This chapter examines some of the factors that are likely to change London’s economy and jobs base over the coming decades. It explores how three “disruptive” factors – automation, migration and low pay – may affect sectoral employment. However, it also looks at the opportunities created, London’s adaptability and at the strength of its sectoral clusters. The next chapter draws these findings together to take an overview of the challenges and opportunities for different sectors.

Automation

As the second machine age unfolds, the scope of technology — and of tasks that can be automated — is expanding rapidly, provoking both utopian and dystopian debates about the ‘end of work’. In thinking about the growth of automation, a two-by-two matrix can be used to map “routine” and “non-routine” roles onto the distinction between “cognitive” (knowledge work) and “manual” (physical labour) tasks. ³

Automation to date has largely been confined to routine manual tasks, and to cognitive tasks involving activities for which machines can be programmed using simple “If A, Then B” algorithms. The limits of automation were set by the potential of a task to be broken up into rule-based elements, the possible complexity of these rules, and the availability of data. “Non-routine” tasks like handwriting interpretation, which requires judgement and the ability to spot patterns in order to cope with almost limitless variations, were resistant to such approaches.

Big data — the production and analysis of large data sets — and improved pattern recognition technology has made such tasks much more well defined as problems, and more capable of automation. Additionally, machine learning allows a computer to take on elements of non-routine cognitive work, like programming itself (once a human has set parameters and specifications) or writing financial journalism stories based on company accounts.

The potential impact of these advances is significant for productivity, employment and daily life. Research by economist Carl Benedikt Frey and machine learning researcher Michael Osborne ⁴ suggests that about 47 per cent of US employment may be susceptible to substitution by technology in the next 20 years.

However, technological change does not operate in isolation, but interacts with other factors that influence the decisions of businesses and investors. Wage costs, labour availability, the price of technology, commercial viability, consumer preferences and regulatory frameworks all play a part. Wage rises in one sector can encourage automation, but the movement of displaced workers may reduce wages and thereby discourage automation in other sectors. Similarly, the deployment of ride-sharing services and autonomous vehicles is dependent not solely on technical feasibility and economic viability but ultimately also on regulatory approval. And if the business case for the capital outlay involved in automation does not stack up, the decision to invest will be deferred, particularly in times of economic uncertainty.

These advances could lead to fundamental changes in the types of tasks that can be automated, with machines becoming capable of undertaking complex cognitive tasks as efficiently as human workers — if not more so. Technology now offers:

• Scalability and pattern recognition: machine-learning algorithms are better able to store, process and analyse complex data sets. For example, Frey and Osborne cite Symantec’s Clearwell system, which “uses language analysis to identify general concepts in documents, can present the results graphically, and proved capable of analysing and sorting more than 570,000 documents in two days” ⁵ — far more quickly and efficiently than the junior lawyers and paralegals who would have previously undertaken this task.
• Reduced bias: human decisions are subject to biases and the influence of conditions that should not be relevant to the decision. Frey and Osborne mention research by Danziger et al. that showed judges being more generous in their rulings following a lunch break. An algorithm, by contrast, is designed to “ruthlessly satisfy the small range of tasks it is given”. This does assume, of course, that no biases are programmed in either intentionally or unintentionally. In the financial sector, for example, artificial intelligence algorithms can access and process more press releases and financial announcements than a human trader, and thus react faster — though automated trading has been blamed for moments of extreme volatility in the markets.

1. Automation is also extending to non-routine manual tasks. In the past, industrial robots have taken over many routine manual tasks, but now the scope of mobile robotics is being extended by three factors: Machine vision allows robots to recognise irregular layouts. This is necessary for (relatively) autonomous orientation and locomotion in unstructured, “messy” places such as roads, construction sites and homes — as opposed to the more predictable layout of a warehouse or factory floor.
2. High-precision dexterity, enabling more complex and delicate manual tasks to be undertaken (for instance in plant operations).
3. Prices, which are falling by 10 per cent annually, both for programming and installation costs and for the cost of the robot itself. ⁶

Nonetheless, significant constraints remain, at least in the immediate future. Frey and Osborne suggest that over the next two decades the susceptibility of occupations to automation will be determined by the extent to which they involve the following tasks, which are considered “engineering bottlenecks” and render automation less likely:

• Perception and manipulation tasks: while machine vision is improving, it is not yet capable of working in the most complex and cluttered contexts — for example in cardiovascular surgery.
• Creative intelligence tasks: the ability to bring to life ideas or artefacts that are “novel and valuable”, as well as creating “unfamiliar combinations of familiar ideas” in ways that require a “rich store of knowledge”.
• Social intelligence tasks: these involve social perceptiveness (being aware of others’ reactions and understanding the reasons for them), negotiation (reconciling differences), care, and persuasion (changing other people’s minds or behaviour).

This report uses Frey and Osborne’s methodology, recoding US occupational classifications to their UK equivalents, and then applying these to London’s workforce, using ONS data that shows employment by occupation (SOC 2010) and by industry (SIC 2007). This enables analysis of London’s economy on the basis of the occupational structure of different industrial sectors.

Figure 2 summarises the proportion of jobs in London that have the potential to be automated in the next 20 years, based on the occupational structure of the workforce. It divides jobs into those that have high (greater than 70 per cent), medium (30-70 per cent) and low (less than 30 per cent) potential for automation within the next 20 years.

The research found that almost exactly one-third of London jobs (33 per cent) are in occupations with a high potential for automation in the next 20 years.

Around one in five (19 per cent) have medium potential, and just under half (48 per cent) are in low-potential occupations. This report talks of “potential” rather than “probability” to reflect the fact that many other discretionary or external factors — such as employer preferences, regulations and wage levels — will determine whether any specific job is actually automated or not.

Table 1 shows the values for the individual occupations, grouped by skill level, that underpin these findings. A number at or near 1 shows an occupation for which all tasks are likely to be automatable in the next 10-20 years. A value nearer zero indicates that the role will be much less susceptible to automation over the same timescale.

High-skilled occupations are the least likely to be automated. There are some notable exceptions, including business and public service professionals.

These are associated with a medium probability for automation overall, reflecting significant variation within the category. Paralegals and legal assistants, for example are more susceptible to automation (reflecting the routine nature of much of their work) while many other occupations — e.g. public relations professionals — are less susceptible.

Middle-skilled occupations show a more mixed picture: administrative and secretarial occupations are associated with some of the highest potential for automation. However, care occupations have relatively low potential, as they encompass social intelligence and dexterity tasks.

Lower-skilled occupations have the highest potential for automation, reflecting a low degree of engineering bottlenecks in their respective activities.

The findings for sales occupations may look anomalous, given the social skills involved. However, the sales category includes some highly automatable tasks (such as till operation); other customer service occupations have less potential for automation. The assessment also works on the basis of what could be automated, so may not fully take into account the social intelligence that is valued but perhaps not strictly required in some roles.

Combining this assessment of the automation potential of different occupations with data on the makeup of different industries’ workforce in London enables an assessment of the automation potential of different economic sectors (Figures 3 and 4). Where the weighted probability is lower than 0.33, the industry concerned is shown in green; where it is between 0.33 and 0.66 it is shown in yellow, and where it is above 0.66 it is shown in orange. No sector shows a high automation potential overall on this metric, reflecting the mix of occupations found in most sectors in London — though it is the globally specialised sectors such as finance, information and communications, and arts and entertainment that have lowest potential.

These sectors also show the most marked difference from the UK as a whole — though London has a lower potential for automation in most sectors owing to its higher proportion of managers, professionals and other roles that involve creativity and social intelligence (Figure 4).

Migration

While automation’s impact will be universal, the impact of Brexit will be specific to the UK and particularly to London. With almost 37 per cent of its population born overseas, ⁷ London is one of the most mixed cities in the world, its cosmopolitanism both reinforcing and mirroring its economic position as a leading global city of the 21st century. 15 per cent of the workforce are overseas EU/EEA citizens, ranging from 5 per cent in public administration to 35 per cent in accommodation and food services (Figure 5). While we do not yet know the likely shape of a Brexit immigration policy, the expectation is that immigration controls will be tightened.

While the government has emphasised that they wish European workers currently in the UK to be able to remain, there is significant churn in the EU workforce, with 10,000 to 50,000 workers leaving each year, as discussed in Centre for London’s Open City report. ⁸ Future restrictions on freedom of movement could make it harder to replace these workers, particularly in low- to medium-skilled sectors where work permits may be harder to come by. In anticipation of Brexit, there is already a slowdown in European workers arriving in London for work, ⁹ and recent ONS data suggested that net immigration to London is also falling (though it remains positive).

It is worth noting that the three sectors most dependent on EU workers – construction, accommodation and food, and administrative and support services – are also among those with the highest potential for automation.

These sectors also offer relatively low pay and often have comparatively informal recruitment processes: consequently, they may find it more difficult to adapt to any new work permit regimes requiring minimum pay levels and applications in advance of arrival in the UK. This in turn may strengthen the business case for deployment of automation.

Low Pay

Low pay is a big challenge for London’s workforce today, pushing many working people into poverty, but measures to tackle it could accelerate the process of automation. Without cheap labour, businesses may rethink their operating model and the business case for automation may be strengthened.

Low pay can be defined in a number of ways. The Resolution Foundation uses two-thirds of median wages as a benchmark, ¹⁰ while the government’s “National Living Wage” — the national minimum wage — uses 60 per cent of median pay. It was set at £7.50 per hour in 2016/17, and is expected to rise to around £9 per hour by 2020.

Centre for London has argued for a separate minimum wage for London, around 20 per cent higher than the national rate. This would suggest a minimum wage of around £9 per hour today. ¹¹ Other measures, such as the voluntary Living Wage promoted by Citizens UK, focus on the costs of living, and have different rates for London and the UK. The current calculations are shown in Table 2.

Figure 6 shows gross hourly pay rates in London for the 10 per cent lowest-paid and 25 per cent lowest-paid workers. As the graph indicates, the National Living Wage, which applies to workers aged 25 and over, affects relatively few London sectors. However, several service sectors — retail, accommodation, food, administration and personal social services — have around 10 per cent of workers paid at or below this threshold (many of whom may be under 25). As the rate rises — it will be around four per cent higher from April 2018 — it will “bite” for more and more workers in these sectors. Recent media reports suggest that it is already having an effect on employment levels. ¹²

Many more workers are paid at or below the London Living Wage; in most sectors at least 25 per cent of workers are paid near or below this rate. If Brexit restrictions on immigration lead to labour shortages in these sectors, pressure for higher pay is likely to grow (though wages have remained stagnant in recent years — at a time of nearly full employment — suggesting that this is not guaranteed).

If wages rise as a result of labour shortages and policy interventions, this may in turn strengthen the business case for investment in automation as a substitute for more expensive labour.

As recent research analysis from the Institute of Fiscal Studies has indicated, the National Living Wage will rise to a point at which it is expected to affect employment levels. Some mid-level sectors are actually more routine and more easily automated than lower-skilled roles, so wage pressure in these sectors may lead to even swifter adoption of automation ¹³ (though London may be less rapidly affected than the rest of the UK as wage rates are higher to start with).

The timeframe for Frey and Osborne’s projections is 20 years. While Brexit negotiations are still underway at the time of writing, it is likely that the UK will leave the European Union within that timescale. Net migration has fallen since the referendum in 2016, while official figures put unemployment levels in London at 4.9 per cent — their lowest level since 1992. ¹⁴ Therefore, it may be that labour shortages begin to bite in some sectors before automation is technically and commercially feasible. The shortfall would most likely lead to wage inflation, and possibly price inflation where labour is a major element of costs. While this would offer low-paid workers a very welcome relief in terms of their living standards, it might at the same time further strengthen the case for and accelerate automation.

Automation has the potential to act as a significant disruptor to London’s economy, with Brexit and regulatory pressure on wages strengthening its impact. Together, these factors may dramatically change the employment profile of some sectors, encouraging and enabling management to substitute robots and algorithms for workers.

Adapting to these challenges will require changes to the way people, businesses, and the government conduct their affairs. But technological change will create new enterprise and employment opportunities too.

Opportunities and resilience

Disruptive change presents challenges, but also offers opportunities. London’s economy has shown itself to be resilient and adaptable to technological and economic change over time, so it should be well placed to adapt to new circumstances. This section will look at the opportunities for job creation, the skills profile of the capital’s workers, and how clustering and specialisation might provide resilience in the face of the challenges discussed above.

Job Creation

Technological change will create new jobs as well as replacing labour by automating existing jobs. New jobs will emerge in professions involved in the process of automation and digitalisation itself (for example in coding and development); from additional demand for services that are made more productive (such as financial advice); and from new products and services that are enabled by automation and increases in productivity (e.g. demand-responsive mass transit systems).

Research undertaken by Nesta ¹⁵ offers some insight into occupations that may experience increased demand as a result of the need for higher-order cognitive skills, interpersonal skills, and system skills. This research is based on the needs of automation processes themselves, as well as other trends such as urbanisation, demographic change, mitigation of climate change, socioeconomic inequality, political uncertainty and globalisation.

The authors of the report offer their assessment of the probability of increased demand for 92 different occupations as a result of these trends (1 being the highest possible probability value and 0 the lowest). This analysis suggests that sports and fitness occupations have the highest probability of increased demand in the future (0.745): in contrast, elementary storage occupations have a probability of only 0.061 for increased demand — the lowest value of all occupations (see Tables 3 and 4).

Table 5 uses this analysis to indicate the probability of increased demand at a two-digit SOC code level, weighted to reflect the composition of these occupational categories in London. All higher-skilled managerial, professional and associate professional occupations have a high probability of higher demand (> 0.5) — especially those requiring social and analytical skills such as teaching, health and science professionals. The middle-skilled occupations offer a mixed picture with some skilled manual trades and leisure-related personal services expected to see higher demand, while for the remaining occupational categories the probability is medium. Sales and customer service, process and transport operatives, and elementary occupations are associated with a medium-to-low probability of higher future demand.

As with the Frey and Osborne analysis of potential for automation, we have used these occupational data to show potential for job growth by industry (Table 6). Information and communication, financial and insurance, public administration, education and health, and (surprisingly) manufacturing are associated with a probability of increased future demand of 0.5 or greater, while all other sectors are below this threshold. The result for manufacturing might reflect the greater role of research and development in this sector in London, and the relatively specialised and skilled profile of the industry in the capital. The lowest values and therefore the lowest probability of increased future demand are in retail, and transportation and storage.

Skills and qualifications

All other things being equal, a more skilled workforce would be expected to be more resilient to changing circumstances, having the intellectual capacity to learn new skills and adapt existing skills to the changing demands of the workplace. ¹⁶

We can assess skills by reference to formal qualifications — GCSEs, A Levels, vocational qualifications, diplomas and degrees. London’s workforce is well qualified in these terms compared to the rest of the UK (see Figure 7). More than half (53 per cent) of London workers have a degree, compared to 31 per cent for the rest of the UK. The contrast is particularly strong in sectors where London has a strong concentration and higher productivity (see next section) — for instance in finance and insurance, where 71 per cent of the workforce in London is educated to degree level compared to 37 per cent for the rest of the UK. This suggests that workers in London will be more capable of taking advantage of new opportunities and adapting to economic and technological change.

The difference between London and the UK becomes even more pronounced when disaggregating qualification levels further for particular industries. Figure 8 charts the educational attainment of workers in the finance and insurance sector. Almost one-third of financial sector workers in London hold postgraduate qualifications, compared to seven per cent of those in the rest of the UK.

However, it is also worth asking whether these formal qualifications needed for jobs today reflect the skills needed to thrive in tomorrow’s workplaces, and adapt to the challenges of automation and digitalisation. Many young people currently attending primary and secondary school may acquire specific skills for traditional middle- and low-skilled occupations — only to see these jobs disappear over the next 20 years.

As we do not know precisely which occupations will be affected and when, “meta-learning” — the ability to learn and adapt through a working lifetime of more than 40 years — becomes as important as specific skills. Cognitive skills — information processing capabilities such as numeracy and literacy — are crucial for meta-learning. These form the basis for acquiring occupation-specific skills or functional skills in further and higher education and later in life.

A growing body of research suggests that cognitive skills matter more than formal qualifications. Research on standardised international tests in high-income OECD member countries comparing cognitive skills suggested that the attainment of formal qualifications “has very little impact on growth if it does not have substantial association with a better cognitive skills score”. ¹⁷

The OECD Programme for International Student Assessment (PISA) — which tests mathematics, science and reading skills among 15-year-olds — indicates that London does not compare particularly well with other city regions, including Hong Kong, Boston, Singapore, Vancouver and Shanghai. ¹⁸ Furthermore, comparing per capita growth rates with attainment demonstrates a fairly strong association between these two variables (Figure 9), including between cities in the same country, though London has a high level of GDP growth compared to other cities with similar PISA scores.

Data on adult skills derived from the OECD Programme for the International Assessment of Adult Competencies (PIAAC) — which tests numeracy and literacy skills among adults — also indicates a stagnation of cognitive skills in the UK between the younger and the older generations. ¹⁹ In contrast to almost all other OECD countries, the younger generations of 16-24 and 25-34 year-olds do not have better literacy and numeracy skills than the 55-64 year olds, despite being much more likely to have attended upper secondary schools and university. In the same study, London working-age residents performed just below the English average (see Figure 10) — roughly in line with the city’s overall deprivation profile.

So, while London’s population appears to be doing well when looking at educational attainment from a formal qualifications perspective, the city’s performance — compared to global competitors — in relation to standardised tests of cognitive skills is far less impressive. While the OECD PISA and PIAAC data is based on residence rather than workplace (thereby reflecting patterns of disadvantage among London residents as well as workforce skills), these findings suggest that London cannot afford to be complacent in terms of its skills profile.

Cluster strength

A further source of resilience should stem from the structure of London’s economy, the sectors that the city specialises in, and the extent to which these sectors demonstrate clustering (or “agglomeration”) that will make them relatively resilient to change.

Table 7 shows the industries in which London’s economic output (Gross Value Added) is concentrated compared to those of the UK as a whole; it also shows the industries where London has a greater proportion of managers and professionals. A value above 1 means that London has relatively more economic output, or more managers and professionals, when compared to the UK as a whole in that industry; a value below 1 means that London’s relative share of the same is lesser than the UK.

The match is not precise, but London’s global city functions — information and communication, financial and insurance services, arts, entertainment and recreation, and professional and technical services — are areas of significant specialism. These industries, which employ 1.8 million people and have accounted for nearly half London’s net job growth since 2000, are also sectors with a high share of managerial and professional staff (whose jobs are relatively resilient to automation owing to the creative and social intelligence skills required).

Firms in London are also more productive. Figure 11 shows London’s sector labour productivity as Gross Value Added per worker on the right-hand scale, indexed to the respective values for the UK as a whole on the left-hand scale (where a value of 100 means that London has the same productivity rate as the UK as a whole). Except for manufacturing, every sector either matches (in the case of real estate) or is more productive in London than in the UK as a whole. In the case of the finance and insurance sector, London is 50 per cent more productive.

Figure 12 brings these two datasets together to compare the sectors in which London’s economic output is concentrated with their relative labour productivity. As the chart shows, it is the same suite of “world city” functions that are both concentrated in London and more productive in London than in the UK as a whole.

These data suggest that London’s specialist sectors are benefiting from agglomeration economies. These make firms in some sectors more efficient and productive because they can “share, match and learn” with others: companies may draw on the same well-stocked talent pool to match their staffing needs, learn and innovate through personal networks, observe the competition, and even share infrastructure. ²⁰ This is confirmed by other research that suggests that agglomeration works much more powerfully for sectors such as professional, scientific and technical services than for wholesale and retail. ²¹

In summary, London specialises in sectors that are highly productive, and which have a high proportion of workers in managerial and professional functions. These jobs are less likely to be automated, and London’s relative productivity suggests that an agglomeration “stickiness” will preserve critical mass in the capital, notwithstanding the regulatory and market access challenges posed by Brexit.

That said, London also retains some economic diversity, although financial services dominate, they have not been London’s fastest-growing sector in recent years — providing further resilience against the sudden decline of one sector.