Does Working from Home Work?

Abstract

Remote work (“WFH”) was often the default mode of working during the recent pandemic, because of lockdown. But beyond this one-off effect, the question remains whether remote working will endure and become part of the “new normal”. We formalize a simple company-employee work-at-home decision model, which takes into account both worker preferences and the company’s strategic incentive to invest in supporting work-at-home practices. The model predictions are then tested on a large sample of global firms across the world, regarding their evolution in WFH intensity and how WFH changes correlate with labor productivity changes. We find that technologies facilitating WFH, and to a lesser extent, active human resources, are needed to make work from home more productive. Said otherwise, the future of WFH depends on how technology will be able to “augment” labor effectiveness.

1. Introduction

By 2022, Shopify, a well-known global next-generation e-commerce platform had secured a perennial flexible work arrangement scheme across its entire workforce, centered around the possibility of working from home (WFH). By that time, Accenture has already implemented an effective flexible work policy, with nearly 100% of employees working remotely. The significance of WFH was also clearly demonstrated during the peak of the COVID-19 pandemic, as social lockdowns pushed many firms and workers to resort to it. Looking at the period pre-COVID, WFH was limited, used by barely 15% of the European population [1].¹ In less than 6 months, during the lockdown of the pandemic, WFH quickly doubled to reach 40% of workers in both the United States and Western Europe [2].

But despite all claims, WFH is failing to automatically scale as a “new normal”. Not everyone is a Shopify’s, and even companies championing WFH are backtracking, such as Disney or Starbucks. Scholars themselves disagree, from the optimistic view of Barrero et al. [3] to the cautious views recently expressed by Harvard Professors Edward Glaeser and David Cluter on the need to go back to the office (2021).²

One reason for some limited WFH effects is simply that not all jobs are suited for WFH [4]. Another reason is that remote work practices have an ambiguous impact on labor productivity, as flexibility benefits are quickly bypassed by the sense of exclusion and by a lack of team performance spillovers [5, 6]. A final issue is organizational, whereby remote workers may hinder the performance of their team colleagues, given a lack of coordination [7]; especially a too high-level of WFH [8].

This research analyzes the trajectories made by large (mostly multinational) firms regarding WFH, pre- and post-COVID, with a goal to better understand the corporate drivers, especially productivity gains-linked to the adoption and expansion of WFH. The originality of the research is threefold. First, most recent studies on WFH productivity resort to workers’ surveys (e.g., [9, 10]). When studies take the pulse from the employer side, they have mostly relied on self-assessed productivity gains, e.g., Erdiesk [11], or have concentrated on very specific firm case studies [12, 13]. Here, we perform a statistical analysis of corporate revenue productivity changes and WFH density changes at the firm-level to derive estimates of the productivity benefit of WFH. Second, we lay out an analytic model of remote work as background to our empirical specifications—many studies rely on conceptual, rather than analytical models to define their empirical strategy, but they lack the insight that, for instance, complementary resources to remote work are critical to boost competitive advantage and boost WFH diffusion. Finally, our study covers a relatively large survey of 4000 firms, in ten countries—this is to be compared to the average sample of about 400 datapoints in the analysis of the 26 studies on productivity impact of remote work by Anapko et al. [14].

We confirm that WFH trajectories are different among firms, with the pandemic playing a catch-up, and most heavy WFH-centric firms exhibiting better revenue growth, but mostly to the extent that they have invested in complementary human resources and technology capabilities. Thus, WFH is not automatically part of “new normal”, but will clearly depend on the firm complementary capabilities in both HR and work technologies.

2. Remote work: evidence to date

WFH dates a long way back and was pushed by workers to limit time loss in commuting, and work–family conflict [15, 16]. Studies also highlight that WFH is an effective way to boost workers’ happiness, and their engagement [17]. According to Anapko et al. [14], the US Congress and the European Union had long approved legislation supporting remote work, but WFH remains rampant as technically challenging for years. For instance, large US firms resorted to remote work as a way to limit commuting during the two main oil energy crises in the 1970s.

The rise of connectivity and new digital technologies, including the internet, mobile, and the first applications of videoconferencing since this new century, have solved most of technical problems linked to WFH and boosted the rollout of more remote work practices. In this context, seminal studies such as Bloom et al. [12] or Anderson and al. [17] had confirmed that WFH could be a powerful tactic for firms, as it reconciles workers’ delight with productivity uplift.

More recently, the COVID-19 pandemic has become a major catalyst to resort to remote work as a main work practices. Results of new studies on the productivity of remote work, however, still raised skepticism as to whether WFH could generate labor productivity gains. For example, a study on UK workers by Mandall et al. [18] concluded that only a minority of UK workers could complete as much work during the first wave of the pandemic as during pre-COVID. In Japan, Morikawa [19] suggested that WFH productivity was only 2/3 of the level achieved at the workplace. Using a sample of 10,000 skilled professionals at a large Asian IT services company, which all shifted to WFH during the pandemic, Gibbs et al. [20] found that total hours worked increased by roughly 30%, but the average output did not significantly change, leading to a productivity drop in total of about 20%. Ipsen et al. [21] have concluded that a majority of Danish WFH workers could complete the same amount of work or more than when working at a workplace, but still 40–45% of them were less productive. Equivalently indecisive results were found for advancing countries such as Indonesia and the Philippines [22, 23].

The merits of those COVID-19 studies have been to highlight that productivity gains are not warranted but depend on a possible few key mediating factors. The first factor is organizational agility: in fact, most firms were possibly ill-prepared to deploy WFH as an urgent action out of lockdown. The second factor is that WFH productivity depends critically on the ability of workers to leverage technologies for effective WFH. In particular, a recent research by Bai et al. [24] demonstrates that digital technologies could boost WFH as well as enhance corporate performance. Tønnessen et al. [25] note that digital ability is not enough—rather digital knowledge sharing is the key driver of remote work on productivity.

Our research contributes to the debate by assessing the WFH trajectories of large firms across a worldwide survey performed by late 2021, at the time of COVID-19 pandemic got controlled by the rollout of effective RNA-based vaccines. The advantage of the study is that it measures WFH intensity at the level of the firm before and after the pandemic. This gives us insights into the dynamic pattern of WFH, which is largely missing in the literature. Second, we capture parallel data on both human resource practices and work technology investment spent that may act as clear mediators on how WFH deployment ultimately correlates with firm performance. In the next section, we lay out a simple analytic model of WFH intensity, that highlights the channels by which firm performance can be affected by remote work. We then continue with the empirical analysis.

3. A simple model of WFH optimal decision

3.1 Framework

We consider a representative worker working only for the focal firm, who values work and leisure equally. The worker receives a salary, w per unit of time worked, L, but her utility out of work is also shaped by the fact that she works remotely, or not. We note L _i = LP_i + LH _i, where.

LH _i (LP _i) is remote (on premises) labor. The worker utility is given by:

$${\mathrm{U}}_{\mathrm{i}}=\left(1-{\mathrm{L}}_{\mathrm{i}}\right)+\left(1+\mathrm{\gamma }\right).\mathrm{wi}{\mathrm{L}}_{\mathrm{i}}$$E1

Where 0 ≤ LH/_LP ≤ 1 is the share of work under remote, 0 < γ < 1 is the preference toward remote work. Regarding the firm, the firm sells a product, Y, with a product demand:

$${\mathrm{P}}^{\mathrm{\kappa }}={\mathrm{Y}}_{\mathrm{i}}$$E2

Where 1/κ is the absolute price elasticity.

As in task based models [26, 27], the product, Y, is made of the aggregation of m tasks in the range (0, 1) that could be done either by on premises or remote labor. Suppose further that tasks are ranked such that remote labor increase its competitive advantage for the high-end of tasks. Firm optimization implies that aggregating over all tasks, there is a portion, 0 ≤ 1/τ ≤ 1, where w.LH_i/Y_i = w.LP_i/Y_i, above which all tasks are done remotely.

As tasks are fully substitutive between remote and on premises, then the aggregation of tasks production function is equivalent to a Cobb-Douglas of the form:

$$\mathrm{Y}=\left(1+\mathrm{t}\right).{\left(\mathrm{\tau }.\mathrm{LP}\right)}^{1/\mathrm{\tau }}.{\left(\mathrm{\tau }/{}_{\mathrm{\tau }-1}\mathrm{LH}\right)}^{1–1/\mathrm{\tau }}$$E3

And (1 + t) ^(τ−1)/τ = (Y/_LH) as we normalize Y_i/_{LP i} = 1 [27].

The firm maximizes its profit, but strategically, has been investing complementary assets, I, upfront, before production with a rental cost, r, to secure the best benefit of remote work. Among others, I includes the collaboration software tools and video conferencing hardware to support connectivity and r is the rental price of I. We note:

$$\mathrm{Y}/\mathrm{LH}=1+1/{}_{\mathrm{\lambda }}.{\mathrm{I}}^{\mathrm{\lambda }}$$E4

and 0 ≤ λ ≤ 1, so that the lower λ, the higher the effectiveness of I to boost WFH productivity

Firm thus decides on I, then selects labor mix based on equilibrium with workers labor supply.

3.2 Equilibrium

As a subgame perfect equilibrium, the solution is found backward.

3.2.1 Final stage: labor demand stage

Thus, using (2) and (3) in the latest stage, the profit maximizing firm choses Y, such that:

$$\mathrm{Y}={\left(\left(1+\mathrm{\kappa }\right).\mathrm{w}\right)}^{1/\mathrm{\kappa }}$$E5

Further, following the production side, and the Cobb-Douglas, the labor demand is implictly given by

$$\mathrm{w}=1/{}_{\mathrm{\tau }}.\mathrm{Y}/{}_{\mathrm{LP}=}1/{}_{\mathrm{\tau }}$$E6

$$\mathrm{w}=\left(\mathrm{\tau }-1/{}_{\mathrm{\tau }}\right)\mathrm{Y}/{}_{\mathrm{LH}}$$E7

or

$$\mathrm{LP}/{}_{\mathrm{LH}}=1/{}_{\mathrm{\tau }-1}$$E8

The optimal labor supply implies:

$$\mathrm{w}=1/{}_{1+\mathrm{\gamma }}$$E9

which means that at labor market equilibrium:

$$\mathrm{\tau }=1+\mathrm{\gamma }$$E10

$$\mathrm{L}=\left({\mathrm{\gamma }}^2+1\right)/{}_{\mathrm{\gamma }\left(1+\mathrm{\gamma }\right).}\mathrm{LH}$$E11

And thus, the optimal split of labor is given by:

$$\mathrm{LP}/\mathrm{L}=\left(1-\mathrm{\gamma }\right)/\left(\text{γ2}+1\right);\mathrm{LH}/{}_{\mathrm{L}}=\mathrm{\gamma }\left(\mathrm{\gamma }+1\right)/\left({\mathrm{\gamma }}^2+1\right)$$E12

3.2.2 WFH investment stage

In the first stage, the profit including investment, I, in remote labor is given by

$$\mathrm{\Pi }=\mathrm{\kappa }.{\left(\left(1+\mathrm{\kappa }\right)/\left(1+\mathrm{\gamma }\right)\right)}^{1/\mathrm{\kappa }}-\mathrm{r{\rm I}}$$E13

As firm wants to maximize profit, first-order condition set to zero thus implies that:

$$\mathrm{\delta \Pi }/\mathrm{\delta }\left(\mathrm{Y}/\mathrm{LH}\right).\mathrm{\delta }\left(\mathrm{Y}/\mathrm{LH}\right)/\mathrm{\delta I}=\mathrm{r}$$E14

And from (4) and (14), one can compute that :.

$$\mathrm{\delta }\left(\mathrm{Y}/{}_{\mathrm{LH}}\right)/{}_{\mathrm{\delta I}}=\mathrm{\lambda }.\left(\mathrm{Y}/{}_{\mathrm{LH}}-1\right)/{}_{\mathrm{{\rm I}}}$$E15

$$\mathrm{\delta \Pi }/\mathrm{\delta }({}_{\mathrm{Y}/\mathrm{LH})=}\mathrm{\kappa }.\left(1+\mathrm{t}\right)$$E16

Inserting (15)–(16) into (14), one finds:

$$\mathrm{r{\rm I}}=\mathrm{\lambda }.\mathrm{t}.\left(1+\mathrm{t}\right).\mathrm{\kappa }$$E17

where from (3) and definition of t, we also have:

$$\left(1+\mathrm{t}\right)=\left(1-\mathrm{\kappa }\right).\left(1+\mathrm{\gamma }\right)/\mathrm{\gamma }$$E18

Integrating (18) into (13), final equilibrium profit is

$$\mathrm{\Pi }=\mathrm{\kappa }.{\left(\left(\mathrm{t}+\mathrm{\kappa }\right)\left(1+\mathrm{\kappa }\right)/\left(\mathrm{t}+1\right)\right)}^{1/\mathrm{\kappa }}-\mathrm{\lambda }.\mathrm{t}.\left(1+\mathrm{t}\right)$$E19

Using Eq. (18), we have thus demonstrated that:

Proposition 1: Firms have an interest to invest in complements to the extent that are able to boost remote above on premise remote work productivity.

Proposition 2: The incentive grows with product market power and workers appetite for WFH.

The proof of the above is done by assessing marginal profit and investment increase in function of the market attractiveness, κ, average productivity of (and preference toward) WFH, t (γ), as well as the degree of effectiveness of investment spent, 1/λ. Also, final profit reduces to κ. (1+ κ) ^1/κ if no gain prevails in selecting WFH over on-premises productivity, as in this case, the firm does not waste resources to invest in no WFH uplift.

4. Empirical research design

4.1 Empirical model

The previous section implies that firm revenue productivity, P.Y/L should be tied to the remote work intensity, especially in the function of remote work complements investments that signal that firms anticipate productivity uplift by shifting toward WFH. We can test this prediction with the following generic empirical equation, at the ith firm level:

$$\mathrm{\Delta }{\left(\mathrm{PY}/\mathrm{L}\right)}_{\mathrm{i}}={\mathrm{b}}_0+{\mathrm{b}}_1.\mathrm{\Delta }{\left(\mathrm{LH}/\mathrm{L}\right)}_{\mathrm{i}}+{\mathrm{b}}_2.{{\text{TECH}}_{\mathrm{i}}}^{\ast }\mathrm{\Delta }{\left(\mathrm{LH}/\mathrm{L}\right)}_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{i}}$$E20

Where u is an error term, Δx/_x is the difference in x measured before and post-COVID; the dependent variable is revenue per employee (PY/L), LH/L is share of remote labor., and TECH_i *Δ(LH/L)_i measures the technology complement to WFH, while the core hypothesis we want to test is that the coefficients: b1,b2 > 0.

The equation above is rather generic and is amended the following way. First, we adjust the equation to account for adjustment costs. The first is that productivity path out of investing in WFH has been recently shaped by the pandemic. The pandemic has indeed led to a urgent shift to WFH due to lockdown, which is not correlated to productivity, but rather to the simple ability to function [18, 19, 21]. If this overshooting takes time to absorb, it may weight on the positive dynamics of productivity gain post-COVID. We thus would include a variable, ADJUST, measuring how well the workforce has adjusted to the new digitally remote work induced by the pandemic.

Second, our model is especially concentrated on technology complements, but other elements may support new flexible works, e.g., such as much more inclusive human resources. As our survey discussion will allude to hereafter, we have measured the degree of HR involvement of firms around three key dimensions beyond purely financial, that is emotional balance, employability, purpose and relationship management. Those three dimensions are important as new working arrangements such as remote work require tools training (“employability”, [28]), induce new forms of interactions and bonding (“relationship management”, [7]), and move the emotional state like fear of exclusion (“emotional balance”, [17]). Following the literature for construct validity [29], we have built an index on those three dimensions based on a ranking from how firms made themselves accountable for those dimensions, on a Likert scale 1–7 (1 not at all, 4 neutral, 7 definite). Recentering on 4 = zero and scaled down by 9, we have built a variable, HR varying between (−1, 1), with −1 (1) meaning that the firm has been not at all (fully) accountable for those practices. Third, regarding technology investment, we separate between degree of deployment (TECHD = 0, no deployment, 1, pilot, 2 = scaling, 3 fully scaled) of remote collaborative technologies to support WFH and degree of investment on those tools (TECHI =0, no investment, 1 = Moderate, 2, Average, 3 = significant). This separation allows to separate input (investment) from output (deployment) as indicator of technology support to remote work.

Based on the above, the amended equation becomes:

$$\begin{array}{l}\mathrm{\Delta }{\left(\mathrm{PY}/\mathrm{L}\right)}_{\mathrm{i}}={\mathrm{a}}_0+{\mathrm{b}}_1.\mathrm{\Delta }{\left(\mathrm{LH}/\mathrm{L}\right)}_{\mathrm{i}}+{\mathrm{b}}_2.{{\text{TECHD}}_{\mathrm{i}}}^{\ast }\mathrm{\Delta }{\left(\mathrm{LH}/\mathrm{L}\right)}_{\mathrm{i}}\\ +{\mathrm{b}}_3.{{\text{TECHI}}_{\mathrm{i}}}^{\ast }\mathrm{\Delta }{\left(\mathrm{LH}/\mathrm{L}\right)}_{\mathrm{I}}+{\mathrm{b}}_4.{{\mathrm{HR}}_{\mathrm{i}}}^{\ast }\mathrm{\Delta }{\left(\mathrm{LH}/\mathrm{L}\right)}_{\mathrm{i}}+{\mathrm{b}}_5.{\text{(ADJUST)}}_{\mathrm{i}}\\ +{\mathrm{b}}_6.{\mathrm{Z}}_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{i}}\end{array}$$E21

We also add a large set of control (vector Z) that are typically known to affect labor productivity outside WFH. We include both a measure of digital technology maturity (DK), and R&D intensity (RD), as both types of capital is also associated with higher labor productivity [30, 31, 32, 33, 34]. The information collected concerns the use of 10 digital technologies: cloud, mobile, broadband access, edge computing, IoT, robotics, machine learning, computer vision, advanced RPA, NLP/NLG. The last four are AI-related. Our measure of maturity is the index from 0 to 10, where each technology used for commercial use is coded 1, 0, 5 if adopted but only used at stage of experimentation, and 0 otherwise. This kind of construct has been used and validated e.g. by Lee et al. [32].

We also include company size, scope (B2C versus B2B versus B2B2C), as well as the fact that companies are, or not, global multinationals. Industry and geographies fixed effects are included, capturing elements like product market features affecting revenue potential, workers skills, and occupation types that affect the technical feasibility of WFH [4].

Our main hypothesis extends to b₁, b₂, b₃, b₄ > 0, and Eq. (21) is estimated both by OLS and mostly by Instrumental variables, given the endogeneity of LH/_L, and the WFH cross-effects. We use the pre-COVID WFH intensity variable as first instrument. The second instrument is the industry weight at the leave-one-out mean. The method works similarly as a generalized moments method [35].

4.2 Sample

Eq. (24) relies on a top executives’ survey designed by a major global consultancy, and administered online by an independent agency, by the fall of 2021. The survey covers a large set of firm practices during three periods (pre-, during-, and post) COVID-19 lockdown pandemic. The survey focuses on large firms, across 10 key countries (Europe: France, Germany, UK, Italy, and Spain); North America (US and Canada) as well as in APAC, China, or Japan. It also covers 17 NACE-2 aggregate sectors. Appendix provides more background on the sample, as well as its representativeness. The full sample comprises 4015 firms, with 2/3 of firms generating 1–10 US billion by end of 2020. Large firms have the advantage to diffuse new practices faster than others [3, 36]. The average company grew revenue by 9% cumulative over 2018 to 2021, thus including the acute phase of the covid-19 pandemic, and 3 years change in labor productivity has been 4.7%, for companies generating EBITDA margin in the range of 15% and return on capital employed of 6% during the 3 years.

WFH intensity in the sample is reported to have been 16.5% by 2018, growing to 24.6% post-COVID. Table 1 provides the distribution of the dynamics of WFH densification among the sampled companies. If there has been a net increase in the portion of WFH practices, the most visible shift has happened at the bottom of the distribution, with low WFH intensive firms catching up, possibly, as per the necessity out of the lockdown during the COVID pandemic. More importantly, we notice that about one-third of companies have also reduced their WFH intensity post-COVID to a level (lower than) before pre-COVID.

WFH intensity	Before COVID	After COVID
Less than 10%	34.8	13.2
Between 11 and 25%	43.4	44.2
Between 26 and 40%	15.3	29.5
Between 41 and 75%	5.1	10.3
Over 75%	0.6	2.0

Table 1.

WFH density (%) distribution in sample, pre- and post-COVID.

Source: survey, authors computation, based on total sample, N = 4015 firms

Regarding remote work technologies, Table 2 shows that about 70% of large companies in the sample had rolled out digital collaborative tools for enterprise use, but only 30% of firms were exploiting those technologies at scale across the whole enterprise. This state of use is in line with other surveys (e.g., [3]). We observe quite some variance by sectors, with sectors such as retail and healthcare being more limited in the rollout at scale, as lots of activities still are made in face-to-face in those sectors.

Stage	Automation (%)	WFH tools (%)
Not started	9	6
Pilot	42	22
Scaling in the organization	32	42
Scaled up in whole enterprise	17	30

Table 2.

State of work technology diffusion large global firms, pre-COVID.

Source: survey, authors’ computation, based on total sample, N = 3930 firms (balance stands for firms for which the technology has no application for their business.

As already mentioned, the absorption by employees of those new technologies may take time. Table 3 reports how employees have been able to settle, even enjoy, the new digital environment (variable, “ADJUST”). Half of the respondents claim that the digital remote shift has been fully absorbed, for about 15% of companies that are still struggling with remote tech absorption.

Employees feeling about digital remote routine:
	Settled (%)	Enthusiastic (%)
Strongly disagree	0.40	0.70
Disagree	1.20	2.00
Neither	13.20	10.40
Agree	35.50	40.80
Fully agree	49.80	46.20

Table 3.

Workflow integration of WFH technologies, 2020, large global companies.

Finally, Table 4 provides a view of HR practices beyond administrative pay, and linked to employability, emotional balance, and relationship/team management. On average, firms are starting to deploy those HR practices, but it remains largely unsettled.

	Not at all (%)	Not really (%)	A bit (%)	Partially (%)	Growing (%)	Almost (%)	Completely (%)	Index (%)
Emotional	0.8	1.4	11.6	26.7	28.8	21.7	9.0	9.0
Relational	0.4	0.8	9.3	25.0	32.3	23.5	8.8	10.0
Employable	0.2	0.8	5.8	18.9	29.5	31.9	12.9	14.0

Table 4.

HR practices, 2020, large global companies.

Index computed with data presented on partially (4), completely = 1; not at all = −1.

Regarding control variables, the survey does not have quantitative measures of investment in digital technologies but provides a split between stage of use, between experimentation and the level of commercial exploitation. The ratio of exploration to exploitation of AI technologies is roughly 65/35 (see also Zolas et al. [36]). Concerning R&D, the average firm spends close to 4.5% of revenue on R&D, or above the typical range of firm [37].

4.3 Results

Table 5 reports the results with fixed country and industry effects included (but not reproduced for clarity). The first two columns report the OLS, and then the last two concern IV estimations. For each type of estimation, we present first the simplest equation without the cross-effects, then including the cross-effects. Regarding the IV, we have performed the first stage regression linked change in WFH with WFH pre-COVID, and “leave the focal firm out” means WFH change at the dyad industry/geography level. We have confirmed the IV relevance for the WFH variable, either through the Hansen J test (P = 0.02), or Stock and Yogo’s F-test (F > 18.2). We then used the fitted value from this first stage to build the new variables WFH and cross-effects as second stage regression shown in Table 5.

Variables	.(1) OLS		.(2) OLS
	Coefficient in %	Stdev	Coefficient in %	Stdev
Constant	3.01***	1.22	3.55*	1.87
WFH (points %)	0.21	0.14	0.19*	0.09
ADJUST (in %)	n.a.	na	2.92***	0.43
TECHD*WFH (points %)	n.a.	n.a.	0.71	0.39
TECHI*WFH (points %)	n.a.	n.a.	0.79*	0.41
HR*WFH (points %)	n.a.	n.a.	1.09**	0.41
RD (points %)	0.42*	0.18	0.37*	0.16
DK (percentage points)	1.02	0.86	1.22*	0.66
SIZE (in billions)	−0.4	0.78	.−0.39***	0.11
B2C+	.−3.12*	1.66	.−3.02*	1.63
B2B2C+	-0.35	0.67	.−0.69**	0.34
Firm is global+	2.54**	1.22	2.21***	0.67
industry dummy	Y		Y
country dummy	Y		Y

Variables	.(3) IV		.(4) IV
	Coefficient in %	Stdev	Coefficient in %	Stdev
Constant	3.09**	1.56	3.28***	1.08
WFH (points %)	0.25	0.16	0.18	0.11
ADJUST (in %)	na	na	3.51***	0.78
TECHD*WFH (points %)	n.a.	n.a.	0.99***	0.28
TECHI*WFH (points %)	na	na	0.83**	0.32
HR*WFH (points %)	n.a.	n.a.	0.74*	0.41
RD (points %)	0.21*	0.12	0.24**	0.1
DK (percentage points)	0.98*	0.46	1.08**	0.49
SIZE (in billions)	.−0.49***	0.15	.−0.52	0.37
B2C+	.−3.02*	1.63	.−2.01*	1.08
B2B2C+	.−0.69	0.45	.−0.32*	0.18
Firm is global+	1.03***	0.37	1.33***	0.68
industry dummy	Y		Y
country dummy	Y		Y

Table 5.

Productivity and work from home change, global firms.

Note: Adjusted R-square are respectively 0.225; 0.412; 0.264; 0.452; variables are measured as difference post- and pre-COVID, or a window of 3 years; ***p < 0.01, ** p < 0.05, *;. +: default is B2B; country: US and industry: hightech.

Regarding controls, a F-test on the joint significance of industry effects, as well as country effects, show that both effects are relevant at 1%. Regarding other firm control productivity variables, global firms exhibit better labor productivity growth than other firms (about 2 point of nominal labor productivity for 3 years), but this effect cancels out when accounting for firm size, as global firms exhibit larger revenue than average, and higher revenue firms tend to have lower labor productivity growth (minus 5 points over 3 years) in this sample. We find that digital technology penetration as well as R&D are associated with higher labor productivity growth (for a joint marginal effect of about 1 to 1.5 points over 3 years). Those effects tend to be lower than the early literature on the effects of digital technology on labor productivity growth, e.g., [30].

Those effects tend to be lower than the early literature on the effects of digital technology on labor productivity growth, e.g., [30].

Coming to our various hypotheses, on the effect of WFH, a right tail F-test, is strongly significant at 1%, confirming that labor productivity change is positively linked to the rollout of WFH practices, as well as technologies/human resources complements supporting those practices, in line with our model predictions. Looking variable by variable, we, however, find that the effect of WFH intensity on labor productivity growth is not significant per se; what matters are really the cross-effects, as well as the ability of employees to absorb those digital technologies for remote working practices.

The above clearly settles the case that WFH is a relevant positive trend as claimed by Barrero et al. [3], but this is a profitable trend for corporations to the extent that employees can get used to this new type of environment, and that firms invest in enough technology and human practice complements to support the shift, to get returns in the form of higher labor productivity.

From the sample, we do not have access to level of the investment made to technology that favors remote work, but the fact is that technology spent and absorption to support effective WFH are a large contributor, at the sample mean, bringing from 6 to 7 points in labor productivity growth over 3 years. This uplift is relatively large to secure a strong return to technology investment made. Finally, besides technology, organization matters. Workers must adapt to those new digital remote technologies to deliver performance. The fact they do not fully adjust to those new digital remote work practices costs about 1 points of productivity growth. Likewise, firms that have a strong proactive human resource approach that alleviates the transition costs to remote work will support a better productivity gain said otherwise, technology augments labor effectiveness.

Regarding the estimates, a firm that manages HR to account for all positive elements of emotional balance and employability, and invests at scale in remote collaborative tools to support the best context for remote work, might generate 3 points of extra labor productivity, versus a firm not willing to invest and support WFH, after controlling for company size, profitability and industry/location. This type of estimate is relatively in line with some of the most favorable remote work case studies reported in the literature such as Bloom et al. [12] for an online Chinese travel portal, or Choudhury et al. [13] during their investigation of a shift from work from home to work from anywhere at the US patent office. Finally, in our sample, about 2/3 of companies has boosted WFH practices since pre-COVID, thus, the net effect should be clearly positive on the economy [38].

5. Conclusion

Working from home is part of the known trend toward more flexible work arrangements. While the pandemic has accelerated the shift out of premises work, the fact is that the trend is not necessarily natural, and will only materialize if there is common incentives by employers and workers to commit to WFH. Despite some high profiles of companies pushing for full flexibility, there are as many brand names reverting back from extensive work-from-home, questioning who will actually buck the trend. The originality of this research is to consider the firm’s viewpoint, as being the ultimate investment maker regarding the adoption and rollout of those technologies. We also test the impact of technology adoption among a large sample of more than 4000 global firms and across 15 countries.

We formalize a task-based model of the firm that only invests in technology and human resources to the extent they can complement WFH productivity uplift. The prediction tested on the full sample, suggests that WFH must interwork with technology and HR support to augment labor productivity gains. We find that the effect of WFH boosts labor productivity on average for our sample of large global firms. The effect is positively linked to the rollout of technology tools, suggesting that the question of the future of WFH is closely linked to the future of technology deployment to complement work tasks [3, 39].

The above study has some notable restrictions. First, regarding sample, the data are survey-based and variables such as WFH are reported as range. Second, company respondents are not fully identified and thus we cannot match the survey with other external sources. Third, data does not give any insights of workers skills and capabilities. This is a critical missing piece of information as the skill distribution will shift both the potential for the technical opportunity of WFH and the ability of workers to quickly absorb and master new technologies. The way we control for skill distribution is through industry control, but this may not be enough, and significant difference may exist at firm-level. Fourth, the sample concerns large global companies, and results may not necessarily be generalizable to smaller firms. Finally, productivity performance is measured by the end of 2021, or post-pandemic lockdown. As shown by the significance of the ADJUST variable, companies may still be partly influenced by the effect of the pandemic, and productivity effect may only arise longer-term.

Acknowledgments

The authors thank Accenture Research for access and some support work on the data. All errors are our own.

Appendix

See Tables A1 and A2.

Revenues change	2 years before COVID-19 (%)	6 months COVID-19 (%)	End of 2021 (%)
More than 10% decline	0	2	1
Between 5% and 10% decline	1	10	4
Between 0% and 5% decline	4	20	27
No change	5	21	13
Between 0% and 5% growth	46	25	24
Between 5% and 10% growth	37	19	22
More than 10% growth	6	3	9

Table A1.

Evolution revenue (pre-COVID = 2018, post-COVID, end of 2021).

Measures	Metrics	Sample	Benchmarks
Financials:	Profit change Q3-Q1,2020	Minus 28% worldwide	Minus 35% for the S&P 500
	Profit growth precovid, 2019/18	7.1% worldwide	7.3% worldwide publicly quoted companies
	profit growth post covid, 2022/21	6.1% worldwide	6.4% worldwide publicly quoted companies
Capabilities
Innovation	innovation spent decline 2020	Minus 8%	Minus 5–10% [5]
Agility	companies versed in agile method	65%	70%, KPMG, 2019
Digitization	Share of tech used	55% for European sample companies	> 50%, EIB, 2020
Sustainability	% ESG engagement US	83% for US sampled companies	70%, Harvard Law School
Work from home	% employees working from home	17% in top 5 European countries,2018	13% same scope, WEF, 2017
		27% in top 5 European countries, 2022	25% same scope, Eurostat, 2022
		19% in US, 2018	14%, firms more than 1000 people; PEW, 2017

Table A2.

Sample representativeness.

Sources: Author based on KMPG, World Economic Forum, Damodoran web site; EIB, McKinsey, Eurostat, Harvard, Pew project.