GenAI@Work

Is generative AI the silver bullet for the

tight labour market in the Netherlands?

www.pwc.nl
Glossary

AI: Artificial Intelligence (AI) refers to the simulation ‘Artificial Intelligence (AI) has been having an impact
of human intelligence in machines that can perform
tasks typically requiring human cognitive abilities, such
for years. It is everywhere around us. Social media and
as learning from data, reasoning, problem-solving, streaming entertainment are such compelling experiences
perception, language understanding and decision-
because predictive analytics (a form of AI) are so incredibly
making. AI systems use algorithms and computational
power to analyse vast datasets, recognize patterns, make good at anticipating what we will like. From the search
predictions, and adapt to changing circumstances. bar that completes your query about a product you
Generative AI: It is a subset of AI that generates text, want to buy to the distribution centre that handles your
code, images, video and other content (‘outputs’) from order and the optimization of your delivery, in so many
data provided to it or retrieved from the internet (‘inputs’)
in response to user prompts.1 It can create new and
seemingly quotidian yet revolutionary ways, AI in all its
original content, enabling machines to autonomously forms – machine learning, “deep” learning, text processing,
produce content that is often indistinguishable from
speech recognition, image recognition, robotics, and real-
human-created work.
time control – has transformed our lives. Through it all, we
Large Language Models (LLMs): LLMs refer to the continue to enjoy its benefits while relying on specialists
development and utilisation of sophisticated artificial
intelligence models, like GPT-4, that are trained on to bring them to us. For most of us, thinking about the
massive datasets using deep learning techniques. uses, risks, implications, and imperatives of AI has been
Such models are capable of various natural language
processing tasks, including text generation, translation,
somebody else’s job. That’s not true anymore.
summarization, and answering questions. The reason is language.’

AI exposure: The exposure scores indicate how close,

in terms of abilities required for specific tasks, different PwC’s Leaders Guide to Generative AI2
occupations are to AI capabilities.

PwC | GenAI@Work 2
Our main conclusions

What did we find? How did we do it? • T hird, we constructed a technology adoption index
• W e focused on language modeling generative AI • First, we created exposure scores to indicate how and ranked the industries according to it.
(similar to ChatGPT, for example). close, in terms of abilities required for specific tasks, - The index is based on four factors retrieved from
• We found that 44% of jobs in the Netherlands are different occupations are to the language modeling the academic literature.
either highly or very highly exposed to language capabilities of AI. - This allowed us to assess the likelihood that a
modeling AI. - We based our assessment on a recent study that given industry will adopt AI rather than just assess
- In some industries such as financial institutions analysed the exposure of different occupations to how exposed it is.
and education, high or very high exposure applies language modeling AI in the United States.3 ° How AI will ultimately affect the given
to more than 75% of jobs. ° This took into account the tasks that make up occupations depends on the extent to which
- In other industries, mainly those whose primary each occupation and the required abilities to industries accept and adopt AI.
occupations involve more physical work, such as perform them. - Finally, we studied the relationship between
accommodation and food serving, exposure to ° Those tasks are matched with the capabilities of technology adoption, AI exposure and labour
language modeling AI is low. language modeling AI. market tightness.
• The occupations and industries that are highly ° We cannot directly tell whether a given
 • We ended our analysis by looking at the broader
exposed to AI tend to be those with a high occupation’s AI exposure will lead to automation implications of the results for society, governments
probability of labour shortages. or augmentation. and companies.
• We found a very strong correlation between - We adjusted these occupations to the
language modeling AI exposure and the technology occupational classification used in the
adoption index. Netherlands, estimating the exposure to
- Industries such as information and communication 113 occupation groups.
and specialized business services are not only • Second, we incorporated specific labour market
very highly exposed to AI but also seem likely to data.
adopt it. - We combined the impact of language modeling AI
- This implies that language modeling AI can raise with expected labour market tightness indicators
productivity in those industries and help to deal for those occupation groups.4
with labour shortages. ° We are the first to consider this for the
Netherlands.
- We assessed the potential of AI to change
the labour market for certain occupations and
industries.

PwC | GenAI@Work 3
Why is language modeling AI expected
to have a large impact on the labour market?
AI is a general-purpose technology. Such Netherlands have consistently been at very high levels
technologies provide a foundation for various and are expected to persist, mainly because of the
applications and drive productivity improvements. The demographic outlook.9 To maintain economic growth
innovations fueled by them can lead to unintended and the standard of living, the Dutch economy will
positive changes in unrelated fields. Examples of need to produce more with fewer human resources.
general-purpose technologies in history include the
steam engine, electricity, semiconductors and the Accordingly, companies are incentivised to adopt
internet. Crucially, they have the capacity to create technologies that can automate tasks, streamline
long-lasting and sustainable changes in the economy processes and increase overall efficiency. And many
and society.5 are on their way to do so. Ninety-eight per cent
of companies appear to think that AI can help
In addition, an important difference between AI and with labour market challenges.10 Automation and
previous technologies is that the impact of language technology can help achieve higher levels of output
modeling AI will be felt mainly by highly skilled with fewer human resources11, leading to lower costs
workers in the service sector. People in those jobs and higher revenue.12 AI, especially language modeling
comprise the largest share of workers in modern AI, is expected to be at the core of this transformation.
economies: 84% of employees in the Netherlands, for
example, work in commercial and public services.6 Could language modeling AI be the silver bullet to
So far, this group of workers has not been negatively reduce the pressure on the labour market?
impacted by technological advancements in terms of
wages or employment.7 However, recent developments
in AI share similar capabilities to perform tasks
found in non-routine and cognitive work that involve
prediction after processing large amounts of complex
input data, detecting patterns, making judgments, and
optimising.8

To fully capture the current potential impact of

language modeling AI on the labour market, we have
to consider labour shortages. Labour shortages in the

PwC | GenAI@Work 4
The future of the Dutch economy has structural challenges

Labour shortages are widespread Figure 1 In all sectors vacancy rates are higher than before the Covid-19 pandemic
in the economy 2019 Q1 2019 Q2 2019 Q3 2019 Q4 2020 Q1 2020 Q2 2020 Q3 2020 Q4 2021 Q1 2021 Q2 2021 Q3 2021 Q4 2022 Q1 2022 Q2 2022 Q3 2022 Q4 2023 Q1 2023 Q2
As of October 2021, there have been more unfilled All economic
vacancies than unemployed people, which is a activities
phenomenon that has not occurred in the Netherlands Accommodation and
food serving
in over 50 years.13 The job vacancy rate in the Agriculture, forestry
Netherlands was the highest among EU countries and fishing
in the first quarter of 2023.14 Since the Covid-19 Health and welfare
pandemic, this phenomenon has been widespread in
Construction
all industries (Figure 1).
Culture, sports and
recreation
On one hand, this is in large part due to the structural
features of the Dutch economy as the Netherlands Education

has a large number of part-time workers. A previous Energy

report from PwC’s Chief Economist Office showed
that the Netherlands average working hours lag Financial institutions

behind the EU’s average.15 Information and

communication

Manufacturing

Other service
activities
Public administration
and services
Renting and other
business support
Specialised business
services
Transportation and
storage
Wholesale and retail
trade
Source: CBS

PwC | GenAI@Work 5
Labour shortages are mainly driven Population ageing is expected to severely change the composition of the
Figure 2 Population ageing is expected to severely change the composition of the Dutch society
by the demographic situation Dutch society
On the other hand, a big part of the problem lies in
the demographic trends of Dutch society. The number 100
of elderly and inactive people in the Netherlands is
90
increasing, while the increase in the working-age
population is lagging behind (Figure 2). 80

In addition to the pressure on the labour market, the 70

ageing population challenges public finances and the
provision of basic services and public goods. Without 60

Age group
productivity improvements, government expenditure,
50
such as on healthcare, is expected to increase at a
faster pace than the tax base. 40

30

20

10

0
200 150 100 50 0 50 100 150 200
Thousands of people
Women 2050 Men 2050 Women 2023 Men 2023
Source: CBS

PwC | GenAI@Work 6
Productivity has been stagnating Figure 3 Labour productivity has barely increased since 2015
The level of productivity is the most important factor
determining the standard of living.16 To maintain its 105
income growth and standard of living, the Netherlands
will need to produce more with fewer workers. In
other words, productivity needs to increase.
100

Labour productivity (2015=100)

This is a challenge for the economy, especially if we
consider that productivity has almost stagnated since
2015 and increased only marginally since the global 95

financial crisis (Figure 3). Innovation and technology

will play a key role if productivity rises in the
coming years. Could AI be the technology to boost 90
productivity and unlock future economic growth?

85
2008 2008 2012 2014 2016 2018 2020

Year
Source: CBS

PwC | GenAI@Work 7
Dutch labour market exposure to AI

In the Netherlands, 44% of jobs are highly or Figure 4 Percentage of total jobs exposed to AI in the Netherlands
very highly exposed to language modeling AI
To evaluate the potential of AI to deal with labour
shortages in the Netherlands, we combine an analysis 40
of occupational exposure to different types of AI17
with data about the Dutch labour market.18

Furthermore, we want to know not only what the

30
exposure of an occupation is to AI but also how many
people work in this occupation in the Netherlands and

Percentage of jobs exposed

if this occupation is facing labour market constraints.19
• The exposure scores indicate how close, in terms
of abilities required for specific tasks, different 20
occupations are to AI capabilities.20
• A typical occupation consists of 20 to 30 distinct
tasks, some of which are more similar to AI
capabilities than others.21 Tasks susceptible to the
impact of AI tend to have objectives that can be 10
clearly specified and are standardised.22
• We then count how many occupations out of the
total labour force in the Netherlands have different
AI exposure levels.
0

We find that the potential for AI in the Netherlands Very low Low Medium High Very high
is immense: 44% of jobs are highly or very highly Exposure level
exposed to language modeling AI (Figure 4). Our Type of AI
Image generation Language modeling Overall AI
results might change, potentially showing a larger
share of jobs exposed, if considering the impact of
language modeling AI in addition to other forms of AI,
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)
complementary software or robotics.23, 24

PwC | GenAI@Work 8
However, there is substantial variation in Figure 5 Percentage of highly or very highly exposed jobs to language modeling AI in the Netherlands by industry
exposure to AI between industries
• H ere we focus on language modeling AI, but
Financial institutions
our results are also available for other types
Information and communication
of AI.
Education
• In specialised business services, public
Specialized business services
administration, information and communication,
education and financial institutions more than 75% Public adminstration and services

of jobs are highly exposed to language modeling AI. Wholesale

• On the other hand, less than a quarter of jobs Other business sectors

are highly exposed in retail, agriculture, forestry Total

and fishing, construction and in manufacturing Metal industry
industries. This is a clear reversal from previous Culture, sports and recreation
technology waves, which had a larger impact on Industry Welfare
industries with a higher proportion of blue-collar Renting and other business support
jobs.25 Health
Other service activities
Are the occupations with high AI exposure Transportation and storage
the same ones where we expect labour Energy
shortages? Chemical industry
Next, we connect exposure to language modeling AI Construction
to labour market constraints.26 Occupations with a Retail
high probability of labour shortages tend to have high
Other industries
language modeling AI exposure, indicating that such Exposure level
Food and beverage industry
AI could make workers in those occupations more High
Agriculture, forestry and fish
productive and help with expected labour shortages. Very high
Accommodation and food serving

0 25 50 75
Percentage of jobs exposed
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 9
 Figure 6 T
 he occupations with high language modeling AI exposure are the same ones where we expect
labour shortages
Lawyers

Teachers of higher education and professors

Social workers Psychologists and sociologists

Financial specialists and economists
Secretaries Accountants Policy advisers Authors and linguists Teachers of general subjects in secondary education
Journalists Business administration and organizational consultants
Receptionists and telephonists Representatives and buyers
Sales and marketing managers Educational experts and other teachers
Accounting staff
Business and administrative services managers Government executives Education managers Primary school teachers
Executive secretaries
Care institutions managers Specialist services managers
Very high Bookkeepers
General managers Software and application developers Engineers (not electrical engineering)
Commercial and personal services managers ICT managers
Biologists and natural scientists
Administrative assistants and general office clerks Business service providers
Database and network specialists
Social workers, group and residential supervisors Production managers in manufacturing, mining, construction, and distribution
Retail and wholesale managers
Logistics managers in manufacturing, mining, construction, and distribution
Language modeling AI exposure

Architects Electrical engineers

Graphic and product designers Librarians and curators
Government officials Police inspectors
Teachers of vocational subjects in secondary education
High User support ICT Childcare leaders and teaching assistants
Physiotherapists Doctors, veterinarians and other health professionals
Transport planners and logistics staff
Retail sales associates Retailers and retail team leaders Traditional and complementary medicine associate professionals
Hospitality managers Providers of other personal services Specialized nurses Jobs in 2026
Performing artists Carers
Travel supervisors Visual artists 100000
Radio and television technicians Nurses (mbo)
Medium Deck officers and pilots
Physical and engineering science technicians 200000
Laboratory assistants Medical specialists
Cashier associates Hairdressers and beauticians
Pharmacy assistants
Production managers in agriculture, forestry and fisheries
300000
Photographers and interior designers
Production leaders industry and construction Life science technicians and related associate professionals
Sports instructors
Waiters and bar staff Medical practice assistants
Security personnel
Low Call center associates Outbound and other vendors
Drivers of cars, taxis and vans Police and firefighters
Cooks Concierges and cleaning team leaders Process Operators
Bus drivers and tram drivers
Printing and crafts workers
Truck drivers Electricians and electronics fitters
Loaders, unloaders and stock fillers Butchers Bakers Auto fitters
Product inspectors Machine fitters
Assembly workers

Kitchen helpers Gardeners, market gardeners and breeders

Very low Garbage collectors and newspaper deliverers
Carpenters Construction Workers
Furniture makers, tailors and upholsterers
Production machine operators Livestock growers
Farmers and foresters
Metalworkers and construction workers Mobile machine operators
Construction workers finishing Plumbers and pipe fitters
Auxiliary workers in agriculture Cleaners

Auxiliary workers in construction and industry

Painters and metal sprayers

Very low Low Medium High Very high

Probability of labour shortages
PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)
Souce: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 10
Starting with occupations in black in Figure 627,
there are occupations with low language modeling AI
exposure, such as waiters, loaders or construction
workers, that also have a low probability of labour
shortages. We do not expect much impact on those
occupations from this type of AI.

Following that, the bottom right corner is relatively

empty. This implies that there are very few, if any,
occupations with high expected labour shortages and
low potential for AI to tackle those.28

In the upper half of the graph, occupations are more

evenly distributed in terms of the probability of partially or fully automated as AI takes on some experience and technical knowledge. This type of AI
labour shortages. In addition, we see that those are tasks and reduces the number of employed people is reducing the cost of specific cognitive expertise.31
occupations mainly in the services sector that tend to required, leading people to relocate to other tasks However, doctors and lawyers also do different tasks
have high exposure to language modeling AI. or occupations.29 Probably these occupations will in their work, of which applying their narrow expertise
not disappear completely, but employees will face is just one of them. For example, tasks that require
On the top right in orange are occupations such as the most pressure to adjust their skill set to the new more interpersonal contact would be much less likely
engineers and teachers. They have very high expected composition of tasks that will be required to remain impacted.32
labour shortages and very high exposure to AI. This competitive. The commonly used phrase ‘AI will not
implies that AI has the potential to play a large role replace humans, but humans with AI will replace At the same time, the probability of labour shortages
in increasing the productivity of workers in those humans without AI’ applies here.30 shown here is based on the predicted supply of
occupations. workers. This is usually more constrained where
These results can be partly explained by the nature significant amounts of training are required, limiting the
In the top left corner coloured in red are occupations of this type of AI, such as LLMs. These models are entrance of new workers into the labour market. That
such as bookkeepers and administrative assistants. trained on massive amounts of data. That is similar would mean both AI exposure and predicted labour
Those occupations are highly exposed to AI, but to a doctor or a lawyer who studies to become a shortages are positively correlated to education and
expected labour shortages are not that high. There specialist in a very narrow field of medicine or law and training.
might be a risk that some of these jobs end up either diagnoses rare diseases or takes on cases based on

PwC | GenAI@Work 11
Box 1: How could financial analysts be affected by language modeling AI?

Financial firms have been using AI for many years.33 Which tasks could be affected by language How could this affect jobs in the finance industry?
By studying the current developments at the modelling AI? Some tasks, such as data collection, fundamental
intersection of finance and AI, we can gather insights • Data gathering and research: given that analysis and reporting, can be at least partially
into the changing role of a financial analyst. the current versions of language modelling AI automated. However, not all roles will become
can organise and summarise large amounts AI-driven.
Which tasks do financial analysts perform now? of information rapidly, it can improve the data
Financial analysts process large amounts of data and gathering process. There are already examples of Output interpretation, communication and
information to advise clients on investment decisions, this on the market, such as BloombergGPT, an AI stakeholder management will likely remain human
usually within their own firm or in another company. launched by Bloomberg earlier this year that has tasks. As Harvard University’s Professor of Finance
made it easier to use the Bloomberg Terminal. Mihir Desai points out, ‘it appears that the client
These are the main tasks that they perform: Rather than writing queries in Bloomberg Query side of finance retains a preference for humans’,38
• Gathering large amounts of high-quality financial Language, which can be unintuitive, financial indicating that the communication between financial
data and information from different databases. analysts can simply write questions in regular institutions and their customers might still be
• Analysing the retrieved data using software. English, thereby making data collection more managed by people. Still, this will leave a large pool
• Modeling and forecasting a company’s financial efficient.34 of highly skilled professionals with more time.
performance based on the gathered data, a set of • Financial analysis: similarly, AI could speed up the
assumptions and market sentiment. financial analysis process itself by quickly breaking Accordingly, it could be that employers train current
• Interpreting the analysis output. For example, down financial statements and market sentiment employees on digital skills that, in combination with
assessing whether the client should invest in the to answer questions such as ‘how is company ‘A’ their financial expertise, could help augment returns
stock of the company whose performance was doing relative to its competitors?’.35 on human capital. Similarly, the financial industry’s
forecasted. • Real-time trading: quantitative financial analysis, demand for quantitative skills in the labour market
• Communicating these results to the client, usually which uses AI to process enormous amounts is likely to expand. The effect is already becoming
in the form of a slide deck or a report that is of data to then automatically profit off market clear, as seen by the inflow of people with STEM
presented in a meeting. inefficiencies, is becoming increasingly popular.36, 37 backgrounds into financial institutions.39
• Creating reports and presentations: generative
AI could also help produce reports and slide decks
faster, a task that is deemed time-consuming by
financial analysts.

PwC | GenAI@Work 12
Moving from occupation to industry language modeling
AI exposure Financial institutions Education
Information and communication

We reproduce the same analysis for all industries Specialized business services
Figure 7 Many service industries have high language modeling AI exposure and high expected labour shortages
by incorporating data that captures how many Public administration and services
different types of occupations exist in eachVery
industry.
high
Financial institutions Education
We see that of the five industries with the highest
Information and communication
labour shortage probability, namely energy, health, Language modeling
Health Specialized business services AI exposure

Language modeling AI exposure

education, specialised business services, and High Wholesale Welfare
Public administration and services
Culture, sports and recreation Very low
information and communication, the latter three also Low
Very high Other business sectors
have very high language modeling AI exposure. This Other service activities
Medium
Language
Very high modeling
implies that language modeling AI has the potential
Medium to Metal industry AI exposure
Language m
reduce labour shortages in those industries. Retail Financial institutions Education
Health Total jobs AI exposure

Language modeling AI exposure

High Welfare
Transportation and storage Wholesale
Information and communication
Renting and other business support Culture, sports and recreation 400000 Very low
Low Specialized businessOther
services
business sectors Energy Low
Chemical industry 800000
Medium
Other service activities 1200000 Very high
Public administration and services
Metal industry
Medium
Very high 1600000
Retail
Very low Total jobs
Accommodation and food serving Transportation
Construction and storage
Language modeling 400000
Food and beverage
Low industry Other industries Renting and other business support
Health AI exposure
Energy 800000
Language modeling AI exposure

High Welfare
Chemical industry
Wholesale
Culture, sports and recreation Very low 1200000
Other business sectors Low
Medium 1600000
Other service activities Very high
Very low Metal industry
Medium Accommodation and food serving Construction
Agriculture, fishingand beverage industry Other industries
forestry and Food
Total jobs
Retail
Total jobs
Transportation and storage
400000
Very
Lowlow Low Medium RentingHighand other business support
Very high Energy
Chemical 800000
Probability of industry
labour shortages
PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021) 1200000

Agriculture, forestry and fishing 1600000

Very low
Accommodation and food serving Construction
Food andVery
beverage Other industries
low industry Low Medium High Very high
Probability of labour shortages
PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

Agriculture, forestry and fishing

PwC | GenAI@Work Very low Low Medium High Very high 13

Probability of labour shortages
PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)
Box 2: Why is Energy an outlier among all industries in our analysis?

Figure 7 shows that the energy industry is a clear On the other hand, financial institutions have a very
outlier: it has a very high probability of labour low probability of a labour shortage and a high
shortages but a low exposure to AI language exposure to AI.
modeling. Why is this the case? • Most of the current tasks of financial analysts
• Most jobs in the energy industry, such as installing could be at least partially automated (see Box 1).
electric grid systems, demand manual labour This is because they revolve around gathering,
skills, which by their nature cannot be performed categorising and analysing large amounts of
by AI and require human involvement. Hence, (financial) data, something that AI can do really
almost half of the jobs in the energy industry well. In addition, generative AI could develop the
have low or very low exposure to language ability to generate the graphs and visuals needed
modeling AI. to present analyses to clients.
• However, the energy industry has a high • Yet, Economics and Business Administration
probability of labour shortages because these remains the second most popular major in the
occupations require technical skills that may not Netherlands, which decreases the probability of
be readily available on the market.40 labour shortages in financial institutions.41 This
is best seen through the famously competitive
recruitment processes of financial institutions, as
the large supply of human capital gives them the
ability to be very selective.

PwC | GenAI@Work 14
Box 3: How could education be affected by language modeling AI?

Education is one of the most discussed industries • Teachers assess the learning process of the variety of explanations of the same concept that
in the context of AI. Experts have been monitoring students through formal evaluations, such as tailor to the student’s intuition, making the teaching
the impact of AI in education for some time now, as exams, presentations or research projects. process more efficient for both the teacher and the
seen in the UNESCO 2019 summit on the matter.42 • Lastly, teachers also receive feedback from the student.46
This prompts the question: what could education students at the end of the education period to • Learning evaluations: surprisingly, AI could also
and educators look like in the age of AI? We look improve the teaching process. help during learning evaluations. For instance,
specifically at educators involved in teaching roles. teachers could ask students to detect mistakes in
Which tasks could be affected by language AI output to assess their mastery of the subject.
Which tasks do teachers perform now? modeling AI?
Teachers offer knowledge, guidance and mentorship Despite initial setbacks, many teachers are already How could this affect jobs in the education
during critical periods of students’ lives. These are developing ways to use language modeling AI and industry?
the main tasks that they perform: similar technologies in the classroom.43 Since education is a fundamentally social process,
• By setting learning goals and monitoring the • Designing the curriculum and teaching the probability of AI completely substituting teachers
learning process, teachers decide what concepts materials: generative AI can be used to assist is extremely low.47 Even if curriculum design is
students should master and at what level of in the design of the curriculum and the teaching entirely automated, there must be someone who
difficulty. materials, saving time for teachers. However, can monitor, talk to, and, most importantly, guide
• Teachers design the curriculum and teaching for now, human input would still be necessary students during these critical stages of life.
materials by selecting the topics that will be taught to inform the AI about learning objectives and
in class, their respective teaching approaches and educational policy standards and detect possible Nonetheless, it is crucial that teachers’ training
the academic literature. mistakes.44 remains up-to-date with the latest technologies.48
• They communicate and reinforce knowledge with • Grading: additionally, generative AI could also help Policymakers must ensure that teachers (especially
the help of teaching material and answer student with grading. This could speed up a task that was those who are less familiar with technology) know
questions throughout the course. These activities previously very time-consuming for teachers, even how to use language modeling AI and its detection
can take various forms, depending on the field outside of class hours. software in the classroom. If this opportunity is
of study, such as theoretical classes or hands-on • Personal tutoring: language modeling AI could worked out correctly, society could augment its
laboratory work. also be used to help students with different return on investment in educators and enjoy a higher
learning styles.45 For instance, it could output a quality of human capital.

PwC | GenAI@Work 15
But is language modeling AI adoption
actually going to happen? And where?
After identifying that the potential is there, we move to instead of relying on generally trained tools.56 the required vacancies.57 Hence, they have a greater
another question: will the potential be realised in the We incorporate this aspect by looking at the share motivation to adopt technology. To capture this
industries exposed to language modeling AI? of big firms and the average wage in an industry. aspect, we include the average probability of labour
• Labour: The need due to labour shortages affects shortages and vacancy rates in an industry.
What are the factors behind technology the usefulness of technology implementation. • Skills: Overall workforce skills and education level,
adoption? Companies and sectors with more pronounced especially in information and communications
Acceptance and adoption of technology depend on labour shortages can get a larger benefit from technology (ICT) fields, are important to acquire
the perception of its usefulness and ease of adoption.49 technology implementation to fill at least some of the skills to more easily use and implement novel
A number of factors are behind these aspects on a firm
and industry level, which we identify in four pillars:50
• Culture: In this category, we incorporate two Figure 8 Industry ranking by the technology adoption index categories
metrics that capture how open industries are to
Information and communication
new technologies: the share of young companies in Education
an industry and the share of companies with union Specialized business services
Health
relations. Younger companies, which tend to be Public adminstration and services
more innovative and less constrained by existing Financial institutions
ways of working, tend to also have higher shares Renting and other business support
Welfare
of AI use.51 In addition, the higher the share of Chemical industry
companies with union relations in an industry, the Other business sectors
Metal industry
more bargaining and negotiation usually takes place
Industry

Energy
before technology is adopted.52 This can already Wholesale
be seen with the Screen Actors Guild strike in the Culture, sports and recreation
Transportation and storage
United States.53 Retail Technology adoption
• Finance: Financial aspects play a critical role in the Food and beverage industry index categories
Other service activities Culture
ease of adoption, especially if big investments are Other industries Finance
required. Previously, large54 and more profitable Accommodation and food serving Labour
companies were early adopters of technology.55 Construction
Skills
Agriculture, forestry and fish
Moreover, in the current age of AI, large companies
have the advantage of conducting more research Technology adoption index
Source: PwC analysis based on data from CBS and ROA.
and training their own models with proprietary data

PwC | GenAI@Work 16
 technology. The prevalence of ICT professionals
has been an indicator of technology adoption in the
past.58 In addition, more educated labour forces also
tend to have better knowledge of the potential uses
of technology.59 To capture that, we look at the share
of ICT professionals and the share of occupations
with higher education degrees.

Industries highly exposed to language

modeling AI also have the likely adoption
conditions
Based on these four pillars, we construct a technology
adoption index. Then we correlate and plot it with the
industry language modeling AI exposure (Figure 9).

We find a strong and statistically significant correlation

between our technology adoption index and language
modeling AI exposure (we reach similar results for
overall AI exposure, see Appendix B on page 30).
This is encouraging, as it indicates that industries
more exposed to language modeling AI also have the
conditions to adopt the technology.60

Our results coincide with previous research on other

forms of AI adoption in Europe and elsewhere, showing
that industries more exposed to AI, such as information
and communication, specialised business services and
financial institutions, have also been the main adopters
of AI and related technology so far.61, 62

PwC | GenAI@Work 17
 Figure 9 Industries highly exposed to language modeling AI also have the likely adoption conditions

Financial institutions Education

R = 0.85, p = 0.00000058
Information and communication
Specialized business services

Public administration and services

Very high
Language modeling AI exposure

Health
High Welfare
Culture, sports and recreation Wholesale
Other business sectors
Other service activities
Medium Metal industry
Retail
Transportation and storage
Renting and other business support
Energy
Low Chemical industry

Very low Accommodation and food serving

Construction Food and beverage industry
Other industries

Agriculture, forestry and fishing

Very low Low Medium High Very high

Technology adoption index
PwC analysis based on data from Felten et al. (2023), Bakens et al. (2021), CBS and ROA
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 18
What should governments and society consider
about the impact of AI on the labour market?
As AI technologies continue to advance and permeate In addition, new jobs have been created after more from technology if they provide tasks that are
various sectors of society, governments play critical major technological advancements, thanks to the complemented by it. If they provide only tasks that are
roles in shaping their development, deployment, and complementarity between humans and technology. substituted, their labour market prospects will not be
impact. Here, we limit our reflections to labour market Sixty percent of employment in 2018 was in positive.72
consequences. occupations that did not exist before 1940.67
But the complexity does not end here. If the tasks that
Technology changes jobs but does not Between 2011 and 2019, a period that coincides with relationship bankers supply can be easily performed
necessarily make them disappear the rise in deep learning applications, employment in by other workers, a flood of new workers might
One of the largest sources of insecurity towards jobs highly exposed to AI has actually increased in become relationship bankers, increasing the supply
AI and technology in general lies in its potential to Europe.68 There is also already evidence of a change and mitigating any possible wage gains that would
automate jobs, increasing unemployment. So far 54% in skill demands in companies more exposed to AI, come from the complementarity between humans and
of companies adopting AI have done so primarily for but, at least until 2018, the effect of AI on employment technology.73
automation purposes.63 demand or wage levels has been negligible.69, 70
This, however, could change with recent and future Finally, the elasticities of demand and the income
New technologies display two opposite effects on advancements in AI.71 elasticity of demand also have to be taken into
employment. On the one hand, substituting human account. If, for example, AI makes the work of lawyers
labour in order to decrease production costs and The economic equation is complicated more automated, reducing the need for labour, this
increase productivity would lower employment. On The number of ATMs in the US quadrupled between could allow legal services to be offered at lower prices.
the other hand, reduced production costs increase 1980 and 2010, but at the same time, the number of It is likely that the demand for some types of lawyers
real incomes and demand. The latter effect fosters bank tellers increased by 10%. How is this possible? will increase as a larger market can now afford certain
production and job creation.64 The rise of ATMs decreased the cost of running a legal services. However, for others, such as criminal
bank branch, increasing the number of branches. So lawyers, the demand should be less responsive to
Until now, technological developments have mainly there were fewer bank tellers required to run a branch. price changes. So, if, on one side, the reduction in
affected wages and wealth distribution, not how many But the increase in the number of branches more prices would reduce the size of the legal business
jobs were available.65 That does not mean jobs have than compensated for that. The tasks performed by sector, on the other, the increase in demand might
not disappeared or largely changed in terms of tasks bank tellers changed as they became more involved (at least partially) compensate for that. The effect on
performed. The impact on task composition has been in relationship management with clients instead of aggregate demand will depend not just on that but also
the main way technology has impacted occupations counting bank notes. That altered the nature of the on how clients who now have access to legal services
so far.66 job, which might not be appreciated by all bank at a lower price will spend their savings.
tellers. To put this formally, workers might benefit

PwC | GenAI@Work 19
Innovations can impact income inequality It is uncertain that AI will have a similar What role do governments play in this labour
But looking only at how much human work could effect on the labour market as previous market transformation?
be replaced by AI misses a large part of the picture. technologies In addition to regulation and ensuring an adequate
Although past waves of technological innovation and First of all, as the meteoric rise of LLMs showed, it is social safety net, it is crucial that governments pay
automation did not lead to an increase in aggregate very difficult to anticipate where the future of AI will attention to AI developments and, if necessary, make
levels of unemployment, there has been a change in take us. Recently, a survey of around 30 economists changes to the tax system.88, 89 Theoretically, if the
income distribution caused by job polarisation.74 showed that the majority believe that AI will not productivity gains from AI are captured mainly by
increase unemployment rates in high-income countries. the owners of capital because the dominant effect
Most of the benefits have been obtained by highly However, the panellists expressed a great degree of is automation in relation to augmentation, our main
skilled workers, who have seen increasing incomes uncertainty in their predictions, noting that ‘we are problem would not be one of scarcity but one of
in the past decades. At the same time, workers in the still in the very early days of AI’.79 A good heuristic distribution. In this case, there might be a strong case
lower and middle parts of the skill distribution had seen to understand the potential impact of AI on specific to increase the tax on capital gains and redistribute
real wages decrease.75 occupations is to ask for whom AI is a substitute and income.
for whom it is a complement.80
As we discussed earlier, the tasks that language AI increases the need for digital, analytical, and social
modeling AI can affect are different than in previous Preliminary evidence points out that, at least in terms skills.90 Sixty per cent of workers will require training
technological waves. Importantly, it can affect white- of performance, AI could reduce inequality within firms before 2027, but only half of them have access to
collar and high-wage occupations, what, in theory, by flattening their hierarchical structure.81 Access adequate training opportunities today.91 The good
could even reduce inequality.76 However, in general, to generative AI tools can increase productivity in news is that only elementary digital skills and analytical
the distributional effects of technology depend more customer service82, professional writing83 and software thinking are sufficient to use and interact with AI
on which workers have tasks that get automated than development roles84, with the greatest impact coming applications.92
on the fact of automation itself.77 from novice and low-skilled workers, who also seem
more likely to adopt those tools.85, 86 The responsibility rests with workers and companies to
It is also telling that the majority of the growth in remain competitive and follow the latest technological
inequality across workers so far has been due However, although generative AI might indeed developments. But if they fail to do so, governments
to increasing average wage differences between democratise access to certain white-collar jobs by can help provide the labour force with the education
companies.78 AI and similar innovations have the reducing the costs of certain expertise87, it might also necessary to adapt. This could involve a combination
potential to further split companies into highly be harmful for more junior employees in the long term. of policy development, funding allocation and
productive and less productive ones, making the By prematurely relying on AI tools in their careers, junior collaboration to ensure the workforce remains
earnings of workers employed in the latter suffer. employees might not learn the fundamental aspects of competitive and adaptable.
their jobs that come with dedicated effort and expertise.

PwC | GenAI@Work 20
What should companies consider when thinking about
AI and the workforce?
According to PwC’s recent Hopes and Fears survey, Exposure to AI is not a management decision.
the majority of workers have a positive view of the It is unavoidable that business partners will start
impact of AI on their jobs.93 In addition, over 85% of embedding generative AI in their processes and that
organisations see the adoption of frontier technologies employees will start experimenting with generative AI
and broadening digital access as the most likely possibilities. This means that priority is not just about
drivers of transformation in their organisation in the bringing this new technology and capability into the
next few years.94 Although there is a lot of potential organisation, but also about managing and directing
and excitement, the diffusion of technological what is already coming and preparing for what’s
improvements is as important as innovation itself. It is next.96
possible that the effects of AI will not build up linearly
but rather at different rates for different countries, That is why it is important to have an AI strategy that
industries, companies and employees. is clear, transparent and communicated to employees
about the available potential of AI and upskilling
Currently, few companies use generative AI tools at requirements. Generative AI has the potential to
scale.95 But there are big opportunities ahead, and it reinvent the way work is done in many organisations,
is the responsibility of companies and their leaders but without the support and energy of all people, these
to harness the benefits of AI. At the center of that will efforts will fail to capture the benefits.97 Realising the
be human and technology cooperation: the future is potential of AI might be challenging, but it will be well
human-led, tech-powered. Companies that are able worth it.
to leverage this collaboration will have the highest
chance of success.

Companies must strategically position themselves to

benefit from the disruption along the way. And this also
includes making a strategic evaluation of the factors
that might influence their AI adoption, including having
the skills required, making the needed investments
and having employees onboard.

PwC | GenAI@Work 21
Appendix A: Methodology

AI exposure scores The final result of this step is a data sheet with AI Dutch labour market data
We use the AI exposure scores of Felten et al. (2023). exposure scores for 113 occupations based on the The labour market data used in this study comes from
This study attempted to determine the occupations ISCO 3-digit codes. The obtained AI exposure scores the Research Centre for Education and the Labour
most exposed to AI by linking ten AI applications (e.g., were standartised to have 0 as the mean across all Market (ROA) of Maastricht University.101 We use their
image generation, language modeling, abstract strategy occupations and 1 as the standard deviation. Hence, industry classification. ROA uses an econometric model
games, real-time video games, etc.) to 52 human they show relative AI exposure between all occupations to forecast labour market data for the year 2026. For
abilities (e.g., oral comprehension, oral expression, in the labour market. From there, we cannot tell exactly each occupation, the data includes the total number of
inductive reasoning, arm-hand steadiness, etc.). what share of each occupation group is exposed to employees in this occupation and how this occupation
AI, but we can understand which occupations in the is distributed among industries.
An important caveat is that the term ‘exposure’ is labour market are most exposed. We classify each
used so as to be agnostic as to the effects of AI on occupation’s exposure scores from very low to very To assess the probability of labour shortages, we
the occupation, which could involve substitution or high as follows: use the Indicator for Future Staffing Bottlenecks by
augmentation depending on various factors associated • Exposure scores less than -1 as ‘very low’. Occupation (ITKB). It reflects the expected labour
with the occupation itself. • Exposure scores between -1 and -0.5 standard market friction by occupation, giving the probability that
deviations as ‘low’. this occupation will be able to match labour demand
ISCO to SOC occupation classification • Exposure scores between -0.5 and 0.5 standard with supply. To obtain the probability of labour shortages
code crosswalk deviations as ‘medium’. in this occupation, we take 1-ITKB.
Because the data in Felten et al. (2023)98 is based on • Exposure scores between 0.5 and 1 standard
the United States, we had to perform a crosswalk from deviations as ‘high’. Merging the data
the Standard Occupation Classification (SOC) system to • Exposure scores above 1 standard deviation as Next, the obtained generative AI exposure data and
the International Standard Classification of Occupations ‘very high’. Dutch labour market data were merged based on
(ISCO). This was necessary to link the AI exposure the corresponding ISCO codes. There were minor
scores with Dutch labour market data. This crosswalking We apply the same classification for occupations and discrepancies due to crosswalking in some occupations,
was done based on the method in Kouretsis and industries with their AI exposure scores and labour and those were adjusted. Figure 10 summarises the
Bampouris (2022).99 In addition, the 771 occupation shortage probability. methodology approach.
groups100 (SOC 4-digit) were aggregated to 113 (ISCO
3-digit) to match the labour market data.

PwC | GenAI@Work 22
Figure 10 Summary of the methodology

Impact of AI on the
Dutch labour market

20-30 tasks per 774 types of 113 occupation

10 AI applications 52 human abilities occupation occupations groups 22 industries

- Lawyers
- Lawyers - Education
- Image generation - W
 ritten - W
 riting - S oftware and
- P rimary school - Specialized 20-30 tasks
- Language modeling comprehension - R eading application
teachers business services per occupation
- Abstract strategy - Inductive reasoning - Categorizing of developers
- Electrical engineers - Financial
games - Memorizing information - Teachers of higher
- Cooks institutions
- Etc... - Etc... - Etc... education
- Etc... - Etc...
- Etc...

Labour market
Occupation-level AI exposure in the Netherlands

Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

AI exposure analysis Occupation AI exposure score * number of jobs classification of occupation group exposure to language
Furthermore, to incorporate the occupational group in this occupation in the Netherlands. We obtain modeling AI. This is important as it not only looks at the
exposure perspective we calculate the AI exposure occupation group AI exposure scores that we technical AI exposure in terms of ability similarity but
scores as follows: standardise to be able to classify them from ‘very low’ also at how prevalent this occupation is in the economy.
to ‘very high’. What we obtain is an economy-wide

PwC | GenAI@Work 23
 For industries, we do a similar analysis. We take the occupation AI exposure score * number of jobs
in this industry, summing all relevant occupations to get the total industry AI exposure score.

Figure 11-1 Top five most relevant occupations in each industry

Accommodation and food serving Agriculture, forestry and fishing Chemical industry

Waiters and bar staff Livestock growers Production machine operators

Cooks Farmers and foresters Process Operators

Kitchen helpers Garderners, market Physical and engineering

gardeners and breeders science technicians
Call center associates Auxiliary workers Life science technicians and
Outbound and other vendors in agriculture related associate professionals
Hospitality managers Mobile machine operators Engineers (not electical
engineering
0 10 20 30 0 10 20 30 0 5 10 15

Construction Culture, sports and recreation Education

Carpenters Performing artists Primary school teachers

Occupation

Construction workers Sports instructors Teachers of general subjects

in general education
Electricians and Waiters and bar staff Teachers of higher
electronic fitters education and professors
Visual artists Educational experts
Construction workers finishing and other teachers
Plumbers and pipe fitters Adminstrative assistants Childcare leaders and
and general office clerks teaching assistants
0 5 10 0 5 10 15 0 10 20

Energy Financial institutions Food and beverage industry

Engineers (not Financial specialists
electical engineering Production machine operators
and economics
Garbage collectors and
newspaper deliverers Bookkeepers Retail sales associates
Process operators
Representatives and buyers Bakers
Production leaders
industry and construction Software and Auxiliary workers in
Physical and engineering application developers construction and industry
science technicians
Science technicians and related Business service providers Transport planners and logistics staff
associate professionals
0 5 10 0 5 10 0 5 10 15 20

Source: PwC analysis based on data from Bakens et al. (2021). Percentage share of total jobs in the industry

PwC | GenAI@Work 24
 Figure 11-2 Top five most relevant occupations in each industry

Health Information and communication Metal industry

Doctors, veterinerians Software and
Engineers (not electrical engineers)
and health professionals application developers
Marketing, public relations Metalworkers and
Specialised nurses
and sales consultants construction workers
Medical practice assistants Database and Transport planners and logistical staff
network specialists
Physiotherapists User support ICT Software and application developers

Physiologists and sociologists Journalists Production machine operators

0 5 10 15 0 10 20 30 40 0.0 2.5 5.0 7.5 10.0

Other business sectors Other industries Other service activities

Furniture makers,
Cleaners Hairdressers and beauticians
tailors and upholsterers
Occupation

Software and application Auxiliary workers in providers of other personal services

developers construction and industry
Garderners, market Social workers, group and
Retail sales associates gardeners and breeders residential supervisors
Loaders, unloaders and stock fillers Adminstrative assistants Carers
and general office clerks
Social workers, group Printing and crafts workers Furniture makers, tailors
and residential supervisors and upholsterers
0 1 2 3 4 0.0 2.5 5.0 7.5 0 10 20 30 40

Public administration and services Renting and other business support Retail

Government officials Cleaners Retail sales associates

Policy advisers Receptionists and telephonists Loaders, unloaders and stock fillers

Police and firefighters Gardeners, market

Retailers and retail team leaders
gardeners and breeders
Administrative assistants Security personnel Cashier associates
and general office clerks
Software and Administrative assistants Auto fitters
application developers and general office clerks
0 3 6 9 12 0 5 10 15 20 0 10 20

Source: PwC analysis based on data from Bakens et al. (2021). Percentage share of total jobs in the industry

PwC | GenAI@Work 25
 Figure 11-3 Top five most relevant occupations in each industry

Specialised business services Transportation and storage Welfare

Accountants Truck drivers Carers

Business administration and Drivers of cars, taxis Social workers, group and
organizational consultants and vans residential supervisors
Engineers (not electrical Administrative assistants and Childcare leaders and
engineers) general office clerks teaching assistants
Marketing, public relations Transport planners and Cleaners
and slaes consultants logistics staff
Loaders, unloaders and Specialized nurses
Barkeepers
stock fillers
0 2 4 6 0 5 10 15 0 10 20

Wholesale

Representatives and buyers

Occupation

Loaders, unloaders
and stock fillers
Transport planners and
logistics staff
Marketing, public relations
and slaes consultants
Retail sales associates
0 2.5 5.0 7.5

Percentage share of total jobs in the industry

Source: PwC analysis based on data from Bakens et al. (2021).

PwC | GenAI@Work 26
Finally, we divide this score by the total number of jobs in this industry to obtain a weighted industry AI exposure score. Then we also standardise these
scores to get exposure from ‘very low’ to ‘very high’ and calculate the share of jobs exposed in each category for all industries (Figure 12).

Figure 12 Percentage of total jobs exposed to AI in the Netherlands by industry

Accommodation and food serving Agriculture, forestry and fish Chemical industry Construction Culture, sports and recreation Education
80

60 40 60
30 Type of AI
60 Image generation
40
30 Language modeling
Overall AI
40 20 40
40
20
20
20 10 20
20 10

0 0 0 0 0 0
Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high

Energy Financial institutions Food and beverage industry Health Information and communication Metal industry
50 80 80
40
60
30
40 60 60
30
30 40 20
40 40
20
20
20 10
20 10 20
10

0 0 0 0 0 0
Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high

Other business sectors Other industries Other service activities Public administration and services Renting and other business support Retail
60
40 50
60 30
30
40
30
40
40 20
20 30
20
20
20 10 20
10
10
10

0 0 0 0 0 0
Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high Very low Low Medium High Very high

Specialized business services Transportation and storage Welfare Wholesale

40
60
60 Type of AI
40 30
40
40
20
Image generation
20
20 20 10 Language modeling
0
Very low Low Medium High Very high
0
Very low Low Medium High Very high
0
Very low Low Medium High Very high
0
Very low Low Medium High Very high
Overall AI
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 27
Incorporating technology adoption factors • T
 he data for wholesale and retail for both retail and
We construct the technology adoption index in wholesale.
three steps: • The data for health and social work activities for both
health and welfare.
1. W
 e aggregate data on eight different indicators,
divided into four pillars:
• Culture: share of young companies and share of
employees not unionised, based on CBS data.
• Finance: share of companies with more than 50
employees and average wage, based on ROA data.
• Labour: average probability of labour shortages,
constructed using the ITKB from ROA, and vacancy
rates, CBS data.
• Skills: share of ICT professionals and share
of occupations with higher education degrees
(bachelors level or higher), based on ROA data.

2. T
 o create the individual pillar scores, we standardised
the indicators and averaged the indicators inside
each pillar, assigning equal weight to each indicator.

3. T
 o construct the technology adoption index, we use
a standardised average of the pillars.

The industry classification differs slightly between

CBS and ROA. For the indicators where the data source
is CBS:
• We use the numbers for manufacturing for all
manufacturing industries.

PwC | GenAI@Work 28
 Appendix B: Main results for overall AI

Figure 13 Percentage of highly or very highly exposed jobs to AI overall in the Netherlands by industry

Financial institutions
Information and communication
Education
Specialized business services
Public adminstration and services
Care
Other business sectors
Wholesale
Total
Metal industry
Culture, sports and recreation
Industry

Welfare
Renting and other business support
Other service activities
Transportation and storage
Energy
Chemical industry
Construction
Retail
Other industries
Food and beverage industry Exposure level
Agriculture, forestry and fish High
Very high
Accommodation and food serving

0 25 50 75
Percentage of jobs exposed
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 29
 Figure 14 Occupation group distribution by overall AI exposure and labour shortage probability

Accountants Lawyers
Financial specialists and economists
Teachers of higher education and professors
Personnel and career development specialists Social workers
Accounting staff Psychologists and sociologists
Representatives and buyers Sales and marketing managers Teachers of general subjects in secondary education
Administrative assistants and general office clerks Journalists Policy advisers Software and application developers
Secretaries Engineers (not electrical engineering)
Bookkeepers Government executives Authors and linguists ICT managers
Executive secretaries Care institutions managers Biologists and natural scientists
Very high Receptionists and telephonists General managers Specialist services managers Education managers
Electrical engineers
Commercial and personal services managers Graphic and product designers Database and network specialists Educational experts and other teachers Architects
Logistics managers in manufacturing, mining, construction, and distribution Production managers in manufacturing, mining, construction, and distribution
Business service providers Librarians and curators
Doctors, veterinarians and other health professionals
User support ICT Physiotherapists
Retail and wholesale managers Social workers, group and residential supervisors Primary school teachers
Government officials Police inspectors
High
Transport planners and logistics staff
Teachers of vocational subjects in secondary education
Overall AI exposure

Radio and television technicians

Traditional and complementary medicine associate professionals
Retail sales associates Carers
Deck officers and pilots Childcare leaders and teaching assistants
Retailers and retail team leaders Specialized nurses
Physical and engineering science technicians
Jobs in 2026
Hospitality managers
Travel supervisors 100000
Medium Performing artists
Laboratory assistants Pharmacy assistants
Medical specialists
Life science technicians and related associate professionals
200000
Providers of other personal services Visual artists
Cashier associates Nurses (mbo) Production managers in agriculture, forestry and fisheries
300000
Photographers and interior designers Production leaders industry and construction
Hairdressers and beauticians Process Operators
Truck drivers Bus drivers and tram drivers
Security personnel Police and firefighters
Low Drivers of cars, taxis and vans
Cooks
Sports instructors Medical practice assistants
Call center associates Outbound and other vendors Auto fitters Machine fitters
Butchers Product inspectors Printing and crafts workers
Waiters and bar staff
Assembly workers Bakers Livestock growers Electricians and electronics fitters
Loaders, unloaders and stock fillers Concierges and cleaning team leaders
Mobile machine operators
Gardeners, market gardeners and breeders
Garbage collectors and newspaper deliverers
Very low Carpenters Construction Workers Farmers and foresters
Production machine operators Furniture makers, tailors and upholsterers

Kitchen helpers
Metalworkers and construction workers
Plumbers and pipe fitters
Construction workers finishing
Auxiliary workers in agriculture
Cleaners
Painters and metal sprayers
Auxiliary workers in construction and industry

Very low Low Medium High Very high

Probability of labour shortages
PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 30
 Figure 15 Industry distribution by labour shortage probability and overall AI exposure

Financial institutions
Information and communication

Specialized business services

Education
Public administration and services

Very high

Overall AI exposure
Health
Very low
High Low
Wholesale
Overall AI exposure

Welfare Medium
Other business sectors High
Culture, sports and recreation Very high
Metal industry
Medium
Transportation and storage
Other service activities
Total jobs
Energy 400000
Retail
800000
Chemical industry
Low Renting and other business support
1200000
1600000

Very low Construction

Food and beverage industry Other industries

Accommodation and food serving

Agriculture, forestry and fishing

Very low Low Medium High Very high

Probability of labour shortages
PwC analysis
Source: PwC analysis based
based on on data
data from from
Felten et Felten et and
al. (2023) al. (2023)
Bakensand Bakens
et al. (2021) et al. (2021)

PwC | GenAI@Work 31
 Figure 16 Industry distribution by the technology index and overall AI exposure

Financial institutions
R = 0.87, p = 0.00000017
Information and communication
Specialized business services
Education
Public administration and services
Very high

Health
High
Wholesale Welfare
Overall AI exposure

Other business sectors

Culture, sports and recreation Metal industry
Medium
Other service activities
Transportation and storage Energy
Retail
Chemical industry Renting and other business support
Low

Very low Construction

Other industries
Food and beverage industry
Accommodation and food serving

Agriculture, forestry and fishing

Very low Low Medium High Very high

Technology adoption index
PwC analysis based on data from Felten et al. (2023), Bakens et al. (2021), CBS and ROA
Source: PwC analysis based on data from Felten et al. (2023) and Bakens et al. (2021)

PwC | GenAI@Work 32
Acknowledgements
(alphabetically)

Authors:
Guntars Upis Chief Economist Office PwC Netherlands
Ricardo Ribas Santolim Chief Economist Office PwC Netherlands

Editorial team:
Astrid van der Werf Chief Economist Office PwC Netherlands
Bastiaan Starink Partner of Workforce PwC Netherlands
Jan Willem Velthuijsen Chief Economist PwC Netherlands
Mona de Boer Partner of Data & Technology PwC Netherlands

Research support and contributions:

Bote Scholtens Marketing & Communications PwC Netherlands
Brenda Verberne Marketing & Communications PwC Netherlands
Christian Eistrup Consulting PwC Netherlands
Danny de Gier Marketing & Communications PwC Netherlands
Femke Alsemgeest Marketing & Communications PwC Netherlands
Joon Yang Chief Economist Office PwC Netherlands
Jose Castillo Chief Economist Office PwC Netherlands
Nicky Roelofsen Clients & Markets PwC Netherlands
Marcel Jakobsen Chief Digital Officer PwC Netherlands
Martine Roelfzema-Uitermark Clients & Markets PwC Netherlands
Monique de Jong Marketing & Communications PwC Netherlands
Richard Volbeda Marketing & Communications PwC Netherlands

PwC | GenAI@Work 33
Endnotes

1 PwC (2023): Policymakers focus on making generative AI safer for 21 Santander (2023): Transformative potential of generative AI: 33 The Economist (2023): Lessons from finance’s experience with
all innovation, impact, and implications for the future workforce artificial intelligence
2 PwC (2023): Understanding the potential opportunities of 22 Webb (2020): The Impact of artificial intelligence on the labor 34 Wu et al. (2023): BloombergGPT: A large language model for
generative AI market finance
3 Felten et al. (2023): How will language modelers like ChatGPT 23 Eloundou et al. (2023): GPTs are GPTs: An early look at the labor 35 IBM (2023): How to improve your finance operation’s efficiency
affect occupations and industries? market impact potential of large language models with generative AI
4 Bakens et al. (2021): The labour market by education and 24 Strategy+Business (2023): Boston Dynamics wants to change 36 American Economic Review (2020): Processing the rise of
occupation until 2026 the world with its state-of-the-art robots quantitative investing
5 Agrawal et al. (2023): Similarities and differences in the adoption of 25 We measure the industry exposure to AI based on the proportion 37 The Economist (2023): Lessons from finance’s experience with
general-purpose technologies of workers in each occupation employed in each industry and the artificial intelligence
6 CBS (2023): Where do most people work? occupational exposures. This leads to industry exposure based on 38 Harvard Business Review (2023): What the finance industry tells
7 Bruegel (2023): A high-level view of the impact of AI on the workforce exposure. Some industries might have their business us about the future of AI
workforce models highly impacted by AI, but this would not be measured in 39 Bloomberg (2017): A sign of the times – how the finance industry
8 White House (2023): The impact of artificial intelligence on the our exposure scores. Take retail, for example. AI is already having a is embracing STEM professionals
future of workforces in the European Union and the United States of very big impact on how product recommendations are made, but this 40 FD (2023): Energy transition ‘unfeasible’ due to lack of personnel
America might not be captured in changes in occupational tasks. 41 CBS (2021): What are the most popular majors?
9 NOS (2023): Never happened in 50 years: more vacancies than 26 We measure labour market constraints using the probability of 42 UNESCO (2019): Beijing consensus on artificial intelligence and
unemployed employers achieving the desired composition of personnel according education
10 PwC (2022): PwC 2022 AI Business Survey to educational background within occupational groups, considering 43 Washington Post (2023): Artificial intelligence is already changing
11 Eisfeldt et al. (2023): Generative AI and firm values the supply-demand ratio for various types of training, based on how teachers teach
12 HAI (2023): Artificial intelligence index report 2023 Bakens et al. (2021): The labour market by education and occupation 44 Ibid.
13 NOS (2023): Never happened in 50 years: more vacancies than until 2026 45 Ibid.
unemployed 27 In Appendix B on page 29, we report the results of a similar 46 Khan Academy (2023): World-class AI for education
14 Eurostat: Job vacancies in number and percentage analysis, but this time focused just on the impact of overall AI. There 47 Education Week (2023): Will artificial intelligence help teachers or
15 PwC (2022): Unlocking the potential on the Dutch labour market are some small shifts, but the results are largely similar, indicating replace them?
16 Economic Policy Institute (2000): The link between productivity that the potential of overall AI comes largely from language modelling 48 UNESCO (2023): Generative artificial intelligence in education:
growth and living standards for many occupations. What are the opportunities and challenges?
17 Felten et al. (2023): How will language modelers like ChatGPT 28 We report the results for a grouping of 113 occupations. In 49 Marangunić & Granić (2015): Technology acceptance model: a
affect occupations and industries? practice, occupations and jobs are even more granular than that. If literature review from 1986 to 2013
18 Bakens et al. (2021): The labour market by education and we could look at very specific groups, probably a few would fall into 50 For more details, see Appendix A on page 22. We focus our
occupation until 2026 this category. analysis on indicators for which data is available at the industry level
19 For more details in the methodology, see Appendix A on page 22. 29 This will depend on the elasticity of demand and the income in the Netherlands, but other aspects have differentiated companies
20 Exposure does not directly predict employment or wage loss elasticity of demand, as increases in productivity can compensate for and industries that adopt technology faster. These include
but instead indicates where things could change in the economy the employment effects of automation. higher productivity, engaging in innovative activities, aligning the
and which workers are most likely to need to adapt. The exposure 30 Harvard Business Review (2023): AI won’t replace humans — but organisational structure with the given technology and information
scores capture the technical feasibility of AI and are limited in their humans with AI will replace humans without AI spillovers and higher R&D intensity. There are also numerous
consideration of other factors. Additionally, ‘least exposed’ to AI 31 Ibid. more qualitative adoption factors for technologies that are difficult
does not necessarily mean that that occupation escapes any impact 32 White House (2023): The impact of artificial intelligence on the to measure, such as user acceptance, habit change, trust, legal
of AI. Lastly, we also do not incorporate new jobs in our calculations future of workforces in the European Union and the United States of requirements and time to mass market saturation.
that could be created because of AI. America

PwC | GenAI@Work 34
51 CEPR (2023): Firms’ use of artificial intelligence: Cross-country 69 Acemoglu et al. (2022): Artificial intelligence and jobs: evidence 91 World Economic Forum (2023): Future of jobs report 2023
evidence on business characteristics, asset complementarities, and from online vacancies 92 OECD (2023): OECD Employment Outlook 2023: Artificial
productivity 70 Babina et al. (2022): Firm investments in AI technologies and intelligence and the labour market.
52 Kostøl & Svarstad (2023): Trade unions and the process of changes in workforce composition 93 PwC (2023): Global hopes & fears survey
technological change 71 OECD (2023): OECD Employment Outlook 2023: Artificial 94 WEF (2023): Future of jobs report 2023
53 NBC News (2023): Actors vs. AI: Strike brings focus to emerging Intelligence and the labour market 95 Bruegel (2023): A high-level view of the impact of AI on the
use of advanced tech 72 Autor (2015): Why are there still so many jobs? The history and workforce
54 White House (2023): The impact of artificial intelligence on the future of workplace automation 96 PwC (2023): Leaders guide to generative AI
future of workforces in the European Union and the United States of 73 Ibid. 97 PwC (2023): Global hopes & fears survey
America 74 Webb (2020): The Impact of artificial intelligence on the labor 98 Felten (2023): Occupational heterogeneity in exposure to
55 Babina et al. (2020): Artificial intelligence, firm growth, and market generative AI
product innovation 75 Johnson & Acemoglu (2023): Power and progress: Our thousand- 99 Kouretsis and Bampouris (2022): Crosswalks Between
56 Santander (2023): Transformative potential of generative AI: year struggle over technology and prosperity. Classifications of Occupations
innovation, impact, and implications for the future workforce 76 International Monetary Fund (2023): The power and perils of the 100 Due to crosswalking difficulties Biologists (SOC code 19-
57 Eisfeldt et al. (2023): Generative AI and firm values “Artificial Hand”: considering AI through the ideas of Adam Smith 1020), Door-To-Door Sales Workers, News and Street Vendors,
58 CEPR (2023): Firms’ use of artificial intelligence: Cross-country 77 Brookings (2023): The Turing Transformation: Artificial and Related Workers (SOC code 41-9091) and Radio, Cellular, and
evidence on business characteristics, asset complementarities, and intelligence, intelligence augmentation, and skill premiums Tower Equipment Installers and Repairers (SOC code 49-2021) were
productivity 78 Economist (2023): Machine dreams Your employer is ( probably ) dropped from the analysis before the SOC to ISCO crosswalking.
59 Babina et al. (2022): Firm investments in AI technologies and unprepared for artificial intelligence 101 Bakens et al. (2021): The labour market by education and
changes in workforce composition 79 CEPR (2023): The impact of artificial intelligence on growth and occupation until 2026
60 One limitation of this result is that our index is constructed as an employment
average of the scores on each dimension. It can be that a specific 80 Financial Times (2023): David Autor: ‘We have a real design
dimension has a very large impact on technology adoption in a choice about how we deploy AI’
given industry. If Culture, for example, is more important than other 81 International Monetary Fund (2023): The power and perils of the
dimensions, we are being overly optimistic about language modeling “Artificial Hand”: considering AI through the ideas of Adam Smith
AI adoption in Education and in Public administration and services. 82 Brynjolfsson & Raymond (2023): Generative AI at work
61 Eisfeldt et al. (2023): generative AI and firm values 83 Peng et al. (2023): The impact of AI on developer productivity:
62 Bruegel (2023): A high-level view of the impact of AI on the evidence from GitHub Copilot
workforce 84 Noy & Zhang (2023): Experimental evidence on the productivity
63 Acemoglu et al. (2022): Tasks, automation, and the rise in U.S. effects of generative artificial intelligence
wage inequality 85 Stack Overflow (2023): Developer sentiment around AI/ML
64 Lorenz et al. (2023): The future of employment revisited: how 86 Santander (2023): Transformative potential of generative AI:
model selection affects digitization risks innovation, impact, and implications for the future workforce
65 Autor (2015): Why are there still so many jobs? The history and 87 Financial Times (2023): David Autor: ‘We have a real design
future of workplace automation choice about how we deploy AI’
66 Brookings (2023): The Turing Transformation: Artificial 88 Abeliansky et al. (2023): Fostering a sustainable digital
intelligence, intelligence augmentation, and skill premiums transformation
67 Autor et al. (2022): New Frontiers: The origins and content of new 89 International Monetary Fund (2023): The power and perils of the
work, 1940–2018 “Artificial Hand”: Considering AI through the ideas of Adam Smith
68 Albanesi et al. (2023): New technologies and jobs in Europe 90 Ibid.

PwC | GenAI@Work 35
 Contacts

Bastiaan Starink Mona de Boer

Partner of Workforce Partner of Data & Technology
PwC Netherlands PwC Netherlands
E: bastiaan.starink@pwc.com E: mona.de.boer@pwc.com

Jan Willem Velthuijsen

Chief Economist
PwC Netherlands
E: jan.willem.velthuijsen@pwc.com

© 2023 PricewaterhouseCoopers B.V. (KvK 34180289). All rights reserved. PwC refers to the PwC network and/or one or more of its member firms, each of
which is a separate legal entity. Please see www.pwc.com/structure for further details. At PwC, our purpose is to build trust in society and solve important
problems. We’re a network of firms in 152 countries with nearly 328,000 people who are committed to delivering quality in assurance, advisory and tax
services. At PwC in the Netherlands over 5,300 people work together. Find out more and tell us what matters to you by visiting us at www.pwc.nl. This content
is for general information purposes only, and should not be used as a substitute for consultation with professional advisors.
36