One size does not fit all: Constructing complementary digital reskilling strategies using online labour market data

Abstract

Digital technologies are radically transforming our work environments and demand for skills, with certain jobs being automated away and others demanding mastery of new digital techniques. This global challenge of rapidly changing skill requirements due to task automation overwhelms workers. The digital skill gap widens further as technological and social transformation outpaces national education systems and precise skill requirements for mastering emerging technologies, such as Artificial Intelligence, remain opaque. Online labour platforms could help us to understand this grand challenge of reskilling en masse. Online labour platforms build a globally integrated market that mediates between millions of buyers and sellers of remotely deliverable cognitive work. This commentary argues that, over the last decade, online labour platforms have become the ‘laboratories’ of skill rebundling; the combination of skills from different occupational domains. Online labour platform data allows us to establish a new taxonomy on the individual complementarity of skills. For policy makers, education providers and recruiters, a continuous analysis of complementary reskilling trajectories enables automated, individual and far-sighted suggestions on the value of learning a new skill in a future of technological disruption.

Technological change: A call for reskilling

Technology is changing the way we work, fundamentally. In economic literature, the periodic warning that automation and new technologies were going to terminate large numbers of jobs is an “evergreen” (Acemoglu and Autor, 2011; Brynjolfsson and McAfee, 2014; Frey and Osborne, 2017). In contrast to recurring fears of mass unemployment, current literature shows that the (digital) technology revolution, rather automates tasks than entire occupations (Autor, 2015). Technological and social transformation change the skill composition of professions (Acemoglu & Autor, 2011). The work that is thereby eliminated has different skill requirements than the newly created jobs, which leads to the paradox of simultaneous unemployment and labour shortage (Autor, 2015). Professional service or admin—“white collar”—jobs are particularly exposed to this trend, which is “hollowing-out” the middle employment spectrum of the labour market (Baldwin and Forslid, 2020). The strategies to reskill are often unclear, as the future benefits of learning a new capacity are uncertain and precise skill requirements for mastering emerging technologies remain opaque (De Mauro et al., 2018).

A conventional policy response, aligning national education systems with changing labour market demand, is increasingly ineffectual as technological and social transformation outpaces national education systems (Collins and Halverson, 2018). Likewise, large employers are struggling to keep their workforces' skills up to date (Illanes et al., 2018). At the same time, the COVID-19 pandemic has tightened company budgets, forced employees to work remotely, and facilitated the global need for reskilling (Stephany et al., 2020a). Workers have begun to assume greater personal responsibility for reskilling, via online courses, distance education tools, and entrepreneurial approaches to work (Allen and Seaman, 2015).

Most recently, just-in-time skills development, motivated by perceived market shifts, has emerged, as formal training courses are unaffordable for some workers (Kester et al., 2006). In addition, female participation is still hindered by cultural aspects in traditional science, technology, engineering, and math (STEM) education (Kahn and Ginther, 2017). Instead research shows that independent professionals, including women, prefer informal, digital, social learning resources like Stack Overflow and tutorial videos to develop new skills (Yin et al., 2018).

From songs to skills: The rebundling paradigm

In the field of reskilling, the individualistic approach of workers to learn new skills in a fragmented, incremental, and digital fashion has, once again, fostered one of the defining paradigms of the Internet: rebundling (McManus et al., 2018). One of the first sectors to witness the powerful emergence of rebundling was the music industry. First, in the early days of the commercial Internet, download platforms allowed music lovers to access songs individually without having to acquire the band’s entire album. The single item (song) was unbundled from the original bundle (album). Later, at a second stage, streaming platforms, like Spotify, reversed the trick by allowing the (re)bundling of previously unrelated items. Users could listen to songs from different artists for one single price. The mastery of this strategy has made digital entertainment companies superstars’ firms (Eriksson et al., 2019).

Similarly, this paradigm has affected the way we learn new skills. Initially, digital technologies allowed education providers to offer topical online courses (Wulf et al., 2014). In a second stage, e-learning platforms like Udacity performed the rebundling and offered a whole set of topical courses for one single price (Bates, 2019). The acquisition of individual skills (programming in Python or designing a logo) has been detached from its original domain of training (studying informatics or graphic design).

Online labour platforms: Laboratories for skills

In fact, newest findings show that independent IT professionals today develop new skills incrementally, adding closely related skills to their existing portfolio, as argued by Lehdonvirta et al. (2019). Their work examines the skill development of freelancers on online labour platforms (OLPs). These platforms are websites that mediate between buyers and sellers of remotely deliverable cognitive work (Horton, 2010). ¹ OLPs can be further subdivided into microtask platforms, e.g. Amazon Mechanical Turk, where payment is on a piece rate basis or freelancing platforms, such as UpWork, where payment is on an hourly or milestone basis (Kässi and Lehdonvirta, 2018). In light of the COVID-19 pandemic and its significant economic repercussions across industries (Stephany et al., 2020a), OLPs continue to increase in popularity due to a general trend of work at distance (Stephany et al., 2020b).

In this role, OLPs might have become early “laboratories” for the de- and rebundling of incremental skills. It could be argued that, for work, OLPs have turned into what streaming providers are for music: Freelancers can sell previously unrelated skills in one single portfolio for one hourly price, as illustrated by Figure 1. The role of the Data Scientist is a prime example of the rebundling of skills from different domains, i.e., visualisation, programming and statistics, as an economically profitable offer.

Figure 1.

Music streaming platforms allow users to rebundle individual songs from different artists for a monthly charge (lhs). Via online labour platforms, previously unrelated domain specific skills appear now rebundled under a new profession, e.g. Data Scientist (rhs).

The general success of the Data Scientist skill bundle is only one of many examples for the profitability of skill rebundling and cross-skilling strategies; i.e., the combination of skills from different occupational domains. Anderson (2017), for example, shows that freelancers with diverse skill portfolios are able to gain higher wages, on average. Similarly, Stephany (2020) illustrates that the acquisition of a new skill from a different, but adjacent skill domain is related to higher asking wages of online freelancers, while the benefit of the individual capacity depends on its complementarity with the skill bundle it is attached to.

The network perspective: Establishing new skill taxonomies

In the situation of rapidly changing skill requirements, systematic oversight is essential. However, individuals often lack foresight into which skills are rising and falling, which skills are most valuable, and, most importantly, which skills their existing portfolio is complementary to. They might get locked into path dependencies that may result in dead ends that prevent them from reskilling into new areas (Escobari et al., 2019).

The work by Anderson (2017) or Stephany (2020) shows how OLP data allows us to monitor skill rebundling in a global workforce by near real-time with up-to-date skill bundles on a granular level. They use the rich toolbox of network analysis for the characterisation of skill relationships. Given a sample of freelancers with multidimensional skill portfolios, the authors construct a human capital network in which skills are nodes and two skills are connected by a link if a worker has both. This skill network provides the researchers with an endogenous categorisation of skills based on their relationship in application and the context dependency of human capital, as illustrated by Stephany (2020) shown in Figure 2.

Figure 2.

In a skill network, 3525 skills are connected if jointly advertised by the same worker. Skills group in eight clusters with different degrees of centrality (node size). Seven clusters emerge: 3D design (top left), admin support (middle right), audio design (bottom left), data engineering (top right), graphic design (middle left), legal (bottom right), Software and technology (top middle), and translation and writing (bottom middle). Recoloured source: Stephany (2020).

Most importantly, the taxonomies emerging from the analysis of large-scale OLP data allow researchers to investigate the complementarities of learning. One and the same skill can inflict different costs and leverage different benefits depending on the learner’s existing skill set that it complements. With the example of programming languages, Stephany (2020) illustrates that learning Java is of limited economic benefit in the field of Data Engineering where learning how to programme in Python, on the other hand, increases worker wages significantly. For the field of 3D design, however, the picture flips and Java yields much higher contributions to worker wages than the allegedly ‘super star’ programming language Python (Grus, 2019). Similarly, complementary costs in learning can be reduced as previously acquired skills may lower the entry barrier into new skill domains, e.g., via the underlying similarity of language logics across programming languages.

Pathways into the future: Sketching reskilling strategies

The global workforce is under constant pressure to reskill, as technological change accelerates and task automation reshuffles occupational skill requirements. Labour market mismatches must be avoided but traditional reskilling, via national education policies, is too slow for the fast pace of technological change and precise skill requirements for emerging new technologies are too fuzzy. In addition, economic tightening due to situations like the COVID-19 lockdown further accelerates digitalisation trends while forcing workers to reskill from home. Workers have already taken matters into their own hands by incremental and just-in-time skill development via social learning tutorial and online education providers.

The first markets witnessing this individualistic re-skilling trend are OLPs. Here, entirely new skill compositions are offered, at times, with very profitable margins for workers. In fact, OLPs have become early “laboratories” for the de- and rebundling of skills from previously unrelated domains. They allow us to observe the rebundling of skills into yet unnamed occupations. The statistical analysis of OLP data holds a tremendous opportunity for revealing endogenous categorisation of skills via skill networks and for giving us insights into the value of complementary reskilling. First investigations reveal that diverse and innovative skill compositions can be extremely profitable for some workers, like Data Scientists (Anderson, 2017; Stephany, 2020).

The complementarity in costs and benefits of learning new skills make reskilling decisions a complex and highly individual endeavour with no single best choice for everybody. The analysis of OLP data can help estimating the complementarities of learning a new skill, which is key for finding individual and sustainable, i.e. continuously profitable, reskilling pathways. It supports establishing a new taxonomy of digital re-skilling opportunities so that job market mismatches can be reduced, at scale. In cooperation with freelance platforms and online education providers, researchers and policy makers should consider using this blueprint to give learners a personalised recommendation on which skills to develop complementing their own capacities. A continuous time series analysis of complementary cross-skilling trajectories (Stephany, 2020), via OLP data, enables automated, individual, and far-sighted suggestions on the value of learning a new skill in a future of technological disruption.