health treatment

recommendation systems that help to customize care plans under consideration of individual

patients’ health status and previous treatments (Davenport and Ronanki 2018).
Also, companies can stay ahead of the competition by transforming products and services. In

R&D heavy sectors, AI can accelerate the product innovation and discovery process through

accessing new value-adding areas (Møller et al. 2018). A widespread concern in society is that

AI will replace a majority of jobs performed by humans. A health app called Noom provides

customized support to their clients in order to help them attain their health goals. From 2017 to

2019 the number of Noom’s employees rose from 77 to 1100. That significant increase shows

that AI can not only help to improve products and services, but also that AI can facilitate the

creation of many new jobs (Lee and Yoon 2021).

The numerous benefits of AI are accompanied by a lot of risks and proves this technology to

be a double-edged sword. Used responsible, it can improve our lives in many ways. Yet even

AI generates business value and consumer benefit, it is also giving rise to umpteen risks and

unwanted consequences which negatively affect individuals, organizations and society

(Cheatham et al. 2019).

On an individual level, AI can be harmful in economical, psychological as well as physical

ways. The underlying black box character is the root cause for unexplained actions the

algorithm might perform, like unreasonably banning a client's credit card or unjustifiably

accusing a person of a crime (Diakopoulos 2016; Dourish 2016). Further, physical threats arise

through autonomous-vehicle malfunction, overreliance on inadequate equipment predictive-

maintenance or the misdiagnose of medical conditions by machine learning models (Cheatham

et al. 2019).

Another risk of AI is the unequal distribution of power. On the one side, governments and

organizations have access to useful tools and resources like client data and highly developed

technology. On the other side is the majority of society which does not have access to these

powerful tools (Someh et al. 2016). 2016, Facebooks’ CEO Mark Zuckerberg was accused of
abusing his power when algorithms of the social media platform censored the photograph of

the “napalm girl”, a historic picture of a naked girl which is iconic for the Vietnam war. Another

example is the use of automated bots on Twitter, Facebook, and Reddit during the 2016 U.S.

presidential election and UK European Union Referendum, which interacted with users and

promoted certain content and viewpoints (Mittelstadt 2016).

A further risk of AI are possible discriminatory effects. The below figure shows the results of

a study about the COMPAS machine learning algorithm by ProPublica which is one of the

most prominent cases about discrimination through AI. The algorithm is used by judicial

systems in the U.S and supports judges in their decision making through assessing the risks of

former prisoners to become delinquent again when released from prison. Figure 2 shows the

error rates of the COMPAS algorithm and its discriminatory bearing. African Americans are

almost twice as likely as white people to be labelled a higher risk but actually did not re-offend,

whereas white people are labelled a lower risk yet did re-offend way more often than African

Americans (Yapo und Weiss 2018).

Figure 2 - Disproportionate error rates, from Ethical Implications Of Bias In Machine Learning

(Yapo und Weiß 2018)

The case of the COMPAS algorithm is just one of many examples of discriminatory effects

induced by AI. The origin of the problem is the underlying bias due to bad data and lack of
inclusivity. If a ML model is trained solely on data of a specific group, it is obvious that the

result is less divers and might neglect certain groups with fewer occurrences in the training data

(Yapo und Weiß 2018).

Further, the close relationship of big data and AI implies concerns about privacy and data

protection. The processing of personal data by algorithms proves to threaten individuals'

privacy, especially when it comes to analysing and predicting socio-economic aspects of

individuals. A survey revealed that two thirds of Europeans expressed concerns regarding data

security (Payne et al., 2015) and also the majority of Americans are worried about the use of

their personal data by companies and governmental institutions (Auxier et al., 2019).

In the following, we will see that the benefits and risks discussed so far do not apply for a

particular sector only but can be found in most common industries in which AI comes into use.

4.2. Benefits and risks per sector

4.2.1. Banking and Finance

Financial institutions are trying to change the way they interact with their clients or customers.

One of the main benefits AI brought to these sectors is the way they interact with their clients

through insights and advice, giving them a personalized experience to their needs and objectives

(Tink, Tink's guide to Improving digital banking through personalised insights 2021). This

contributes to a better customer or user experience, increasing the probability of having a client

staying for a longer time and recommending the services to its peers – more than 50% of bank

clients think personalized experiences are drivers to trust the institutions. These technological

mechanisms can speed up decision making when it comes to allow or not for a loan – risk

assessment mechanisms and underwriting processes get faster with machine learning

algorithms since these can manage multiple data sources at the same time. This means a faster

loan generation, and clients get happier in a shorter time (Deloitte US 2021). Also, regulatory
reporting becomes easier, clearer, and more accurate with the help of AI. There is no need for

a lot of manual interventions like mappings and reviews, regulatory changes will be easier to

address due to the speed machines take in adapting and the quality of the reports increases (PwC

2020). Finally, fraud detection and anti-money laundering mechanisms are developed through

machine learning techniques as well, preventing financial institutions to process illegal

transactions and detect potential crimes. These are used to detect fraud patterns and do real-

time analysis of movements to perform risk-monitoring (McKinsey 2020).

Regarding the risks, the amount of data collected by these institutions to create tailor-made

products can make them breach privacy laws – if they hold the data for longer and for more

purposes than the ones held in the contracts with the customers (Atkins e Luck 2021). Also,

because of the huge amount of data collected, banks and other financial institutions get more

prone to be victims of cyberattacks and information leaks. These algorithms are designed to

help in decision making, and if they are not well done, final outcomes might be biased or not

accurate. Financial institutions need to be careful when dealing with these outcomes and make

sure that they are transparent, accurate and aligned with the culture and objectives of the firm

(Deloitte, AI and risk management: Innovating with confidence 2018). If we say that

personalized banking experiences might be a driver for increasing customer loyalty, the fact

that there are fewer human interactions can be a reason to decrease that same metric, and

therefore institutions need to look carefully at this – a few years ago, changing from one banking

institution to another would be a high-cost process, while today, with the developments seen in

AI, this is way easier and with increasing customer demand for good services these financial

institutions will need to step up their game and provide reasons for their customers to stay

around longer (Forbes 2019).

4.2.2 Healthcare
“An ounce of prevention is worth a pound of cure.”, said Benjamin Franklin, one of the

Founding Fathers of the United States of America, back in 1735.

One of the main benefits AI and robotics are giving to this sector is early detection of diseases

such as cancer. The American Cancer Society says 12.1 million mammograms are done per

year in that country, and the usage of AI is making review and translation 30 times faster and

with an accuracy of 99% (Wired 2016). Diagnosis is also easier with the help of these

algorithms. 80% of health data comes unstructured, making it hard to read, meaning that only

20% of the data is easy to read by computers (such as numerical data or records pre-organized

by humans) (Healthcare Data Institute 2015). IBM developed Watson for Health, and this tool

processes and stores far more data than any human, allowing for a quicker and more accurate

diagnosis. Also, predictive analytics tools can inform and support clinical decisions and help

doctors to prioritise tasks on the treatment (PwC 2017). AI may also benefit the treatment

process for patients and help doctors manage the treatment plans – example of AiCure, a

platform that helps people with long-term conditions to comply with their medications by

visually recognizing the face of the patient, the medications it is taking and to confirm the

success of the ingestion. It also offers virtual assistance to the patients and assesses the

progression of conditions over time (Vasishtha 2018). Finally, it accelerates the process of

putting medical solutions in the market. The California Biomedical Research Association

estimates that it takes 12 years for a drug to go from discovery to the patients. With the help of

AI, these processes last less, and drugs are becoming available in a shorter period (PwC 2017).

Just look how fast the Covid-19 vaccine took from discovery to the market – the pandemic

started in the end of 2019 in China, reached the rest of the World by the beginning of 2020 and

by the 8th of December 2020 the first person in the UK received the first shot of the Pfizer
vaccine (BBC 2020). This is one of the great examples showing how can technology help the

healthcare industry.

Regarding the risks, one that arises is data bias. Training AI models needs a huge scale of input

health data, and if the data used for training does not fit to the population to which the solutions

are being applied to. Insufficient or bad quality data can also lead to this bias (Sunarti et al.

2020). According to the same authors, there are privacy issues regarding the hold of sensitive

health data – the privacy of the individual is an ethical obligation. Finally, we need look

carefully for how these algorithms are built and how accuracy might not be a good indicator of

performance. Imagine a program that aims at identifying if a tumour is malign or benign that

has an accuracy of 99% – this means that for every 100 predictions, it gets 99 right. What if the

one left is a malign tumour that is predicted as benign? The objective of this algorithm is to

identify the bad cases, and not the good ones. Therefore, imbalanced classes must be looked

after when dealing with these algorithms, and sometimes the technology will not be enough to

make decisions.

4.2.3 Agriculture
Similar to the healthcare industry, AI can be used beneficially for the classification and

prediction of crop diseases. Based on input parameters about the physical constitution of the

plant, diseases can be forecasted and appropriate measures for prevention and recovery taken

on an early stage (Tilva et al. 2013). Image processing coupled with an artificial neural network

for example helps to classify seedling diseases (Huang 2007) or to detect the percentage of

infection in leaves (Sannakki et al. 2011)

Further, rule-based crop management systems provide an interface for general management of

all sorts of crops and give advice regarding crop selection, fertilizer application and pest related

issues (Bannerjee et al. 2018).

Next to the monitoring of diseases and pests, the handling of harvested crops is also a crucial

aspect of agriculture. Bannerjee et al. (2018) refer to various AI powered food monitoring and

quality control mechanisms for the storing, drying and grading of harvested crops.

AI is also employed for soil and irrigation management. Rule based expert systems evaluate the

design and performance of micro irrigation systems (Brats et al. 1993) or recommend crops

depending on land suitability (Sicat et al. 2005). Further, AI is used for estimating soil moisture

(Arif et al. 2013) and predicting rainfall using atmospheric inputs (Manek and Singh 2016).

Moreover, AI models are applied to predict crop yield, which is beneficial for estimating crop

costs and developing marketing strategies (Bannerjee et al. 2018). Overall, such AI applications

provide tremendous support in the decision making for farmers and create value in terms of

enhanced crop yield, efficiency and environmental sustainability.

In terms of social sustainability, however, risks due to the dehumanizing character of AI arise.

One thinks of driverless machines and other robots that increase the risk replacing traditional

farmer jobs. In order to prevent social inequality, the ethics of AI need to be considered similar

to other sectors and concerns over data privacy, transparency and unintended consequences of

the technology require significant attention (Lakshmi and Corbett 2020).

4.2.4 Retail
Artificial intelligence has reinvented the retail landscape and is expected to continue that trend.

AI is expected to boost wholesale and retail gross value by $2.2 Trillion by 2035 (Statista 2021).

The most notable area in which AI has brought great benefits to retail is the customer

experience. AI-assisted conversational assistants help customers navigate questions, FAQs, and

troubleshooting and redirect them to a human expert when necessary, improving the customer

experience by providing on-demand, always-available support while streamlining staffing

(Deb Marotta 2020). Chatbots, for example, can respond to many questions at the same time.
This is a lifesaver for companies with overburdened call centres and long wait times. It allows

customer service departments to perform more, resulting in a better client experience

(Salesforce 2021). Another significant benefit that helped retailers like Amazon become the

behemoths today is AI's personalisation to retailing. Personalisation in advertising refers to the

use of data or consumer insights to improve an ad's relevance to its intended audience. This can

include information like demographics, interests, purchasing intentions, and behaviour patterns.

Increasing the relevancy and personalisation of adverts is becoming a primary priority since it

significantly improves the user experience and customer retention (IBM 2021). Customers can

benefit from artificial intelligence in retail by making product discovery more straightforward

as well. Customers may now take a picture of a product they like in the real world and use it to

find an online store that sells it (Forbes 2020). Another central area in which AI can create

significant benefits for retailers is in making operations more effective. AI can be utilised in

forecasting, demand planning, assortment, allocation optimisation, and return optimisation in

the supply chain. "When you are shipping billions of packages every year and working with

tens of millions of products, you can't do it in that manual process," said Steve Gurney, head of

worldwide general merchandise at Amazon Web Services (National Retail Federation 2021).

Another area where AI has succeeded is in streamlining warehouse and in-store store

operations. From the amazon robots that help Amazon employees in the packaging process in

warehouses to in-store where using AI, shops can easily optimise their space and inventory.

Existing consumer preferences, product location, season and weather conditions, expiration

dates, and other factors are considered by algorithms to put shelves and products where visitors

anticipate to find them intuitively (CHI Software Developmet 2021).

Nevertheless, AI does bring some disadvantages to the retail industry. For once, privacy

violation and data abuse can destroy an organization’s reputation and customer trust. More than

half of executives express "serious" or "severe" concern about AI's ethical and reputational
hazards in their firm. That means that developing an AI ethical risk program that everyone buys

into is required before AI can be deployed at all. For companies that use AI, this must be a

priority (Harvard Business Review 2021). Another risk is the replacement of the workforce.

According to economists at MIT and Boston University, robots could replace as many as 2

million more employees in manufacturing alone by 2025. “This pandemic has created a very

strong incentive to automate the work of human beings,” state Daniel Susskind, a fellow in

economics (Time 2020).

4.2.5 Construction
The adoption of AI in the construction industry is quite low compared with other industries,

even though it encompasses many possibilities and potential use cases. AI can be beneficial for

optimizing project schedules and for enhancing project planning. I addition, image recognition

and classification on work sites can identify and assess unsafe work behaviour. Moreover,

analysed sensor-data can be used to understand signals and patterns in order to provide real-

time solutions, prioritize preventive maintenance, reduce costs and prevent unplanned

downtime (Bianco et al. 2018).

Automation can replace traditional manual observation, which usually tends to be time-

consuming and prone to errors. In terms on safety, AI helps to detect and predict potential risks,

not only on the construction site, but also when it comes to project management, streamline

operations and budget planning. In addition, robots can deal with unsafe operations and replace

humans in dangerous work environments (Bolpagni et al. 2021). AI can also amplify the

efficiency of the construction execution process through new approaches like process mining.

Repetitive routine tasks can be taken over by robots which work continuously without taking a

break at almost the same quality and productivity (Pan and Zhang 2021).

Important to mention is the role of Building Information Modelling (BIM), which serves as a

“[…] digital backbone to work with AI.” (Pan and Zhang 2021, p.7). Through the collection of
large amounts of data about all aspects of the project, real-time analysis can support to

streamline the complex workflow, make processes more efficient and cut costs. In combination

with AI techniques, computer vision promotes the understanding of data in images or videos

and is used for the inspection and monitoring of complex construction tasks and structural

conditions. It provides actional information about construction safety and can perform

automated damage detection which leads to a safer work environment (Pan and Zhang 2021).

On the other side, the advantageous of digitalization are accompanied by the exposure to

cybercrime and privacy intrusion with potentially huge economic and financial consequences.

Examples of cyber threats in the construction industry include malware, social engineering and

phishing. In addition, construction work is often conducted in unsecure environments. Small

mistakes by AI can comprise the safety of construction workers and lead to life-threating

accidents. Furthermore, the location of construction sites is often secluded and lack power and

internet connectivity. However, AI mostly relies on good internet connectivity and power

supply, which poses another threat to the usage of this technology (Abioye et al. 2021).

4.2.6. Hospitality
The hospitality industry is expected to reach USD 44.38 billion by 2026, and there is an urgent

need to revolutionize it. The most promising approach is to invest more in AI technologies in

the industry and therefore improve its customer service and experience, which they rely upon

heavily. The hospitality industry has adopted digital technology long ago due to the significant

amount of data generated. For hospitality specifically, AI’s primary purpose is to explore and

analyse guest data, aid in decision-making, and manage guests’ complaints. Nevertheless, the

industry could take more advantage of the use of AI and incorporate it into different areas within

hospitality (Roy 2021).

The main benefit for the hospitality industry in adopting AI is that, by doing so, it can offer

services that are somewhat more accurate, timely and efficient, when in comparison to relying
solely on people’s capabilities. Along with this is AI’s ability to provide customers with better

experiences aligned with their interests. This is feasible as AI streamlines processes, analyses

them, and collects valuable data from different sources, therefore improving its

recommendations. Another substantial benefit AI brings for hospitality is its ability to enhance

customer profiles based on previous guests’ history, preferences, and satisfaction, consequently

generating a more loyal customer base (Qualetics 2020). Nonetheless, AI is also beneficial in

the sense that it aids those working in the industry. Germany-based Model One has been testing

a robot nicknamed Sepp to answer simple questions and deliver basic information to customers.

IBM Watson was the mastermind behind Sepp’s creation, and the robot is capable of

understanding people’s requests as well as learning new information. It can, for instance,

provide weather information and let guests know at what time breakfast is served. Likewise, in

Virginia, USA, Hilton has an AI member of staff. Like Sepp, robot Connie can provide helpful

information to their guests and learn from its interactions. However, Connie’s most impressive

capability is its ability to make gestures, just as people do. IBM’s Watson vice president and

chief technology states, “When it is asked ‘where’s the elevator?’, it says it’s down the hall to

the left while pointing down the hall to the left” (Fomby 2019).

Nonetheless, AI-driven robots are not the single domain in which AI is positively affecting the

hospitality industry. In 2018, Avvio, a tech company, launched Allora, the world’s first booking

platform entirely run by machine learning. Traditional booking platforms are unsuccessful in

delivering a personalized experience to their guests. Therefore, Allora consolidates multiple

insights from different users and optimizes their experience by finding the best hotel and

experience. The platform considers thousands of users’ preferences based on geography,

booking history, and other circumstances that impact the hotel selection (Allora 2021). A survey

done by the online platform Booking unveiled that 75% of guests prefer self-service options

thus, making chatbots another great benefit of AI in the hospitality industry. Chatbots are
capable of assisting with current reservations and answering common questions concerning

hotel policies as well as transportation, changes in dates, check-in and check-out times and

payments. Booking’s chatbot is fitted to manage 50% of customers’ “post-booking

accommodation-related requests”. If the chatbot is incapable of answering, the person will be

redirected to a customer service member (Fomby 2019).

Despite all the great benefits that AI has delivered, there are still some risks concerning the

implementation of AI in the hospitality industry. AI is still a very vast field, and although there

has been significant development in recent times, the field is still very fresh, and AI is still

developing. The previous vice president and AI leader of Google, Andrew Moore, even stated,

“AI is currently very, very stupid”. In fact, the term AI Stupidity is used to illustrate AI’s

inability to make sound decisions by only relying on the data that is available. As AI is based

on human input, people are likely to provide inaccurate or biased data, thus leading to inaccurate

or biased decisions. Additionally, businesses are further concerned about data privacy issues.

Despite it being mandatory to follow data privacy laws and their ethical use, data collected

during user interactions could be gathered for devious reasons. Hence, there is a significant risk

of violating data privacy. However, if businesses were to obey every law and regulation, AI

could become a significant source of competitive advantage (Fomby 2019). Likewise, data

privacy issues are also a concern for customers. Many are reluctant to rely on information

delivered by AI-based technologies completely, as the provided data depends on the program’s

quality and algorithms that make the technology work. Therefore, many customers will

continue to seek human help, even when the required information is available (Roy 2021).

Another liability for implementing AI in the hospitality industry is that this type of technology

is expensive to implement and costly to maintain. While technology is more easily attainable

and there are several options available, most hospitality businesses may not have the budget to

invest in AI-driven technologies, thus causing them to lose their competitive advantage (Koo,
C. et al. 2021). Finally, the most pressing risk of implementing AI in the hospitality industry is

unemployment. Most people believe AI will replace humans due to new developments in

technology. Workers in the hospitality industry are fearful for their jobs and anticipate AI will

take over the more obsolete tasks, thus leaving many unemployed. A study developed by

McKinsey Global Institute reveals that intelligent robots will replace 30% of the world’s

working population by 2030. Given the many different tasks that require minimal effort in

hospitality, it is likely that most people will no longer be required and will, in fact, be replaced

by some AI technology (Bughin et al. 2018).

4.2.7 Industrial Products

According to a Deloitte poll on AI adoption in manufacturing, 93 percent of businesses believe

AI will be a key driver of growth and innovation in the industry (Deloitte 2019).

Artificial intelligence has, in many ways, revolutionised the industrial product industry. For

instance, with the introduction of autonomous industrial robots, production improved

significantly. Production can now be operational 24/7, while human beings need rest and

regular maintenance. Robots do not get tired or hungry, and they can operate on the assembly

line 24 hours a day, seven days a week. This enables the growth of manufacturing capacity,

which is becoming increasingly essential to fulfil the expectations of global consumers (Rowse

2019). Additionally, robots are more efficient overall. Artificial intelligence technology ensures

that products satisfy the necessary quality and regulatory requirements. Manufacturers can

accomplish this by incorporating AI technologies such as machine learning and big data into

their equipment, such as tracking sensors (Global Trade 2020). Safety is another topic that AI

significantly improves in this sector. Humans are prone to making mistakes and are fallible.

Errors and mishaps happen on the factory floor and in any building or processing setting; this

is a problem that AI and robotic aid can almost completely eliminate. Furthermore, remote
access control necessitates a reduction of personnel, mainly when the activity is hazardous

(Rowse 2019). AI also enables factories and industrial complexes to minimise operational costs.

According to Deloitte, manufacturing is predicted to generate 1,812 petabytes (PB) of data per

year, outnumbering communications, banking, retail, and several other businesses (Deloitte

2019). Consequently, it can use and develop predictive that programs aid the sector in multiple

stages of the business. For example, Data is collected in real-time to monitor the state of

equipment in predictive maintenance scenarios. The idea is to uncover patterns that can assist

forecast and ultimately prevent failures; AI systems are increasingly being employed to achieve

this goal using learning algorithms. Plants can be more strategic when analysing equipment

state and anticipating when maintenance should be conducted when predictive maintenance is

automated (Stefanini Group 2020). Additionally, producers might synchronise production

schedules to increase output. According to a report by McKinsey, an AI predictive maintenance

model can boost productivity by 20% (McKinsey 2017). It can also save up to ten percent on

maintenance costs. Aside from production, AI plays a vital role in other sectors of

manufacturing. Distribution and supply networks, monitoring, customer behaviour, and

changing patterns are all examples. As a result, AI in manufacturing ensures that businesses

can anticipate market shifts. They can then strategize for better manufacturing and other cost

management processes with this information. Manufacturers can also utilise AI algorithms to

forecast market demand (Global Trade 2020).

Nonetheless, it also does have its disadvantages and challenges. First, the costs of implementing

and maintaining artificial intelligence are substantial. For small businesses and start-ups, the

budget is often prohibitively expensive. Even while artificial intelligence reduces labour costs,

installation, and maintenance costs (Global Trade 2020). Another disadvantage of artificial

intelligence is that it is vulnerable to cyber-attacks. According to a recent World Economic

Forum research, cyber-attacks are among the top five global stability threats (World Economic
Forum 2019). For any manufacturer who uses AI software, this kind of information might be

frightening. Finally, the scarcity of talent and expertise. Because these technologies necessitate

complex programming frequently, it is critical to factor in expert availability. Furthermore,

because such hands are in high demand, the cost of hiring them will be expensive. “Demand

for workers with AI talent has more than doubled over the past three years, with the number of

AI-related job postings as a share of all job postings up about 119%.” (Indeed 2018).

5. Investment in Artificial Intelligence

5.1 Investment in Artificial Intelligence

Artificial Intelligence changed and reshaped the behaviour of most companies, among it also

on their investment approach and strategy. It is important to mention that the aspect of

investments into AI must be divided into two segments: one the one side investments are

conducted internally, therefore focusing on establishing and implementing own concepts

through recruitment of personnel and purchase of assets. On the other hand, investment of AI

is conducted through funding, Mergers & Acquisitions, or strategic funding’s in external

companies, that serve as an asset, to amplify the product portfolio or as a value addition to

existing technologies and processes. The following part will analyse different aspects of

investment activities within AI industry, focusing on the internal investments that companies

have conducted and on differences between external investment throughout the last decade.

5.2 Overall Information of AI Investments

IDC research in 2021 have estimated a global investment value of almost 342 billion USD only

in 2021, forecasting further growth in the upcoming years by breaking the 500 billion USD

mark until 2024. Mayor part (88%) relate to spending’s on AI based Software, followed by

expenditures in AI based hardware (Needham 2021). Within the industry, investments for “AI

services” show the fastest development as it is forecasted at a CAGR (Compound Annual

Growth Rate) of 21% and a total market volume of 50 billion USD by 2025 (Kenyon 2021).
Throughout the last 10 years, companies’ investment in the own AI structures, processes and

human resources have increased drastically as strong raising revenues of AI Enterprises indicate

(Columbus, Forbes 2018). Whereas majority of companies had to make large budget cuts for

operations during the Coronavirus Pandemic, investment in technology and mainly in AI were

maintained or even increased over time (Kark, Gill and Smith 2021). Gartner Research data of

2020 indicate that 66% of organizations decided to actively fund new and existing AI related

approaches to enhance “[…] customer experience […], retention, and revenue growth – along

with cost optimization […]” (Stamford 2020) i. Moreover, 50% of companies of the Life

Sciences, Energy, Retail Consumer Products, Telecommunications, Government and

Automotive industry stated that they are “[…] progressing their AI efforts as planned or even

quickened the pace of deployments” during the economic shutdown caused by COVID-19,

indicating the overweighting benefits generated by AI based approaches (Liu 2020).

Based on OECD research of “Venture Capital Investments in Artificial Intelligence”, the AI

industry is one of the most prosperous industries throughout the last years by growing an

average of 34% Year to Year and being responsible for almost 75 billion USD in VC (Venture

Capital) investments into AI companies alone in 2020 (Tricot, OECD 2021). Investments in US

and China based companies are responsible for almost 80% of the monetary value of the

investments creating an enormous gap towards the EU27 countries that represent an aggregated

9% of total investment volume. Not only has the amount of investment grown over the years

(from 500 in 2021 to almost 3900 in 2019) (Tricot, OECD 2021), but also the average ticket

size per each investment as it almost doubled in most of the regions. Moreover, changes can be

identified in the average ticket size between 2012 and 2020. The amount of tickets with a size

of 10 to 100 million USD per investment has almost been doubled, whereas as strong decrease

(-17%) in investment tickets bellow 1 million USD con be recognized (Tricot, OECD 2021).

Reason for such development is connected to changes in the approach towards start-ups, as the
aspect of long-term growth and maturation through large amounts of cash to finance their

operations is accepted.

Furthermore, Crunchbase studies indicated that since 2000, investments in AI related

companies have increased up to six times, making it one of the fastest growing industries

(Columbus, Forrbes 2018). Further information from the OECD reports indicates that in 2020

over 20% of the overall investments conducted by Venture Capital are related to an AI focused

company (Tricot, OECD 2021). Enlarged investments are mainly related to potential high

return of investments due to the growing demand on customer side and constantly developing

technological market standards.

5.3 Leading Investment Companies & Major Investments

Most of the leading successful Venture Capital companies are based in the US and evolve

companies such as Sequoia Capital, Y Combinator and Andreessen Horowitz. GlobalData

announced Sequoia Capital as the most successful VC investor in the AI space in 2020 by

having participated in 52 deals and investing over 400 million USD in 2020 alone (GlobalData

2021). The market is heavily disputed as market giants such as Meta Platforms, Amazon or

Alphabet Group have also increased their efforts by acquiring and strategically investing in

emerging companies with high technological standards. Google, nowadays Alphabet Group,

are the largest investors among the leading technology elite by having acquired over 30 AI start-

ups and having spent over 4 billion USD (Hurst 2020) on M&A activities since 2009. Among

their top investments, the acquisitions of DeepMind in 2014 for over 500 million USD (Shu

2014) and the acquisition of Onward. Objective of the investments was to elevate the quality of

the offered services by automating their processes and improving the customer experience on

the respective platforms. Facebook, nowadays Meta Platforms, have acquired AI based

companies such as AI.Reviere (Wiggers 2021), Bloosbury AI (for 23-23 million USD) (Ha

2018) and Scape Technologies (for 40 million USD) (O'Hear 2020) to improve their existing
NLP, Machine Learning and Virtual Recognition services (Shu 2014). Such investments by

large entities have proofed that certain know-how and human resources can only be obtained

by acquiring smaller markets players, driving companies values to higher dimensions.

The traditional bootstrapping, therefore, the financing of future operations with own capital

(Kenton 2020), is not a common practice among most AI start-ups, due to enormous costs

connected to human resources, hardware, and license fees for software. This creates the

opportunity for other type of investors such as Business Angels and Early-Stage investors,

which gained on popularity throughout the last years within the AI industry. Among the

successful Early-Stage investors Venture Capital companies such as Y Combinator or M12 can

be found, which focusses on funding tickets bellow 50 million USD (GlobalData 2021).

It is important to differentiate by the final purpose of the investment. The activities of larger

companies mostly tend to improve the already existing technology behind the own products,

whereas the investment activates of smaller and medium sized companies also aims to expand

the product portfolio.

On the one hand, this can be seen in large deals such as the acquisition of Nuance by Microsoft

in 2021 (Baker, Porter and Dina 2021). The company was acquired for almost 20 billion USD

to improve Microsoft conversational AI focused platform with its cutting-edge NLU (speech

and text recognition) and NLP technology. Not only the Tech Giants have shown interest in the

emerging industry, as for example players such as Panasonic acquired the supply chain-based

company Blue Yonder in 2020 for 7 billion USD (Blue Yonder 2021). The electronics provider

aims to “[…] aim to optimize the overall supply chain not only within single companies but

also across companies.” (Panasonic 2021) An additional deal enhancing the strategy of

acquiring external companies to improve the own product performance can be identified in the

Zoox acquisition by Amazon in 2020. The E-Commerce giant acquired the autonomous driving
system for 1.2 billion USD to accelerate own developments for autonomous delivery vehicles

with the objective of solving the last-mile issue and cutting mayor cost of delivery (About

Amazon 2020).

On the other hand, the diversification of the portfolio can be seen in a company such as Zebra

Technologies. The Illinois based designs, manufactures, and sells automatic identification and

data capture products and amplified its product portfolio by acquiring Antuit.ai in 2021

(SupplyChainBrain 2021).

With Antuit omni-channel approach, Zebras Technology will be able to offer the services

through new channels, increasing the value for its customers and their end-customers.

Moreover, Ipsos also present a similar approach based on the most recent acquisitions of the

companies Infotools (IPSOS 2021), Synthesio (IPSOS 2018) and Intrasonics (IPSOS 2021) for

over 60 million USD through the last three years. In results in an expansion of services by

acquiring players that focus on social media and audio, therefore implementing new sources of

data to complement the conventional approach of the French market research company. An

additional example consists in the MarTech (Marketing Technologies) leader Hootsuite. The

Canadian company offers a unified solution of Social Media and Marketing Management,

unifying different services of the industry. With the recent acquisition of HeyDay!, a

conversational AI provider, for 60 million USD, Hootsuite plans to expand its operation into

also helping its client automize its communication towards end client by using state of the art

Natural Language Processing and Natural Language Understanding systems (Hootsuite 2021).

5.4 Investment per Industries

Based on the high penetration of the adaption of the digitalization by the beginning of the 20 th

century, it can be assumed that majority of the nowadays known industries are able to

implement AI approaches. The industries differentiate by the degree of implementation of AI

approaches, therefore also defining the scope and value of the companies with AI based
solutions. It is important to state that the AI industry it’s an industry itself, yet the usability of

it is always connected with other industries.

Based on OECD research of “Venture Capital Investments in Artificial Intelligence”, industries

indicate strong differences in terms of ticket size and popularity during the timeframe of 2012

and 2020. In quantitively numbers, the industries of IT infrastructure and hosting (2012-2020:

4063 deals = 19.8% of all deals), Media/Social Platforms/Marketing (2012-2020: 3351 deals =

16.3% of all deals), Business processes and support services (2012-2020: 2944 deals = 14.3%

of all deals) and Healthcare, drugs, and biotechnology (2012-2020: 2545 deals = 12.4% of all

deals) are the largest industries up to today (Tricot, OECD 2021). When comparing with the

aggregated total value of all investments between the timeframe of 2012 and 2020, changes can

be identified. The industry with the highest investment is the “Mobility and Autonomous

Vehicles” industry, accounting for over 29% of the monetary value of the investments since

2012, yet slightly decreasing during the last years. Surprisingly none of the before mentioned

industries are equalling the relative amounts of deals with the relative monetary value of the

investments: IT infrastructure and hosting = 10% and steady performance of the years,

Media/Social Platforms/Marketing = 11%, decreasing performance over the years and even

reaching 6% in 2020, Business processes and support = 11% and steady performance of the

years and Healthcare, drugs, and biotechnology = 10% and increasing performance of the years

(Tricot, OECD 2021).

High investment in the Mobility and Autonomous Vehicles are connected to the factor of high

cash burn and low margins. The large AI-based transportation services such as Uber, Bold, Lyft

and Didi received additional mayor cash injections to compensate the extensive cash burn

during the initial phase for their operations, technology assets and marketing spending’s

(Lehtonen 2021). In addition, Googles’ autonomous car manufacture Waymo had an impact by
raising almost 2.5 billion US in a second external investment round in 2021 (Alamalhodaei

2021).

5.5 Investment by Geography

When analysing the source of the investment related activities a very homogenous distribution

among different regions can be identified. Mayor origin of deals is connected to the US and

China, as they account for almost 72% of all closed deals between 2012 and 2020, accounting

for almost 80% of the monetary value of the investments (Tricot, OECD 2021).

When comparing the US to China, US takes the clear role as the more active investor as they

account for 174 billion USD in investments during the period of 2012 and 2020, accounting for

more than 50% of all VC investments over this period. Throughout the years, China, and other

countries such as UK, EU27, Japan and Israel started to increase the amount of investment, yet

the American VCs still account 43% of all total investments in 2020 (Tricot, OECD 2021).

EU27, mainly due to the investments from German and French VCs, performance throughout

the years had a positive performance, as all countries increased the aggregated amount of

invested money to a total of 7 billion USD and a total participation in 800 deals in 2020 (Tricto

2021).

It is important to differentiate between the factor of origin of the investing VC and the origin

of the to be invested company. When comparing the activities on national terrain, China

represents 70% of the investments in local firms, whereas the US only accounts for 60%. Mayor

differences can be identified when it comes to investment outside the own country, as the US

based VCs account for almost 20% - 24% of the globally conducted investments, excluding

themselves and China. In comparison to this, China, accounts for only 5% of the globally

conducted investments, excluding them itself and the US. This big difference indicates that the

efforts from China are mainly focused on investing locally. Moreover, the investment landscape
in China changed due to foundation of government-led incubators and the raise of strong

Chinese technology firms such as Alibaba, JD or Baidu (Tricot, OECD 2021).

5.6 Investment Trends in the AI Industry

The heavily increasing investments throughout the whole world strongly indicate a prosperous

future for the industry. Yet, as presented in 1.3 and 1.4, certain industries have experienced a

decrease in demand, influenced by the geolocation or the change in customer requirements.

Assessing future trends in the AI industry can be approached from different perspectives, as it

can be analysed from an industry point of view or on a more technological point of view.

When performing an analyse of the industry point of view, industries such as the Mobility and

Autonomous Vehicles will continue growing due to the raising demands for cars working based

on renewable energies and the constantly increasing fuel prices caused by limited natural

resources.

When performing an analyse of the technological point of view, three mayor trends will be the

main challenge according to the Yang Lu from Antai College of Economics and Management,

Shanghai Jiao Tong University: development of platforms, algorithms, and interfaces (Lu

2019). According to Professor Lu, future developments should focus on creating platforms that

can perform at a higher level, therefore processing larger amount of data in a shorter time. Such

requirement is closely connected with Hardware AI providers such as NVIDIA, Intel or Google

that are already working on next generation (GPU = Graphics Processing Unit instead of CPU

= Central Processing Unit) devices to fulfil those demands. In addition to this, platforms shall

develop own approaches to combat increasing to prevent malicious processes and threads,

making them event more secure against Cybersecurity related issues (Doshi-Velez and Kim

2017). Yang Lu indicates that the future development of the algorithms should aim changing

from an “artificial intelligence” towards an “humanoid intelligence”, therefore preparing

algorithms to adapt to changing circumstances in the social world and combine it with the

material world (Lu 2019). Last but not least, the development in regards of the interface should

combine the factor to a very elaborated and professional back end with an user friendly front-

end to prevent any kind of usage problems on the platform and therefore also decrease

unnecessary expenses on Customer Support.

6 Artificial Intelligence in Banking

6.1 Opportunity for Banks

Today, banks are not questioning whether they should focus and invest in AI techniques and

algorithms or not – this decision was taken long ago, but the adoption is only starting.

According to a survey from Deloitte that wishes to analyse the importance of AI in the success

of the organizations in the next two years, 86% of the respondents said that AI will be very or

critically important (Deloitte US 2021). However, it is important to understand what AI is and

how it might impact businesses instead of just follow the hype around the topic – there is the

need for a strategic guide. While some banks still think that AI is useful to make experiments

on some products and processes, others are already taking these experiments into large scale

and powering their businesses through intelligent tools that automate back-office tasks such as

Know-Your-Customer (KYC) procedures or credit scoring decisions, but also communication

channels like chatbots.

“The world’s most valuable resource is no longer oil, but data” (The Economist 2017), and that

is the greatest opportunity for banks. Financial institutions have access to the purchasing records

from each of its clients, and when analysed these can give powerful insights on default and

credit risk, cross-selling (e.g., propose an insurance product after the purchasing of a car), and

many other applications that will be later explored.

First, this section has an initial part that analyses how a bank is organized and how the value

chain is composed, and then it will analyse the rising importance of banking data because of

increasing internet and mobile applications adoption, and a snapshot of how high this will be

was already given here. The second part of this section will be built on the work done in the

first one, plus the results of qualitative interviews answered by experts in financial services in

the Portuguese landscape, and a quantitative survey to understand the perspectives of the clients

of a bank regarding the communication channels they use, and their need of going to a physical

branch. In that second block, this work will present what are the current applications of AI in

the financial services sector and how banks are already leveraging these techniques. Moreover,

it will define if the banking ecosystem is moving towards a scenario of cooperation or

competition between FinTech firms and banks, and what we can expect from the existence and

operation of physical branches. Specifically for this last topic, the main conclusions will be a

merge of the opinions of experts and the expectations of customers.

6.2 Research Methodology

This project is integrated in joint research on how AI is impacting digital businesses, focusing

on measuring, and understanding this impact specifically in the financial services sector.

The main objective is to answer the following questions:

(1) How are banks organized, which are the main products and processes, and how can data

help?

(2) What is the state of the art and the impact of Artificial Intelligence in the banking sector?

(3) Where is the banking sector going in terms of the relationship with FinTech firms and

the existence of physical branches?

To answer the proposed questions, this study will focus on a qualitative research methodology

through an interview made to experts and a quantitative survey made to the general population.
These interviews were performed to gain multiple insights on how banks are leveraging AI

techniques to scale their business, and that is why we chose to interview specialists in

management roles at financial institutions in Portugal. In appendix A it is present what are the

questions and the profiles from the experts interviewed.

To guide the experts through the questionnaire, the following topics were accessed: 1) is AI a

must for banks to deliver better and increase the customer base and retention; 2) what is the

state of the art of AI in banking right now, regarding the main applications and benefits and

risks; 3) what are the future trends shaping the future of AI in the banking industry; 4) will the

technological developments come from in-house or through cooperation with FinTech

companies; 5) what is the future of physical branches and their role in society.

The qualitative survey focused on understanding the customers’ expectations about digital

banking services, and whether they believe or not that most activities and product will be

transferred to the online sphere. To do so, the survey consisted in 178 answers from individuals

with a diversified demographic and academic background and tried to find: 1) will products

like loans move to the online or will they stay on the branches; 2) which are the preferred

communication channels; 3) will virtual assistants’ impact increase in the future. In appendix

B it is described how the survey was organized.

7 Banking and the generation of data

7.1 Understanding a Bank’s Organization

Before deep diving on the main applications, trends, and future of AI in the Banking industry,

it is important to understand the organization of a bank and its main components, as well as the

principal operations and revenue streams (i.e., products and services), to better see how AI can

impact these. After defining each area and the main products and services, it is possible to build

the full value chain of the bank.

According to (Deloitte US 2021), we can define three main blocks of action:

1. Front office: this area comprises all the operations and interactions that directly connect

with the final customer, through agents such as salespeople or branch managers (Deloitte

2020).

2. Products and services: in here are included all the products and services offered by banks

to its clients, such as checking and savings accounts, loans, mortgages, the distribution of

debit and credit cards, insurance, advisory, etc. This applies to both individual and

institutional clients.

3. Back office: this area is made by the administrative and support people that do not face

clients, with areas such as IT and Data & Analytics, accounting, regulatory reporting and

compliance, fraud and money laundering prevention, audit, and service optimization.

Given this and the experts’ input, we have the ingredients to build our banking value chain and

start exploring the applications of AI on each sector of it. Thus, we can define the value chain

by having a back-office, a middle-office that will do the connection between both, and a front-

office. In the back-office, we will have the IT, accounting, and the people and talent teams. In

the middle-office, we will have the operations and risk management teams. In the front-office,

we will have the marketing and sales, products and services, and customer relationship teams.

Figure 1: The banking value chain (based on Author)

This definition is important to then resume the main applications of AI within the chain, as well

as its trends and their impact in the ecosystem and the future of branches. It is also important to

understand this breakdown since although most of the areas will benefit from AI, not every
product and service or department will benefit from it in the same way. However, we can

conclude that by having so many products and services that connect directly to clients, this will

generate huge amounts of data that needs to be treated. In the next section, it is presented if the

data generated by banks is valuable or not.

7.2 The Value of Banking Data

First, there is the need to introduce the concept of Big Data. This is a concept associated with

three Vs: volume, variety, and velocity. Volume matters, since big data comes associated with

high volumes of unstructured and low-density data, and this might come from several sources

and not only the banking transactions, proving the second V. At last, data is generated quicker

and so there is the need to treat this in real time (Oracle 2021). The usage of mobile banking

applications skyrocketed in the last months, and the root of that might be the pandemic. The

market penetration for all generations of ages increased from December 2020 to May 2021. For

Generation Z, it went from 86% to 95%; for Millennials, from 83% to 91%; for Generation X,

from 73% to 85%; for Baby Boomers, from 42% to 60%; for Seniors, from 18% to 27% (Forbes

2021). The increase in the usage of mobile applications rose, meaning that all providers are now

collecting more data that needs to be treated in real time to become valuable.

As said before, if there is a type of data that really defines someone is the record of the purchases

done. Let’s imagine the following example: there is someone that loves tennis. Considering an

individual that has an account on a social network and follows all the tennis players in the world,

plus the pages that share content about tournaments and points. What will tennis brands target

when they do advertising? People like this individual. And these brands will spend money

targeting it, trying to get it to spend money on tennis products.

Now let’s go for the data available at the bank. This individual is 40 years old and never did a

purchase on a merchant related to tennis, never rent a court to play and never bought a ticket to
a tournament. This means that the marketing money allocated to this user will be useless, but

the bank has valuable data and needs to treat it.

It is estimated that 30% of the actual revenue streams of a bank disappear in the following 20

years due to the rise of neo and challenger banks, regulatory changes and pressure, and the

increased number of transactions and investments done in cryptocurrencies (Accenture 2018).

This is an alarm – banks should push hard to collect valuable insights from data and do not let

their customers slip away. Also in this report, there is evidence of a large player in the United

States that added 1.2 billion USD in annual revenue by partnering with merchants to perform

marketing campaigns powered by banking data insights to increase conversion in their sales.

The conversion rate of the bank rose to 3.5%, contrasting with the 0.4% average on the industry.

McKinsey found that the potential annual value of AI, traditional or advanced, to the banking

industry can sum up to 1 trillion USD, representing 15% of sales. The sectors and activities of

the bank that can benefit the most from this are related to marketing and sales and can have a

value up to 625 billion USD. Inside this department, the most relevant use cases where AI can

add value are related to customer service management, with optimized call centres that predict

the times with more demand or voice recognition to immediately have a specialist speaking to

that person to avoid churn, but there is also opportunity for pricing and promotion with the

application of algorithms to introduce personalized pricing and increase conversion. Risk

management can also benefit a lot from AI, with a potential annual value up to 372 billion USD,

through the usage of analytics to detect and prevent fraud and money laundering, and to better

preview potential defaults on loans. Then, also human resources, finance and IT, and other

operations can benefit from AI, but at a much lower scale (McKinsey 2018).

So yes, banking data is valuable, and has a lot to explore. What drives AI is the amount of data

available, and that is already happening in the banking sector. In the next section, we will deep
dive on the current applications of AI in the banking industry and how banks are currently

leveraging the huge amounts of data collected to scale their businesses.

8 The state of the art of Artificial Intelligence in Banking

Artificial Intelligence is already applicable in the whole value chain of the bank. However, only

in the last few years we have been seeing these techniques extend to the touchpoints of the bank

with its customers. The first developments in AI started in the back-office operations, with the

creation of algorithms and procedures to automate tasks such as credit scoring, fraud detection

and anti-money laundering schemes – the rise of Robotic Process Automation (RPA), that will

combine several AI techniques as we will see further. These algorithms can be useful for several

duties, such as mortgage processing, KYC procedures, report automation, and many more

examples, allowing to reduce costs, human errors and increase efficiency. That also means that

banks were already using data from the front-office to feed these algorithms but were not

treating it in a way that would benefit their relationships with their customers, mainly in

managing expectations and creating services that are tailored and personalized to the customers’

needs.

To automate tasks such as credit scoring and fraud detection, these algorithms need to analyse

millions of records to make better predictions. This is done through Machine Learning

algorithms. Also, banks are leveraging AI techniques to perform the exact same back-office

tasks that are repeated and time-consuming. This includes scanning documents and digital

signatures, reducing the paperwork, and is powered by Computer Vision techniques. Finally,

NLP is being used to power voice and chat assistants, to help customers to interact in an easier

way with the bank.

Resuming, and according to the experts’ inputs and the study performed, we can state that the

main AI techniques being used in the banking sector right now are Machine Learning,
Computer Vision and NLP. We will see which are the use cases mentioned for each technology

described.

8.1 Machine Learning

Predictive analytics and classification problems are powered by ML techniques. These are

helping the back-office tasks of a bank getting smoother and more accurate, with less human

intervention and giving results faster. If one were to analyse the transactions of a bank one-by-

one, doing fraudulent transactions would be easy. These techniques are helping not only to

better predict credit defaults and to monitor risk, but also to verify the real identity of a

customer.

Money laundering refers to the process from which criminals can disguise illegal transactions

as legal sources of property and income, transforming this income into legal money (Banco de

Portugal 2021). Therefore, banks need to focus on anti-money laundering (AML) mechanisms

to prevent fraud and terrorism financing. With the pandemic of COVID-19 people shifted more

of their transactions to the digital space, through e-commerce platforms and with digital means

(such as wallets or NFC payments), rising the number of actions related to AML – this number

was already rising, but the pandemic was a factor to deepen this problem. In e-commerce, for

instance, criminals see a good opportunity to pose as legal merchants and can easily collect

money from the less informed buyers (McKinsey 2019).

Also, ML is powering credit scoring to better predict defaults and decide whether a customer is

trustworthy or not. With the expansion of the banking business, credit card and personal loan

applications are also rising, and the decision needs to be done immediately to better serve

customers. Banks use the purchasing records of their customers to make instantaneous credit

decisions, and thus speed up this process. This is also enabling the approval of loans through

digital channels, without the need of going to a physical branch, which increases the
convenience of the service. With more data, algorithms get more accurate, and the risk of

default gets lower with time. Both use cases specified contribute heavily to better manage and

thus reduce risk.

Finally, authentication methods such as Face ID (that allows to do a login in a mobile banking

app) or fingerprint scanning are also using DL algorithms to make sure that no one can access

an account besides the real owner of it. iPhone users can log in to their accounts or make

payments using the Face ID system provided by the phone, that places thousands of infrared

dots on the face of the person that is trying to authenticate and compares it to the original picture

provided (Apple 2021).

8.2 Computer Vision

Computer Vision allows to analyse pictures and videos and turn them into valuable and verified

information due to the recent development of DL algorithms for financial services (Forbes

2019).

The first main advantage of this technology is the digital onboarding of clients, performed with

in-app KYC, eliminating the need to go to a branch to open an account. Customers are required

to make a video-call (in the case of Moey!, the digital bank created by Crédito Agrícola in

Portugal) – where they will speak with a real human, take a selfie and scan an identity card, all

without leaving the call (ECO 2019) – or to just upload the photos of the required documents

(in the case of Revolut) – a photo of the national identity card, plus a selfie, and the algorithm

does the match (Revolut 2019). These speeds up the process of getting a new client and

diminishes the risk of losing it during the onboarding process.

Another main application of this technology in the banking sector is a complement to the digital

KYC procedures, and is the digital scanning of documents, previously done manually by

humans. The documents, when in paper, are easy to get lost, to be damaged or to be mistreated,
and the technology is here to help reducing the human error and increase efficiency – by

automatically understand which is the type of document that is being scanned and what is the

information stored in it, like in Revolut’s onboarding. When opening a bank account online, the

user needs to enter its personal information but also needs to submit some documents, that

sometimes come in a picture format. Computer Vision algorithms, combines with NLP that will

be explained after, will take care of the situation, and make sure that all the information is

treated and stored.

8.3 Natural Language Processing

Picking where we were, the text present on the documents scanned during, for example, KYC

operations needs to be analysed and compliant with the other data provided in editable fields

such as the name or the date of birth – when the identity card is uploaded, that information

needs to match the one provided manually by the client. Also, legal teams are leveraging NLP

techniques to review large bulks of long documents – the algorithms are created to identify

what is marked as important and retrieve that information to the analysts. Finally, other client

data different from the transactions done by card or transfers that reaches the bank can be an

opportunity to explore cross-selling or identify the client’s needs – for example, the posts a

client makes about its bank on social media turns it easy to detect whether it is happy with the

service or not.

The other use case of NLP within the banking industry is associated with customer service, that

today is required on demand. Banks are using chatbots and virtual assistants to communicate

with their clients since these need their problems to be solved in real time. For instance, a client

can use a chatbot to perform basic banking operations such as check the account balance or

make a transfer, but also to solve technical issues and get information about products, services,

schedules, and so on. However, in the survey performed, 73% of the respondents said that they

would not use voice commands to perform basic operations such as checking balances or make
transfers. But the perspectives are good, since the share that says they would do it in 10 years

from now is 63%, showing that the trend exists. Also, 71% said that chatbots are indifferent,

mostly irrelevant, or irrelevant when asked to classify that as a feature of a bank.

This is where we are today (and will continue to be). In the next section, we will see if these

technological developments will come from in-house or from third party providers.

9 FinTechs: Friend or Foe

FinTechs such as challenger banks that operate in a Business-to-Consumer (B2C) model cannot

be seen as an enemy of banks right now since they do not have the same business model. A

bank captures deposits to transform these into loans or investments in securities, and that is the

main activity – the interest rate spread gives a profit. Then, of course, they have several other

revenue streams such as commissions charged on account and card maintenance, commissions

on trading accounts, insurance products or investment products – and usually, challenger banks

do not charge for any of these. If we look at the business model of a challenger bank like N26

or Revolut, we can see that they do not offer the same loan products that incumbents offer, and

thus cannot be seen as a big rival in this segment. For instance, Revolut has some of its clients’

deposits safe since they are deposited in other big banks (Revolut 2021), showing that even

these are depending on the incumbents to survive right now. Also, looking at this breakdown

of the incumbents’ revenues and checking that challenger banks do not have these as revenue

streams and are incurring in huge operational losses – Revolut had $280 million in 2020

(Bloomberg 2021) – we are allowed to ask if these business models are good after all.

However, when we talk about FinTechs that operate with a Business-to-Business (B2B) model,

the panorama changes. These are firms that are offering their services to banks, in a clear

statement that they exist to cooperate and there is work for everyone. If we look at the

Portuguese banking landscape, we can see that: Caixa Geral de Depósitos is using Tink’s

account aggregation and payment initiation engines to provide an open banking experience to
their clients (Tink 2019); Crédito Agrícola invested and partnered with Meniga to power their

digital transformation experience and launch Moey!, a fully-digital bank (Echo Boomer 2019);

Feedzai, a Portuguese unicorn valued at 1.5bn USD (ECO 2021) established in 2011 that now

is the market leader in fighting criminal transactions by providing software to financial

institutions, is helping banks such as Citi, Lloyds Bank, ABN AMRO and Santander to better

monitor risk and prevent fraud.

Given this, all the experts said that, although some developments might come from in-house,

most of the back-office and routine tasks will be automated through partnerships with FinTechs.

Around 46% of banks are planning to create more partnerships with FinTechs in 2022, while

this value was 32% in 2020. Also, in the first half of 2021, the United Kingdom’s FinTechs

received more than £4.1 billion, while the total investment in 2019 was £3.3 billion (Lloyds

Bank 2021). Analysing the experts’ opinion and the facts above, we can conclude that FinTechs

are here to help maturing solutions, save costs, get more efficient processes, and allow banks

to scale their businesses. Therefore, they will stay around and cooperate even more with banks

in the next years.

All these technologies and developments are impacting the workplace, as well. While

algorithms automate back-office and routine procedures, it might be the case that the job at a

physical office or branch also changes. In the next section, we will discuss the future role of

physical branches and how these are being impacted by digitalization.

10 The Future of Physical Branches

Right now, physical branches are here to say, although the absolute number of locations per

million people is dropping in North America, the United Kingdom, and Europe, showing the

evidence that these were too much and representing unnecessary costs (McKinsey 2019). These

will not disappear in the short run – the ones that keep operating will serve as strategic locations

to serve a higher number of clients. With the empowerment of AI in the industry, customers are
also embracing a more digital culture as we could see in chapter 2.2., meaning that most of the

operations that were previously done in the branch are already done online – and the back-office

operations performed at the branch are being substituted or at least helped by AI. However, the

landscape changes when we think about the future – specifically 10 years from now.

In the survey performed to study the customers’ expectations about banks, there is evidence of

their will to still go to a physical branch to take care of a specific product – loans. In the

investigation performed, today the physical branch is the preferred channel to buy a loan,

whatever the category (personal, student, home, car, and business). 72% would use a physical

branch to purchase a home loan, and the same percentage applies to business loans. Lower

dimension loans such as the personal or the student ones have lower percentages of customers

desiring to have them in the branch – 45% and 43% respectively.

That will change in the future. 10 years from now, the app would be the preferred channel to

buy any type of loan, while the physical branch is the least preferred for every category except

for the home and business loans (this was expected after analysing today’s results). In the future,

84% would use digital channels to buy a personal loan (the website or the app), being the app

the most preferred channel. 73% would use digital channels to buy a home loan, completely

reverting the trend found today. More detailed results can be found in the table in appendix B.2.

Concluding, we can expect the number of branches to keep decreasing in the next years, but

these will not disappear. Most services will be transferred to the online sphere and these

locations will be transformed in customer service centres not so focused on sales, but the human

touch still needs to be there in the long run and will be for sure.

11 Conclusion
After all the contents analysed there is a solid ground to answer the research questions proposed

in the beginning of this paper. The first question was linked to the bank’s organization and the
value of data. We can organize the value chain in three major blocks: 1) the back-office,

composed by the IT, accounting, and people and talent departments; 2) the middle-office,

composed by the risk management department that deals with situation such as fraud detection,

and the operations team; 3) the front-office that deals directly with the final customer, and is

composed by the sales and marketing teams, the products and services offer, and the customer

relationship managers. All these products and services commercialized by a bank will generate

huge amounts of data that are serving as fuel to the back and middle office engines. Data is,

indeed, of great value since it allows to characterize customers, group them into clusters,

classify their default probability and reduce the default rate, cross-sell products, and many other

examples – that will be developed principally for front-office activities.

The second question aimed at finding which are the principal applications of AI in the banking

activity and their impact. Here, we found that the major technologies are ML, Computer Vision

and NLP. These allow to develop solutions concerning fraud detection and anti-money

laundering, authentication methods, KYC, scanning and finding information on documents, and

virtual assistance to the customer. A great benefit to the bank comes associated with a lower

default rate, thus increasing the revenue. Also, this speeds processes such as an account opening

or document scanning, saving costs and working faster. For the final customer, products and

services are available on-demand and in real time and coming at lower costs. However, there

might be the case of bias in the ML algorithms, for instance, and that is a possible limitation

from these implementations. The sum up, the biggest impact right now is coming from the back

and middle-office tools that exist to save on basic procedures and to better manage risk. The

great potential of data will drive marketing and sales in the future, with a more accurate client

targeting and a better strategy to increase retention.

Finally, physical branches are going to transform themselves into customer service centres,

located strategically, to decrease the number of branches spread through a country. Human
touch will still exist, but through less physical places, and this shift to the online sphere allows

to save on people and infrastructure costs and increase productivity.

Bibliography
2020. About Amazon. June 26. https://www.aboutamazon.com/news/company-news/were-
acquiring-zoox-to-help-bring-their-vision-of-autonomous-ride-hailing-to-reality.

Accenture. 2018. Bank Data: Untapped high return asset or junk bond? Accessed November 10, 2021.
https://www.accenture.com/_acnmedia/pdf-74/accenture-bank-data-asset-value-or-junk-
bond.pdf.

Advani, Vaishali. 2021. What is Artificial Intelligence? How does AI work, Types and Future of it?
October 19. https://www.mygreatlearning.com/blog/what-is-artificial-intelligence/.

Alamalhodaei, Aria. 2021. TechCrunch. June 16. https://techcrunch.com/2021/06/16/waymo-

alphabets-self-driving-arm-raises-2-5b-in-second-external-investment-round/.

Apple. 2021. Acerca da tecnologia avançada do Face ID. Accessed November 20, 2021.
https://support.apple.com/pt-pt/HT208108.

Apple Machine Learning Research. 2017. Hey Siri: An On-device DNN-powered Voice Trigger for
Apple’s Personal Assistant. October. Accessed November 1, 2021.
https://machinelearning.apple.com/research/hey-siri.

Arnold, Brenda. 2021. Everyone’s talking about AI. But what do they think about it? September 28.
https://innovationorigins.com/en/everyones-talking-about-ai-but-what-do-they-think-about-
it/.

Atkins, Scott, and Kai Luck. 2021. AI for Banks - Key Ethical and Security Risks. September 9. Accessed
November 7, 2021. https://www.law.ox.ac.uk/business-law-blog/blog/2021/09/ai-banks-
key-ethical-and-security-risks.

Baker, Liana, Kiel Porter, and Bass Dina. 2021. Bloomberg. April 11.
https://www.bloomberg.com/news/articles/2021-04-11/microsoft-is-said-to-be-in-talks-to-
buy-nuance-communications.

Banco de Portugal. 2021. Money laundering and terrorist financing. Accessed November 11, 2021.
https://www.bportugal.pt/en/page/branqueamento-de-capitais-e-financiamento-do-
terrorismo.

BBC. 2020. Covid-19 vaccine: First person receives Pfizer jab in UK. December 8. Accessed November
8, 2021. https://www.bbc.com/news/uk-55227325.

Bloomberg. 2021. Revolut losses hit 280 million in 2020 with staff costs surging. June 21. Accessed
November 28, 2021. https://www.bloomberg.com/news/articles/2021-06-21/revolut-losses-
hit-280-million-in-2020-with-staff-costs-surging.

2021. Blue Yonder. September 17. https://media.blueyonder.com/panasonic-completes-acquisition-

of-blue-yonder/.

Britannica. 2021. artificial intelligence. November 9.

https://www.britannica.com/technology/artificial-intelligence.
CHI Software Developmet. 2021. Artificial Intelligence in the Retail Industry: Improving Shopping
Experience. September 1. https://chisw.com/ai-in-retail/.

Columbus, Louis. 2018. Forbes. January 12.

https://www.forbes.com/sites/louiscolumbus/2018/01/12/10-charts-that-will-change-your-
perspective-on-artificial-intelligences-growth/?sh=630350754758.

—. 2018. Forrbes. January 12. 10 Charts That Will Change Your Perspective On Artificial Intelligence's
Growth.

Davis et al. 2014. The Visual Microphone: Passive Recovery of Sound from Video. Accessed October
30, 2021. https://people.csail.mit.edu/mrub/papers/VisualMic_SIGGRAPH2014.pdf.

Deb Marotta. 2020. Artificial Intelligence: How AI Is Changing Retail . https://global.hitachi-

solutions.com/blog/ai-in-retail.

Deloitte. 2018. AI and risk management: Innovating with confidence. Accessed November 6, 2021.
https://www2.deloitte.com/content/dam/Deloitte/uk/Documents/financial-
services/deloitte-uk-ai-and-risk-management.pdf.

—. 2019. Deloitte Survey on AI Adoption in Manufacturing .

https://www2.deloitte.com/cn/en/pages/consumer-industrial-products/articles/ai-
manufacturing-application-survey.html.

—. 2020. Impact of COVID-19 on front-office oversight at financial institutions. Accessed November

16, 2021. https://www2.deloitte.com/content/dam/Deloitte/us/Documents/finance/us-
impact-of-covid-on-front-office-oversight-final.pdf.

Deloitte US. 2021. Artificial Intelligence: Transform the future of banking. Accessed November 3,
2021. https://www.economist.com/leaders/2017/05/06/the-worlds-most-valuable-resource-
is-no-longer-oil-but-data.

Doshi-Velez, Finale, and Been Kim. 2017. "Towards A Rigorous Science of Interpretable Machine
Learning." Februar 28.

Echo Boomer. 2019. Moey!. Já chegou o primeiro banco 100% mobile. Accessed November 24, 2021.
https://echoboomer.pt/moey-ja-chegou-o-primeiro-banco-100-mobile/.

ECO. 2019. Há um “novo banco” digital. Saiba como funciona o moey! September 11. Accessed
November 20, 2021. https://eco.sapo.pt/2019/09/11/ha-um-novo-banco-digital-saiba-como-
funciona-o-moey/.

—. 2021. Unicórnios com ADN português já valem mais de 34 mil milhões de euros. November 23.
Accessed November 29, 2021. https://eco.sapo.pt/2021/11/23/unicornios-com-adn-
portugues-ja-valem-mais-de-34-mil-milhoes-de-euros/.

Forbes. 2019. Customer Loyalty and Retention Are In Decline. October 13. Accessed November 8,
2021. https://www.forbes.com/sites/shephyken/2019/10/13/customer-loyalty-and-
retention-are-in-decline/?sh=3e3d8de44329.

—. 2019. How Machine Vision Can Transform Financial Services. September 9. Accessed November
19, 2021. https://www.forbes.com/sites/cognitiveworld/2019/09/09/how-machine-vision-
can-transform-financial-services/?sh=2bdd4bad1e88.
—. 2021. Mobile Banking Adoption In The United States Has Skyrocketed (But So Have Fraud
Concerns). Accessed 11 16, 2021.
https://www.forbes.com/sites/ronshevlin/2021/07/29/mobile-banking-adoption-has-
skyrocketed-but-so-have-fraud-concerns-what-can-banks-do/?sh=6fddf3855dc6.

—. 2020. Seven Ways Artificial Intelligence Is Disrupting The Retail Industry. August 21.
https://www.forbes.com/sites/forbestechcouncil/2020/08/21/seven-ways-artificial-
intelligence-is-disrupting-the-retail-industry/?sh=7e00cc9b56ae.

Future Today Institute. 2021. 2021 Tech Trends Report: Volume 1. Accessed October 10, 2021.
https://www.dropbox.com/s/pc5g3dww1ckhbjb/FTI_2021_Tech_Trends_Volume_1_AI.pdf?
dl=0.

Global Trade. 2020. AI is Transforming the Manufacturing Industry: Pros and Cons. October 1.
https://www.globaltrademag.com/ai-is-transforming-the-manufacturing-industry-pros-and-
cons/.

2021. GlobalData. March 18. https://www.globaldata.com/sequoia-capital-top-vc-investor-ai-space-

2020-says-globaldata/.

Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. 2016. Deep Learning. MIT Press.
https://www.deeplearningbook.org/.

Ha, Anthony. 2018. TechCrunch. July 3. https://techcrunch.com/2018/07/03/facebook-confirms-

bloomsbury-ai-acquisition/.

Harvard Business Review . 2021. Everyone in Your Organization Needs to Understand AI Ethics. July
26. https://hbr.org/2021/07/everyone-in-your-organization-needs-to-understand-ai-ethics.

Healthcare Data Institute. 2015. Healthcare Data Institute. February 18. Accessed November 27,
2021. https://healthcaredatainstitute.com/2015/02/18/big-unstructured-datas-contribution-
to-healthcare/.

2021. Hootsuite. August 3. https://www.hootsuite.com/newsroom/press-releases/hootsuite-

acquires-conversational-ai-leader-heyday-for-ca-60-million.

Hurst, Aaron. 2020. Information Age. February 18. https://www.information-age.com/google-

revealed-acquired-most-ai-startups-since-2009-123487752/.

IBM. 2021. How AI is changing advertising . March 23. https://www.ibm.com/watson-

advertising/thought-leadership/how-ai-is-changing-advertising.

—. 2020. Natural Language Processing (NLP). July 2. Accessed October 28, 2021.
https://www.ibm.com/cloud/learn/natural-language-processing.

Indeed. 2018. Demand for AI Talent on the Rise. March 1.

https://www.hiringlab.org/2018/03/01/demand-ai-talent-rise/.

2021. IPSOS. September 15. https://www.ipsos.com/en/ipsos-acquires-infotools-and-further-

upscales-digital-capabilities-fuel-growth-0.

2018. IPSOS. October 30. https://www.ipsos.com/en-nz/ipsos-reinforces-its-social-media-

intelligence-capabilities-acquisition-synthesio.

2021. IPSOS. April 30. https://www.ipsos.com/en-in/ipsos-acquires-intrasonics.

Jackson, Peter. 1998. Introduction to Expert Systems. Addison-Wesley Longman Publishing.

Jurafsky, Daniel, and James Martin. 2000. Speech and Language Processing: An Introduction to
Natural Language Processing, Computational Linguistics, and Speech Recognition. Prentice
Hall.

Kark, Khalid, Jagjeet Gill, and Tim Smith. 2021. Deloitte. February 03.
https://www2.deloitte.com/xe/en/insights/focus/cio-insider-business-insights/impact-covid-
19-technology-investments-budgets-spending.html.

Kenton, Will. 2020. Investopedia. November 07.

https://www.investopedia.com/terms/b/bootstrapping.asp#:~:text=Bootstrapping%20descri
bes%20a%20situation%20in,revenues%20of%20the%20new%20company.

Kenyon, Tilly. 2021. AI Magazine. August 09. https://aimagazine.com/ai-applications/ai-spending-

will-reach-usdollar342bn-2021-says-idc.

Lehtonen, Scott. 2021. Investor's Business Daily. October 14.

https://www.investors.com/research/uber-stock-buy-now/.

Liu, Shanhong. 2020. Statista. July. https://www.statista.com/statistics/1133024/impact-of-covid-19-

on-ai-investments-by-industry/.

Lloyds Bank. 2021. Financial Institutions Sentiment Survey 2021. Accessed December 1, 2021.
https://lloyds.turtl.co/story/financial-institutions-sentiment-survey-
2021/page/1?utm_source=FISS-landing-
page&utm_medium=Website&utm_campaign=FISS21&utm_content=copy-download.

Lu, Yang. 2019. Taylor Francis Online. Februar 06.

https://www.tandfonline.com/doi/full/10.1080/23270012.2019.1570365.

Machine Learning Mastery. 2019. A Gentle Introduction to Computer Vision. March 19. Accessed
October 29, 2021. https://machinelearningmastery.com/what-is-computer-vision/.

McCarthy, John. 2007. WHAT IS ARTIFICIAL INTELLIGENCE? Stanford University.

McKinsey. 2020. AI-bank of the future: Can banks meet the AI challenge? September 19. Accessed
November 7, 2021. https://www.mckinsey.com/industries/financial-services/our-insights/ai-
bank-of-the-future-can-banks-meet-the-ai-challenge.

—. 2017. Manufacturing: Analytics unleashes productivity and profitability. August 14.

https://www.mckinsey.com/business-functions/operations/our-insights/manufacturing-
analytics-unleashes-productivity-and-profitability.

—. 2019. Rewriting the rules: succeeding in the new retail banking landscape. February. Accessed
December 4, 2021.
https://www.mckinsey.com/~/media/mckinsey/industries/financial%20services/our%20insig
hts/rewriting%20the%20rules%20in%20retail%20banking/rewriting-the-rules-succeeding-in-
the-new-retail-banking-landscape.ashx.

—. 2018. The executive's AI playbook. Accessed November 14, 2021.

https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/the-
executives-ai-playbook.
—. 2019. Why anti-money laundering should be a top priority for financial institutions. Accessed
November 15, 2021.
https://www.mckinsey.com/~/media/mckinsey/industries/financial%20services/banking%20
blog/why%20aml%20should%20be%20a%20top%20priority%20for%20financial%20institutio
ns/why-aml-should-be-a-top-priority-for-financial-institutions.pdf.

Medium. 2020. The 5 Primary Areas of Robotics. July. Accessed October 29, 2021.
https://rancholabs.medium.com/the-5-primary-areas-of-robotics-6aff30f2cf39.

Minsky, Marvin. 1960. Steps towards Artificial Intelligence. October 24. Accessed October 20, 2021.
https://web.media.mit.edu/~minsky/papers/steps.html.

National Retail Federation. 2021. AI’s role in retail . September 30. https://nrf.com/blog/artificial-
intelligence-in-retail.

Needham, Mass. 2021. IDC. August 04. https://www.idc.com/getdoc.jsp?containerId=prUS48127321.

Nilsson, Nils. 1998. Introduction to Machine Learning. Stanford: Stanford University.

—. 2009. The Quest for Artificial Intelligence. Cambridge: Cambridge University Press.

O'Hear, Steve. 2020. TechCrunch. Februar 8. https://techcrunch.com/2020/02/08/scapebook/.

Oracle. 2021. What is Big Data? Accessed November 11, 2021. https://www.oracle.com/pt/big-
data/what-is-big-data/.

O'Reilly. 2021. AI Adoption in the Enterprise 2021. April 19. https://www.oreilly.com/radar/ai-

adoption-in-the-enterprise-2021/.

Oxforf Internet Institute. 2020. "Global Attitudes Towards AI, Machine Learning & Automated
Decision Making." https://oxcaigg.oii.ox.ac.uk/wp-
content/uploads/sites/124/2020/10/GlobalAttitudesTowardsAIMachineLearning2020.pdf.

2021. Panasonic. September 17.

https://www.panasonic.com/global/corporate/cns/gpi_blueyonder.html.

Pega. 2019. "What Consumers Really Think About AI: A Global Study."

Punjani, Ali, and Pieter Abbeel. 2016. Deep Learning Helicopter Dynamics Model. December 16.
Accessed November 1, 2021. http://www.cs.toronto.edu/~alipunjani/pdf/icra15-helimodel-
cameraready.pdf.

PWC. 2020. AI Predictions 2021. October . https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-

predictions.html.

PwC. 2020. Artificial Intelligence for reporting: Proof-of-Concept to explore use of AI to improve
reporting speed and consistency. Accessed November 7, 2021.
https://www.pwc.com/sg/en/consulting/assets/artificial-intelligence-for-reporting.pdf.

—. 2017. What doctor? Why AI and robotics will define New Health. June. Accessed November 7,
2021. https://www.pwc.com/gx/en/industries/healthcare/publications/ai-robotics-new-
health/ai-robotics-new-health.pdf.

Revolut. 2021. O meu dinheiro está seguro? Accessed November 20, 2021.
https://www.revolut.com/en-PT/help/profile-plan/security-personal-data/is-my-money-safe.
—. 2019. What is KYC and why do we need it. Accessed November 21, 2021.
https://blog.revolut.com/what-is-kyc/.

Rowse. 2019. 7 Benefit of AI in Manufacturing. January 24. https://www.rowse.co.uk/blog/post/7-

manufacturing-ai-benefits.

Russell, Stuart, and Peter Norvig. 1995. Artificial Intelligence - A Modern Approach. Prentice Hall.

Salesforce. 2021. Rise of the Chatbots: How AI Changed Customer Service.

https://www.salesforce.com/products/service-cloud/best-practices/how-ai-changed-
customer-service/.

Shu, Catherine. 2014. TechCrunch. January 14. https://techcrunch.com/2014/01/26/google-

deepmind/.

Stamford, Conn. 2020. Gartner. October 01. https://www.gartner.com/en/newsroom/press-

releases/2020-10-01-gartner-survey-revels-66-percent-of-orgnizations-increased-or-did-not-
change-ai-investments-since-the-onset-of-covid-19.

Standford. 2016. Defining AI. https://ai100.stanford.edu/2016-report/section-i-what-artificial-

intelligence/defining-ai#_ftnref1.

Statista. 2021. "In-depth Report: Artificial Intelligence 2021 ."

Stefanini Group. 2020. The 5 Advantages AI Can Offer Industry 4.0. October 2.
https://stefanini.com/en/trends/news/the-5-advantages-ai-can-offer-industry-4-0.

Sunarti et al. 2020. Artificial intelligence in healthcare: opportunities and risk for future. August 24.
Accessed November 8, 2021.
https://www.sciencedirect.com/science/article/pii/S0213911120302788.

2021. SupplyChainBrain. September 2. https://www.supplychainbrain.com/articles/33691-zebra-

technologies-to-acquire-antuitai.

Szeliski, Richard. 2021. Computer Vision: Algorithms and Applications. September 30. Accessed
October 30, 2021. https://szeliski.org/Book/.

The AI Journal. 2020. AI in a post-COVID-19 world. August. https://aijourn.com/report/ai-in-a-post-

covid-19-world/.

The Economist. 2017. The world’s most valuable resource is no longer oil, but data. May 6. Accessed
November 3, 2021. https://www.economist.com/leaders/2017/05/06/the-worlds-most-
valuable-resource-is-no-longer-oil-but-data.

—. 2020. What companies really think about AI. January 1.

https://applied.economist.com/articles/what-companies-really-think-about-ai.

Time. 2020. Millions of Americans Have Lost Jobs in the Pandemic—And Robots and AI Are Replacing
Them Faster Than Ever. August 6. https://time.com/5876604/machines-jobs-coronavirus/.

Tink. 2019. How CGD built a cutting-edge PFM app in 6 months. Accessed November 22, 2021.
https://tink.com/blog/digital-banking-solutions/caixa-cgd-dabox-app/.

—. 2021. Tink's guide to Improving digital banking through personalised insights. Accessed November
2, 2021.
https://resources.tink.com/hubfs/05%20Resources/Guides/Money%20manager%20campaig
 n%202021/Improving%20digital%20banking%20through%20personalised%20insights%20-
%20Tink%20guide.pdf?_gl=1.

Tricot, Roland. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-

technology/venture-capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

—. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-technology/venture-

capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

Tricto, Roland. 2021. OECD. September 30. https://www.oecd-ilibrary.org/science-and-

technology/venture-capital-investments-in-artificial-intelligence_f97beae7-
en;jsessionid=cASduqeohzs3I4drpyPz7qIH.ip-10-240-5-8.

Vasishtha, Dhruv. 2018. How AiCure is bringing computer vision to health care: an interview with CEO
Adam Hanina. March 5. Accessed November 8, 2021.
https://medium.com/@dvasishtha/how-aicure-is-bringing-computer-vision-to-health-care-
an-interview-with-ceo-adam-hanina-ae92484eb614.

Wiggers, Kyle. 2021. VentureBeat. October 11. https://venturebeat.com/2021/10/11/facebook-

quietly-acquires-synthetic-data-startup-ai-reverie/.

Wired. 2016. This AI software can tell if you're at risk from cancer before symptoms appear2016.
August 26. Accessed November 27, 2021. https://www.wired.co.uk/article/cancer-risk-ai-
mammograms.
World Economic Forum . 2019. These are the biggest global risks . January 16.
https://www.weforum.org/agenda/2019/01/biggest-global-risks-facing-our-world/.

World Economic Forum. 2020. Here's how opinions on the impact of artificial intelligence differ
around the world. December 18. https://www.weforum.org/agenda/2020/12/mixed-views-
of-the-impact-of-artificial-intelligence.

—. 2019. This is what the world's CEOs really think of AI . June 25.
https://www.weforum.org/agenda/2019/06/ai-will-shape-our-future-how-do-we-ensure-it-
is-used-responsibly/.

Zhang, Baobao, and Allan Dafoe. 2019. “Artificial Intelligence: American Attitudes and Trends.”

Appendices

Appendix A – Interviews to Experts

Appendix A.1. – Questions

1) Do you think that AI, today, is a must for banks and there is a race to become AI driven and

get in the front row to better serve customers and gain new ones, but also to create more

efficiency in procedures and operations?

2) What is the state of AI in the banking industry right now? Which are the sectors and

activities already leveraging AI techniques, the use cases and the main benefits and risks?

3) Will these developments come from in-house, or will they rely on third party providers? Is

this a crucial time to include FinTechs in the financial ecosystem?

4) What do you think will happen to physical branches in the near future? And in the long run?

5) What can we expect for the near future (i.e., main trends)? Which sectors and activities are

starting or will be benefiting even more from AI, and which are the use cases?

Appendix A.2. – Profile of the experts

Expert 1 is a C-Level executive at Santander Portugal, working in this institution for almost 20

years. He was already part of the executive committee and the leadership of the insurance
branch. He holds a degree in management and has studied across several schools such as

INSEAD and Harvard Business School.

Expert 2 is the Chairman at Banco CTT, but also an advisor for Morgan Stanley, visiting

professor at Nova SBE and a research associate at the London School of Economics. He holds

a PhD in Economics from the University of Chicago and previously worked in companies like

Oliver Wyman and the Portuguese Treasury and Debt Management Agency.

Expert 3 is a Director at Banco de Investimento Global. He is responsible for the Digital

Strategy and Development department at the bank and is responsible for the Venture Capital

side. He is also a visiting professor at Católica SBE and the co-founder of Portugal FinTech.

Appendix B – Survey

Appendix B.1. – Original Structure of the Survey

Artificial Intelligence (AI) is changing the way products and services interact with their final

customers. Banks today have a huge challenge in complying with all the technological advance

in course and have to move faster than ever.

This research is being conducted by a Nova School Business & Economics student, to complete

a thesis on "Artificial Intelligence in Digital Business", and it aims to find out what is the

perception of final consumers on the current and possible features a bank might offer.

Your help matters! Expect to complete this survey in no more than 5 minutes. Thank you!

All the answer are anonymous and will be analysed as a group of observations, and not

individually.

Q1 – Gender (Man, Woman, Prefer not to say)

Q2 – Age (< 18, 18 - 24, 25 - 34, 35 - 44, 45 - 54, 55 - 64, > 65)
Q3 – Education (Basic education, secondary education, bachelor’s, post-graduation, master’s,

PhD)

Q3 – Current situation (not employed, student, student-employed, employed)

Traditional banks are the banks we are used to see when we are walking down the street, with

a physical branch. In Portugal, there are major players in this industry such as Santander,

Caixa Geral de Depósitos or novobanco. Out there, we could use the examples of ABN Amro

(Netherlands), HSBC (UK), Deutsche Bank (Germany), Crédite Agricole (France), among

others.

Q4 – In how many traditional banks do you have an account open? (0, 1, 2, 3 or more)

Q5 – What is your favourite channel to manage your account? (App, website, phone call, e-

mail, going to the physical branch)

Q6 – What are the characteristics you value the most in your bank? Scale them as: Not relevant,

mostly not relevant, indifferent, mostly relevant, relevant. (Security, pricing, app features, card

design, client support)

Q7 – You need to talk to your bank because something is not working with your bank account.

How would you classify the following channels to contact your bank? Scale them as: Not

relevant, mostly not relevant, indifferent, mostly relevant, relevant. (Phone call, going to the

branch, chatbot, e-mail)

“Neobanks, sometimes referred to as “challenger banks,” are FinTech firms that offer apps,

software and other technologies to streamline mobile and online banking. These FinTechs

generally specialize in particular financial products, like checking and savings accounts. They

also tend to be nimbler and more transparent than their megabank counterparts, even though

many of them partner with such institutions to insure their financial products.” Forbes 2021
This definition includes firms such as Revolut, N26, Monzo, Starling Bank and Moey.

Q8 – Do you have an account open in a neobank like Revolut or N26? (Yes, No)

Q9.1 – On what degree do you agree with the following sentence: I opened the account

because… Scale them as: Totally disagree, somewhat disagree, neither agree nor disagree,

somewhat agree, totally agree. (… it is free, … it pays no commissions on purchases abroad,

… it is trendy, … the app is cool)

Q9.2 – What could describe the main reason for not having this type of account? (I never heard

of it, I don’t need to have one, I don’t feel safe)

Q10 – How would you value the following banking features of an app? Scale them as: Not

relevant, mostly not relevant, indifferent, mostly relevant, relevant. (Checking balances, free

transfers, virtual debit card, savings account, budgeting, account aggregation, trading,

exchanging cryptocurrencies, robot-advisor, chatbot, instant push notifications)

Q11 – Would you change your primary account from your principal bank to a neobank because

of innovative features such as the ones described above? (Yes, no)

Q12 – Select the channels you would use to buy the following loans today: Physical branch,

website, app. (Personal loan, student loan, home loan, car loan, business loan)

Q13 – 10 years from now, where do you expect to buy the exact same loans: Physical branch,

website, app. (Personal loan, student loan, home loan, car loan, business loan)

Q14 – Would you use voice commands to check your account balance or to make a bank

transfer? (Yes, no)

Q15 – 10 years from now, do you think you will use voice command to check your account

balance or to make a transfer?

Appendix B.2. – Key results

Answer to questions 12 and 13 from the survey

Appendix C – Figures
Appendix C.1.: The organization of a bank (source: (Deloitte US 2021))

Appendix C.2.: The main applications of AI in the banking industry