Digital Technology Uses among Microenterprises Why Is Productive Use So Low across Sub-Saharan Africa?

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


Policy Research Working Paper 10280
This paper explores the use of digital technologies, their association with performance outcomes, and the main constraints to greater use among microenterprises. The study uses a sample of more than 3,300 firms across seven Sub-Saharan African countries, of which over 70 percent are informal and over half are self-employed enterprises with no full-time workers. The analysis finds that productive use of digital technologies is low: less than 7 percent of firms use a smartphone, less than 6 percent use a computer, and roughly 20 percent still do not use a mobile phone. Even fewer firms use digital tools enabled by these access technologies: among firms with smartphones, less than half use the internet to find suppliers, and only half with a computer use accounting software or inventory control/ point-of-sale software. Women are less likely to use all digital technologies than men. A greater range of uses based on internet-enabled computers or smartphones relative to uses based on 2G phones are conditionally associated with higher job levels. However, there may be a tension between higher productivity and more jobs: the highest productivity firms are not generators of the highest jobs, and vice versa. That formal high-sales and high-jobs firms are more strongly associated with the use of internet-enabled tools than high-productivity firms suggests that relaxing constraints preventing the latter from using more such digital tools and expanding sales and jobs could be important. Among these constraints, more than seven in ten non-users indicate that lack of attractiveness ("no need") is the main impediment to productive use of digital technologies. The most important conditional correlates of smartphone and computer adoption are related to having a loan, having electricity, having business linkages with large firms as customers, and managers having vocational training. This paper is a product of the Africa Chief Economist Unit and the IFC Economic Policy Research Unit. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/ prwp. The authors may be contacted at mdutz@ifc.org.

Introduction
This paper presents findings on the adoption and use of digital technologies (DTs) by microenterprises across Sub-Saharan African (SSA) countries. The paper explores associations between enterprise and entrepreneur (owner) characteristics, use of basic 2G-2.5G feature phones versus computers and internet-enabled 3G-4G smartphones and more specialized DTs enabled by internet, and business outcomes linked to productivity, sales, export status and the generation of more and better jobs. It explicitly focuses on informality and the extent to which top-performing firms include both informal and formal firms. The findings are based on a sample of over 3,300 firms across seven SSA countries, namely Ghana, Kenya, Mozambique, Nigeria, Senegal, South Africa, and Tanzania. Over 70 percent of enterprises are informal, with over half of the sample being selfemployed household enterprises with no full-time paid employees.
The paper provides new evidence, both with respect to associations between adoption of various DTs and firm performance and with respect to correlates of adoption. With respect to adoption, the paper explores the correlates of adoption of smartphones and computers while controlling for firm level characteristics such as the degree of informality, being in an urban versus rural location, access to finance, the level of education of the owner, and whether the owner has any vocational training.
As indicators of output-related performance outcomes, the paper uses labor productivity, sales, and whether the firm has customers abroad (a proxy for exports). The paper also measures the performance of enterprises against the objective of creating more and better jobs: as a measure of more jobs, the number of full-time employees plus owners generated by each firm, and as earningsrelated measures of better jobs, wages per full time workers and profits per owner. The latter variable is likely to be especially important to assess whether entrepreneurial firms generate sufficient profits to support the well-being of their owners.
The paper explores the associations between these performance variables and the use of access technologies as well as additional more specialized DTs enabled by access to internet. The distinction between adoption of an access technology such as smartphones and the use of specialized DTs is important: a smartphone can be used by firm managers and workers in ways that have no productive bearing on firm performance, such as watching videos, shopping online or engaging in communications with friends. Only when access technologies are used for purposes relevant for business performance is a positive association with performance to be expected. Fortunately, the data provide a rich set of information on various types of DTs as well as different uses of internet, unavailable elsewhere, especially for the type of largely informal micro firms covered in the survey. The paper explores the use of DTs that are especially useful for the firms' external transactions with upstream and downstream product providers and consumers, financial intermediaries, and labor markets: for example, whether the firm uses internet to look for suppliers and better understand and market to customers, to access the financial system, or to recruit workers, in addition to more basic uses of non-internet-enabled mobile phones for communication and mobile money. The paper also explores uses of DTs to enhance internal-to-firm transactions, including to improve the internal management of the firm, namely by using accounting and inventory control/point of sales (POS) software. It thereby explores some dimensions of management practices that are related to those that have been shown to improve firm performance (e.g., Bloom et. al. 2014Bloom et. al. , 2012.
The data are available only as a single cross-section. The findings are restricted to statistically significant associations and do not generally allow inferences to be made about causality, neither with respect to adoption of technologies nor with respect to their impact on business performance. However, a detailed description of the incidence of DTs among microenterprises and an assessment of statistically significant associations between firm characteristics, use of DTs and firm performance-both unconditional and conditional-provide valuable inputs to enrich the existing understanding of the micro-entrepreneurial environment in SSA countries. This analysis also highlights the desirability to generate and analyze better panel data in the future.
Some of the key hypotheses with respect to use of DTs and business performance, and correlates of adoption that are tested in the paper are: • Specific internet-enabled DTs that provide services beyond access are expected to have stronger positive correlations with performance outcomes relative to adoption of smartphones, and in turn relative to use of non-internet enabled DTs. DTs that are used to reduce the cost of operations and improve efficiency within the firm (such as accounting or inventory control/point of sales software) are expected to be more strongly positively correlated with better firm performance relative to DTs that are useful for firms' external transactions. This is because DTs used primarily within the firm are less subject to network effects while those that are especially useful for firms' external transactions can generally only be effective if similar DTs are extensively used by firms' external counterparts. • There is expected to be considerable heterogeneity among informal and formal firms, with the best-performing informal firms as well-performing as the best-performing formal firms and as intensive in the use of relatively sophisticated DTs. • Adoption of smartphones and computers is expected to be positively correlated with businesses located in urban areas, having access to electricity, higher levels of owner education, and having access to finance. Firms with younger owners are more likely to adopt smartphones. Firms with a female owner are expected to have a negative association with smartphone adoption, based on widespread evidence of a gender gap in digital access.
The main findings of this paper include: • Few microenterprises across SSA countries use internet-enabled access technologies. Less than 7 percent use a smartphone, less than 6 percent a computer, and roughly 20 percent still do not use any mobile phone. • Even fewer firms use digital tools enabled by these access technologies. Only two in five firms use a mobile phone to communicate with suppliers or customers, and less than one in five use them to pay suppliers or receive payments with mobile money. And among those with smartphones, less than half use internet to find lower-cost or more favorable suppliers, and less than a third use internet for ecommerce. Only half of firms with a computer use accounting software or inventory control/POS software. • Across all DTs, women are less likely to use them than men. The largest divides between men and women are in the use of computers (with men 3.3 times more likely to use them), using internet to find suppliers (2.4 times more likely), and using accounting software, inventory control/POS software, and the general use of smartphones (each 2.3 times more likely). • A greater range of more sophisticated DT uses based on internet-enabled computers or smartphones relative to DT uses based on only 2G phones are conditionally associated with higher job levels. There is a positive progression in the number of internet-enabled DTs associated in turn with higher productivity, sales, and job levels. Six internet-enabled and three non-internet-enabled DT uses are the only significant conditional correlates of higher job levels. Firms using accounting and inventory control/POS software are associated with a roughly 1.6person larger firm size (0.47 and 0.44 log points, respectively) than those not using them. • There may be a tension between higher productivity and increased employment. Firms characterized by the highest productivity are not generators of high employment, and vice versa. While this tension between high-productivity, skill-intensive technologies (automation) and jobs has been emphasized among SMEs and large enterprises, this is the first evidence of such tension among microenterprises. 2 • Formal "scale" (high sales and jobs) firms are more strongly associated with the use of internetenabled DTs than formal and informal high-productivity firms. A relevant policy research question is whether alleviating constraints in greater use of DTs among high-productivity firms could enable them to expand their scale in line with their higher productivity. • Over seven in ten non-users indicate that lack of attractiveness ("no need") is the main constraint to productive use of DTs. A relevant policy question is whether this is because there are no apps available that are useful to them in their local language that meet their productive needs, because their general skill level does not enable them to understand how they could productively use available apps, or because the available quality of service is so poor that access to internet is not useful to them. • The most important conditional correlates of smartphone and computer adoption among available variables are related to having a loan, having electricity, having business linkages with big firms as customers (being embedded in value chains that require upgrading), and having relevant skills (managers having vocational training).
The rest of the paper is organized as follows. Section 2 presents an overview of the relevant literature. Section 3 presents an analysis of general characteristics of the data as well as digital divides by owners' age and gender. Section 4 explores whether the uses of DTs matter in terms of their association with business performance outcomes. Section 5 explores differences and similarities in performance outcomes across formal and informal enterprises to better understand the heterogeneity across microenterprises in the available SSA countries. Section 6 explores why productive use is so low, despite the strong positive associations between uses of DTs and performance outcomes. A final section concludes with suggestions for further research work.

Literature review
The literature on firms' uses of DTs focuses on two general questions. First, what are the economic consequences of DT adoption and use on performance dimensions such as productivity, production, export orientation, employment, and the skills composition of employment? Second, what are the drivers of adoption for productive use? The literature applied to emerging markets so far has largely focused on the economic consequences of adoption of smartphones. There is an important followon issue regarding what specific productive purposes broadband internet is used for, which includes the use of a variety of more specialized DTs enabled by broadband-which this paper seeks to explore.
The recent literature has provided evidence that the adoption of DTs by enterprises generally leads to reductions in various types of costs and frictions (e.g., search, replication, transport and monitoring costs as well as networking and organization costs). These efficiencies, in turn, are expected to generate better economic outcomes, including higher productivity, higher sales, and better and more jobs (Goldfarb and Tucker, 2019). However, the materialization of these benefits may require some time as well as complementary investments McAffee, 2014, Brynjolfsson et. al 2020). There is also evidence that the impact of DTs exhibits heterogeneity across different types of technologies as well as across localities. For example, DeStefano et. al. (2019) find that while conditional correlations between broadband adoption and various firm performance variables are positive and significant, there is no causal effect of broadband on plant exit, productivity, sales growth, employment or employment growth. They find weak evidence of impact on sales, but this finding is not robust to variations in the specification of econometric approach. The finding of a positive causal impact of DT adoption on productivity is also not universal.  Dutz, Almeida, and Packard (2018) show that low-skilled workers can also benefit from the more intensive use of internet due to the output expansion effect from increases in productivity and consequent lower prices: while firm investments in information and communication technologies (ICT) capital and firm-level use of faster internet result in a substitution effect whereby some low skilled workers are replaced by the new technology, a sufficiently strong output effect results in a net increase in the use of low skilled labor as well as higher skilled labor. More generally, as long as adoption of DTs expands production volumes and does not totally eliminate the need for lowerskilled workers including through new tasks, it results in more jobs over time, including both higher and lower-skilled jobs. The output expansion effect requires sufficient responsiveness of demand to the lower prices, which is more likely in lower-income countries where demand for most products is still far from being satiated and therefore more likely to be highly price-elastic (Bessen 2019); the output expansion effect is additionally facilitated through exports. Iacovone et al. (2016) show that over the period 2008-2012, firms that increased ICT use in Mexico had higher growth in productivity. They also find that firms that face higher competition and make more intensive use of ICT have a better performance. Fernandes et. al (2019) find that the rollout of internet (captured by the availability of internet interacted with intensity of use) in China increases a firm's likelihood of exporting and its export-output ratio. Further, rollout of internet also results in increased total factor productivity and use of labor. They also find that increased availability and use of online services leads to a higher increase in exports in industries more reliant on an online presence.
Work on SSA includes Hjort and Poulsen (2019), who examine the impact of availability of faster internet made possible by the arrival of submarine cables on jobs by skill and education levels in twelve countries. 4 They find that arrival of fast internet causes an overall increase in the probability of holding a job as well as an increase in the probability that an individual will hold a skilled job. The probability of holding an unskilled job does not decrease. Most important, an examination of the impact of faster internet arrival on workers by educational attainment rather than by job categories highlights that the estimated increase in the employment rate is of comparable magnitude for those with primary school, secondary school, and tertiary education in all samples studied. Three subsequent country case studies -on Nigeria, Senegal, and Tanzania -have also added to this rapidly growing evidence base by exploring the impact of mobile internet availability (3G or 4G coverage) -instead of fixed terrestrial broadband -on jobs and welfare. The studies take advantage of geospatial information on the roll-out of mobile internet towers over time combined with at least two rounds of household data over a six to seven-year period. The job estimates from Nigeria show that internet availability had positive impacts: labor force participation and wage employment increase by 3 and 1 percentage points respectively after three or more years of exposure in areas with internet availability relative to those with no coverage (Bahia et al., 2020). The job estimates for Tanzania are similarly significant, with richer jobs data allowing a more detailed exploration (Bahia et al., 2023). The job estimates for Senegal are also positive for some job categories: 3G availability is associated with 5 percent higher formal employment than non-covered areas, though availability does not have a significant impact on overall employment .
The answers to the second question, namely drivers of adoption, generally emphasize two factors: first, the business environment, including policies affecting the degree of competition, and second, firm characteristics, and in particular firm capabilities. For example, using data for 25 industries in 25 European countries and a difference-in-difference methodology, Andrews, Nicoletti and Timilotis (2018) find that both market incentives (for example, administrative burdens on start-ups and barriers to entry in services, both measured from the OECD product market regulation index, and barriers to exit and reallocation, as measured by their employment protection Index) as well as firm capabilities (low managerial quality, lack of ICT skills and poor matching of workers to jobs) explain cross-industry-country variations in adoption of various DTs such as enterprise resource planning and customer relationship management software, and cloud computing.
A more recent categorization focusing on developing countries has emphasized that the main factors associated with the adoption of smartphones and computers by enterprises, as well as of more sophisticated DTs relying on these access technologies, are related to ability to pay and willingness to use, as well as other factors: 5 Ability to pay for DTs • Affordability of DTs, namely the ability to pay for DTs, is linked to both their costs (relative to expected enhanced returns to enterprise revenues) and access to finance, including the 4 For a review of the recent literature on Africa, see "Impacts of digital technology use on jobs and poverty" in Begazo, Blimpo and Dutz (2023, forthcoming), and on developing countries more broadly, see Hjort and Tian (2021). 5 For a review of the theoretical and empirical literature on technology adoption and diffusion, see Comin, Cirera and Cruz (2022), chapter 6. For adoption drivers of DTs including with network effects (like digital platforms), see Cusolito (2021). The SME digital lag in OECD countries is attributed to similar factors though with different relative emphases, including insufficient capital and missing complementary assets (ability to pay), lack of information and awareness and skills gaps (willingness to use), see OECD (2021) and references therein. extent to which local financial systems enable financing of the underlying access technologies, associated equipment, and software. • Affordability of complementary infrastructure can be interpreted as part of the ability to pay of users-as the affordable availability of internet is facilitated by the affordable availability of complementary non-digital infrastructure. This includes the cost, availability and reliability of electricity, as well as roads and logistics for certain types of apps requiring associated transportation services.
Willingness to use DTs • Attractiveness, namely the underlying demand and willingness to use DTs, is affected by the availability of information on the existence of DTs, by how easy it is to use them, and by whether they meet the productive needs of users. Whether the DTs meet the productive needs of users, in turn, is affected by whether they are designed for the skills level of the user, in the language that the user speaks, and whether they facilitate on-the-job learning. It also is affected by required quality of service levels of internet connectivity such as sufficiently high speed (how fast data moves), low latency (no discernible lag in requested items to appear) and reliability (always on in a stable manner). Importantly, attractiveness is also affected by the uncertainty of the economic benefits of adoption associated with risk/loss aversion preferences and biases in the behavior of individual users, including factors such as over/under-confidence, status quo bias, misinformation, and trust. • Capabilities affect the willingness of enterprises to use DTs. Capabilities include both people skills and enterprise technological capabilities. People skills include the ability of workers, managers, and owners to extract value from DTs, and is affected by levels and quality of basic education, and follow-on technical and vocational education and training, as well as experientially based know-how. Enterprise capabilities include the need and ability of firms to use DTs, and are linked to the complexity of inputs, production processes, and quality of outputs required by markets, including managerial and organizational practices, and to the extent of technology use, search, and research & development activities undertaken (enabling the enterprise to accumulate relevant know-how assets beyond those that can be bought through markets). Given the low literacy rate coupled with low quality of education across most African countries, DTs can serve as a means of knowledge diffusion to improve the quality of education and bridge skill shortages in Africa to facilitate a higher productivity growth and job creation.
Other factors • Other elements of the business environment, including relative availability and cost of skilled labor and specific types of capital, and competition incentives to spur technology adoption and generation of new technologies, are linked to access to wider domestic and international markets and to level-playing-field government regulations (e.g. that avoid favoring incumbents and impeding entry or expansion, and rather enable any start-up to enter and experiment, expand rapidly, and exit and re-enter if the initial business idea was not well executed and did not sufficiently respond to market needs). • Socioeconomic factors include whether social norms and rules make ownership of access devices difficult for women.
Smartphones, computers, and several other DTs have network effects, namely the value of the DT is increased with the number of users including those accessed through digital platforms. Network effects lead to additional factors affecting the uptake and use of DTs, including: • Coordination issues, including wait-and-see behavior relative to others in the network, and either weak or lack of interoperability and equal access across markets (linked to standards and regulation). • Trust issues, linked to possible shortcomings at the government legal and regulatory levels (e.g., lack of appropriate liability laws, online consumer protection, e-signature, personal data protection, cybersecurity), and at the platform operator level (e.g., third party certifications, simple dispute resolution procedures).

Data
This section presents the data and main patterns of use of DTs by microenterprises. The first subsection presents general characteristics of the data. The second sub-section presents differential uses of DTs by microenterprises owned by men and women, disaggregated further by younger versus older owners.

General characteristics
The Research ICT Africa (RIA) After Access business survey is focused on micro-size enterprises. The RIA survey was implemented during 2017-18. The data set analyzed in this paper includes 3,325 observations covering seven Sub-Saharan African countries, namely Ghana, Kenya, Mozambique, Nigeria, Senegal, South Africa, and Tanzania. 6 The RIA survey also covers Rwanda and Uganda, but sufficient missing data on variables of interest have precluded its inclusion in this paper. 7 There is considerable heterogeneity across enterprises according to their sector(s) of operation (Table 1). 8 Almost two-thirds (63 percent) of surveyed microenterprises are active in trading activities and almost one-quarter (24 percent) are active in other services. 9 Almost one in ten firms (9 percent) is active in producing agricultural products, 10 while less than one in twenty (4 percent) is manufacturing products.
Formal enterprises are defined as legally registered either at the national registrar general or country-level revenue authority, at the local authority or municipality, or both. Informal enterprises have neither national nor local registration. Most firms are informal: more than two-thirds (73 percent) are informal, with the largest share of informal firms being in agriculture (85 percent). Almost half of firms (41 percent) providing other services are formal.
Most surveyed firms are in urban or peri-urban areas (62 percent), as that is where most selected enumeration areas are located. However, there is significant variation: almost seven in ten firms manufacturing products and providing other services (68 percent each) are in urban or peri-urban locations, while a majority of farms (55 percent) are in rural locations.
Most microenterprises are relatively young, with the median firm in operation for four years. However, there is considerable variation in firm age, with some farms in operation for well over ten years, bringing the average up beyond ten years (10.3 years). Trading firms are on average the youngest. In terms of age of the owner, there is not much difference between the mean and the median age, suggesting that a roughly similar number of owners are younger or older, with a sizable number of owners less than the median age of 36. Finally, almost half of firms (44 percent) are women owned. 11 Firms in agriculture are the most likely to be owned by women (49 percent), those in manufacturing the least likely (29 percent).  Table 2 presents variables related to each enterprise's soft assets. They include the aptitudes and capabilities of the business owner-managers, as well as access to electricity, finance, and more sophisticated upstream and downstream business linkages, disaggregated by sector of operation. Whether the owner is a "transformational" entrepreneur is a proxy for their inherent aptitude for productive entrepreneurship: owners self-select as transformational entrepreneurs due to the profit-making opportunity that owning a business provides as opposed to a necessity or subsistence choice to supplement earnings or because there is no preferred wage job available. Less than three in ten owners (28 percent) are transformational. The highest share of owners with transformational aptitude are providing other services (35 percent), while the lowest share are traders, overwhelmingly subsistence entrepreneurs. 11 Firms with several owners are designated as women-owned if more than half of owners are women. Firms that have equal numbers of men and women owners are designated as being owned by men. There are 300 such firms (corresponding to about 9 percent of the sample). Of these, 116 report zero owners for both men and women; moreover, these are observations for which data on labor productivity are missing, so they are not included for most of the statistical analysis. Further, 175 observations have one male and one female owner, 5 firms have 2 of each, and there are three enterprises with 8, 9, and 10 male and female owners, respectively. The number of years of formal schooling and whether the manager holds any vocational training certificates are proxies for the capabilities of the owner-manager. The median manager has only completed primary education (6 years of schooling), highlighting the low general level of education among firm managers. Across sectors, the highest average level of schooling is attained by managers of providers of other services, though this level (9 years) does not even reach completion of secondary schooling. Low levels of education are complemented by relatively low levels of vocational training, with less than one in five managers having a vocational training certificate (18 percent). The lowest share of managers with vocational training certificates is in agriculture (10 percent), while the highest is in other services (33 percent).
Access to other productive assets is also relatively low. Almost half the firms (44 percent) report not having access to electricity. Firms providing other services have the highest access to electricity (73 percent), no doubt linked to the fact that almost 70 percent are in urban or peri-urban locations (though this does not hold for firms manufacturing products). Conversely, less than one in three firms in agriculture (31 percent) have electricity, indicating that not only the majority of rural farms but also a number of farms in urban and peri-urban areas do not have electricity. Regarding access to financial assets, what is striking is how few microenterprises have ever had a loan, just 3 percent of firms overall. Manufacturing firms have the largest access to loans (6.5 percent of firms), double the share of remaining firms, but still very low. On the other hand, it is non-manufacturing firms, and especially farms and providers of other services, that have the highest share of firms that have ever had a line of credit or credit facility from any supplier (19 percent), though this is also relatively low. These agricultural lines of supplier credit are no doubt linked to the purchase of seeds, fertilizers, and pesticides, and possibly for a lucky few to the provision of farm insurance.
Finally, regarding access to learning opportunities from more knowledgeable business partners upstream and downstream in the firm's value chain, though large suppliers in principle offer the largest scope for learning, just 16 percent of microenterprises have large formal businesses as their main suppliers. This indicates the low levels of integration of microenterprises in formal value chains, with over four out of five farms having either informal businesses or small formal businesses as their main suppliers. Even fewer firms have access to other forms of linkage-related learning: less than 2 percent of firms overall have their most important upstream suppliers located abroad, just over 4 percent of firms overall have big enterprises as downstream customers, and just over 3 percent of firms overall have non-local customers situated beyond surrounding towns and villages. In addition to the overall share of firms benefiting from such business linkages being low, there is not much variation across sectors of operation in these opportunities of learning: just over 5 percent of manufacturing firms have business linkages with foreign suppliers, just over 8 percent have big firms as customers, and just over 5 percent have non-local customers (together with providers of other services). Note: The table is based on unweighted data. It reports shares of all firms within each category, except for the manager's years of schooling variable. Entrepreneurs are labeled "transformational" if they answered "My own business pays more than being employed" to "What was the main reason to start a business for you?" while the remaining (subsistence) entrepreneurs answered either "To make additional money to my salary" or "Otherwise I would have been unemployed". Manager's education is measured in years based on: "What is the highest level of education of the business manager?" The coding is as follows: "None" =0, Primary=6, "Secondary"=13, " Tertiary: Diploma /Certificate"=15, "Tertiary: Bachelors"=16, "Tertiary: Masters"=19. For vocational training, respondents answered yes to "Do business owners have vocational training certificates?" Firms with electricity are based on a yes/no answer to "Does the business premises have electricity?" "Had loan" is in response to "Has the business ever had a business loan from a bank?" "Supplier credit" is in response to "Does the business have a line of credit/ credit facility with suppliers?" Large suppliers are firms that responded "large formal businesses" to "Who are your main suppliers?" "Foreign suppliers" are firms that responded "abroad" to "Where are your most important suppliers located?" with the other possible responses being "locally (surrounding towns and villages)" and "from all over the country". "Big firm customers" are firms that responded "big enterprises" to "Who are your customers?". "Non-local customers" are firms that responded "from all over the country" to "Where are the most important customers of the business located", with the other possible response being "locally (surrounding towns and villages)." Table 3 reports unconditional means of the use of specific DTs by businesses, disaggregated by sector of operation. A first set of variables describes access technologies, namely whether the firm has a basic 2G or 2.5G featurephone, or a mobile broadband-enabled smartphone (3G or 4G). It also includes whether the firm uses one or more computers as well as whether the firm has a websiteboth are also access technologies, with websites being a more general DT to get the attention of upstream and downstream companies, end-use customers, financial institutions, workers, and government. Across all firms, 73 percent still use a 2G or 2.5G phone that does not provide broadband internet access. It is striking that less than 7 percent of firms use a smartphone, less than 6 percent use a computer, and roughly 20 percent still do not use any mobile phone. The lowest users of smartphones are farmers and traders, with less than 4 percent of farms and less than 5 percent of trading firms using them. Even fewer farms and trading firms use computers, less than 3 and 4 percent, respectively. Finally, websites are even less used among microenterprises, no doubt less useful relative to social media apps such as WhatsApp for getting the attention of other users.
Two other sets of variables cover more specialized DTs for external-to-firm transactions and internalto-firm functions enabled by the adoption of access technologies. The use of external-to-firm transactions is largely driven by the lower search costs, cheaper and expanded market coordination, and lower transportation costs enabled by these DTs. However, they require both sides of the transaction to have a mobile phone. On the other hand, internal-to-firm management tools that reduce costs and enable users to enhance their management skills, namely accounting and inventory control/point of sale (POS) software, do not require another external party to also use these tools; an exception is using mobile money to pay the firm's workers, as the workers also require a digital device to receive payment. Of eight DTs for upstream transactions with suppliers and downstream transactions with customers, five uses are possible with a 2/2.5G phone, namely communicating by voice or SMS with suppliers or customers, advertising via SMS, and making or receiving payments via mobile money. The other three uses facilitate more interactive learning but require a smartphone to access internet to search and find suppliers or better understand customers, and for selling and purchasing products online through ecommerce platforms. Information on the use of any phone for mobile banking and on the use of smartphones for online banking, and on the use of internet to recruit workers is also provided as part of DTs facilitating external-to-firm business functions.
A key finding is how few microenterprises overall use digital tools for productive uses. Even though four out of five firms have a mobile phone (80 percent of firms use either 2/2.5G or 3/4G phones), only two out of five (42 percent) use these digital devices to communicate with suppliers or customers, and less than one out of five use their phones to pay suppliers (15 percent) or receive payments (18 percent). And even among those with smartphones, while most use internet to understand customers (roughly 6.1/6.5 or over nine out of ten), far fewer use internet to find lowercost or more favorable suppliers (2.9/6.5 or just over four out of ten), and even fewer use internet for ecommerce (1.8/6.5 or under three out of ten). The overall use of management tools to learn and boost business performance is somewhat higher, though still only about half of firms with a computer use accounting software (3.0/5.8) or inventory control/POS software (2.7/5.8).
Across sectors, providers of other services followed by manufacturers are the largest users of smartphones (13.3 and 10.3 percent, respectively) and computers (12.0 and 10.3 percent), and in turn are the largest users of more specialized DTs. While farms and trading firms lag in most uses of DTs, over one in five farms use their mobile to receive payments (20.9 percent, or over 30 percent of farms with any mobile phone), ahead of manufacturing firms. This suggests a possible entry point for broader use of DTs over time, as well as providing a digital transaction record for future financing based on a credit score (linked to histories of their output sales and input purchases), especially promising for all smaller firms that lack sufficient traditional collateral or resources from family and friends.

Table 3. Uses of DTs
Note: All responses are shares (%) of firms responding based on unweighted data. "Use 2/2.5G mobile" is derived from "yes" responses to "Does the business manager have a mobile?", irrespective if it is for private, business use or both, and "no" responses to "How does the business access internet: Mobile broadband (3G/4G, wireless)." Smartphone users answer "yes" to this question. "Use computer" is a non-zero response to "How many computers does your business have?" Reported answers to "What do you use the internet for?" include "looking for suppliers online", "e-commerce (selling products and services online)", "internet/online banking", and "recruitment". Reported answers to "Does the business use mobile money for…" include "paying suppliers", "receiving payments from customers", and "paying employees". Understanding customers is "agree" (as opposed to "not sure" or "disagree") response to the question "Regarding the internet/social media use, it helps to understand our customers better". The management-related questions are "Does your company use accounting software?" and "Does your company make use of inventory control/point of sale (POS) software?" (asked in the computer section of the questionnaire).

ACCESS TECHNOLOGIES EXTERNAL-TO-FIRM TRANSACTIONS INTERNAL-TO-FIRM Upstream and downstream transactions
Finance Management across all sectors is a self-employed household enterprise with no full-time employees. The median firms producing other services and manufactured goods are the most productive, and the median agricultural firm is the least productive (measured as monthly value added divided by the sum of full-time workers and the number of owners, in US dollars). These median productivity figures align as economic theory would predict with total sales, number of jobs, and entrepreneurial (owner) profits. Higher-productivity manufacturers and other service producers have higher median sales with a higher share of firms exporting, generate more jobs on average (a larger number of firms with more jobs pulling up the mean), and have higher median profits per owner. Conversely, lowerproductivity agriculture firms have the lowest median sales, generate the lowest number of jobs on average (together with trading firms), and have by far the lowest median profits per owner.

Table 4. Performance outcomes
Note: All values are based on unweighted data. Labor productivity is measured as value added (total sales minus raw materials and intermediate inputs plus water and electricity used in production) divided by the sum of full-time workers and the number of owners. Exporting reflects shares of firms active in each sectoral area that report having international customers, in response to the yes/no question "Does the business have customers located in other countries (selling goods or services abroad)?" Profits is measured as value added minus salary & wages. Labor productivity, sales, profits, and wage values are monthly (based on a typical month for expenses or annual divided by 12 for sales and profits) and in US dollars. Table 5 details digital divides across gender and age for the use of selected DTs with and without access to broadband internet. Across almost all DTs, the digital divide between men and women is larger than the divide between younger and older owners, where young owners are those aged 30 years and younger. Across all DTs, women systematically are less likely to use them than men. The largest divides between men and women are in the use of computers (with men 3.3 times more likely to use them, 7.9 percent of men vs 2.4 percent of women), using internet to find suppliers (2.4 times more likely), and using accounting software, inventory control/POS software, and the general use of smartphones (each 2.3 times more likely). On the other hand, while older owners are less likely to use most DTs than younger ones, older owners are more likely to use 2/2.5G phones and computers than younger owners.   While 13 percent of younger men-owned firms use a smartphone, only 3 percent of older womenowned firms use one. The digital divide is even larger with computer use, with less than 2 percent of young women-owned firms using one and more than 4 times more younger men-owned firms using one (8 percent).

Performance outcomes
This section explores whether the use of DTs matter in terms of their association with business performance outcomes, namely labor productivity, total sales and exporting, and jobs, both more jobs (full-time workers beyond the owner-manager) and better jobs (profits per owner and wages per worker). A first sub-section explores unconditional associations for different types of DTs for users versus non-users. A second sub-section presents regression results where other relevant enterprise-level variables and country effects are controlled for.

Unconditional associations
Tables 6a and b report means and medians of labor productivity, total sales, and exporting status (share of firms with some customers abroad), and of total employment (owners plus full-time workers), profits per owner, and wages per worker, respectively, for users versus non-users of DTs. The patterns for medians and means are relatively similar. First, users of DTs have higher mean and median productivity, sales, exporting status, total jobs, and profits per owner outcomes than nonusers. This holds for all DTs when comparing the median user versus non-user firm. 12 This relationship holds less consistently for worker earnings (wages per worker), as the share of earnings distributed across workers likely depends on many other factors in addition to total earnings and DT uses. Second, the differentials in performance outcomes between users and non-users of DTs with internet-enabled 3G-4G phones or computers are significantly larger than for basic DTs with noninternet-enabled (2G-2.5G) phones. While the productivity differential for users of 2G-2.5G phones relative to non-users is only 10 percent higher for the mean (an average of 888 vs 795) and 70 percent higher for the median user (116 vs 67), the differential for users of smartphones relative to non-users is larger by almost three times for the mean (2172 vs 771) and almost four times for the median user (364 vs 97).
The largest differentials in performance outcomes are associated with the use of internet-enabled DTs to facilitate specific GBFs. The largest productivity differentials between users and non-users are associated with management-related business functions, namely accounting and inventory control/POS software (the median user of accounting software having more than 5 times higher productivity, 773 vs 145), and with using internet to find suppliers (the average productivity of users being almost 3.5 times higher, 2728 vs 808). For total sales, the largest differential is again associated with using accounting software for both the median user and the average across users (almost 13 times higher for the median user, US$4,038 in sales per month vs $315, and almost 14 times higher for the average, $23,654 vs $1,724). Interestingly, using internet to recruit workers has a particularly large differential between users and non-users for firms that have some sales abroad, perhaps because internet enables lower cost hiring of some workers from countries where foreign customers reside, such as for hotel services (almost eight times higher, 39% of users exporting vs 5% of non-users). For jobs, using internet to recruit workers also has the largest differential between users and non-users (five times higher, 5 vs 1 job for the median firm and 10 vs 2 jobs for the average). For earnings, the largest profit per owner differential between users and non-users is again associated with using accounting software (more than nine times higher for both the median user, $1,398 per month vs $155, and for the average profit across users, $12,507 vs $1,368). For wages per worker, the largest differential between users and non-users is associated with using internet to find suppliers (roughly three times higher for both the median user, $150 vs $46 per month, and for the average wage across users, $270 vs $93). Figure 2 summarizes productivity, sales, and jobs average outcomes for users versus non-users of selected DTs. The figure visually highlights that firms using DTs with internet-enabled computers or smartphones have higher productivity, sales, and jobs outcomes than non-users, and higher outcomes than those enterprises using only 2G phones. Firms using smartphones have 2.8 times higher productivity, 6.0 times higher sales, and 1.9 times the number of jobs than non-users.
Computer users have 2.4 times higher productivity, 7.5 times higher sales, and 2.5 times higher jobs. Across all DTs, the highest sales premia from DTs are associated with the use of accounting software (13.7 times higher than non-users) and the use of internet for online banking. The use of accounting software and inventory control/point of sales (POS) software are associated with some of the highest jobs premia (3.8 and 3.5 times higher than non-users, respectively), after the use of internet to recruit workers (not illustrated).

Conditional associations
Firms that use a wide range of DTs with internet-enabled computers or smartphones are associated with higher conditional jobs outcomes than non-users, and higher conditional outcomes than those enterprises using only 2G phones. Annex Tables 1a-f and Figure 3 report findings of OLS (ordinary least squares) regressions between DT uses and performance outcomes (namely productivity, sales, export status, and employment, profits per owner, and wages per worker) that explicitly control for whether the enterprise has ever had a loan, has access to electricity, is run by transformational A range of DT uses are positively and statistically significantly (at least the 5 percent level) conditionally associated with productivity, total sales, jobs, and entrepreneur earnings (profits per owner). Regarding the use of DTs with 2G-2.5G phones, they include using a mobile phone to communicate with customers and using SMS to advertise with customers. Using a mobile phone to communicate with suppliers is significant for productivity, sales, entrepreneur profits and wages per worker but not for jobs. Using mobile money to pay suppliers and to receive payments from customers are positively and significantly associated with productivity, sales, and entrepreneur profits but not jobs. Using mobile money to receive payments from customers is also positively and significantly associated with wages per worker. Regarding the use of DTs with computers or with internet-enabled 3G-4G phones, they include the use of computers as an access technology and using internet to better understand customers, and for sales and jobs, they also include using accounting software and using internet for online banking. Using a smartphone as an access technology is significantly associated with productivity, sales, profit per owner and wages per worker but not jobs. Using internet to understand customers is significantly associated with productivity, sales, and entrepreneur profits, but not jobs. The largest positive significant correlates of higher levels of jobs are using internet for recruitment, using accounting software, and using inventory control/POS software. Using internet to find suppliers is significantly associated with wages per worker. Positive and statistically significant correlates of exporting status include using SMS to advertise with customers and using internet to pay employees.
Additional variables that are generally (across most DT uses) positively and statistically significantly correlated with productivity, total sales, jobs, and entrepreneur earnings include the gender and age of the manager (being male and younger), the manager having had vocational training, the owner being a transformational entrepreneur, the enterprise having more sophisticated upstream and/or downstream linkages with suppliers and customers, and regarding enterprise characteristics, older and formal firms, as well as whether the firm has ever had a loan and has electricity. The level of schooling of the manager is a significant positive conditional correlate of productivity, sales, and wages per worker but not jobs. And whether the manager had vocational training is a significant positive conditional correlate of productivity, sales, profits per owner, and jobs.

Figure 3. Firms using DTs with computers or smartphones are associated with more jobs
Note: Conditional correlates of productivity, sales and jobs that are significant at least at the 5% level based on OLS with robust standard errors using un-weighted data. Controls include manager age and gender of owner(s), schooling and vocational training, firm age and age squared, having electricity, having had a loan and having a line of credit/credit facility with suppliers, whether the owner is transformational, linkages with more sophisticated upstream suppliers or downstream customers, informal and urban/rural status, and sector, plus country fixed effects. Green bars represent those DTs with computers or smartphones; blue bars represent those with non-internet-enabled DTs. Data cover 3,325 firms across Ghana, Kenya, Mozambique, Nigeria, Senegal, South Africa, and Tanzania. Productivity is measured as value added (total sales minus raw materials and intermediate inputs plus water and electricity used in production) divided by the sum of full-time workers and the number of owners. Employment is the number of full-time employees plus owners. ** denotes variables that are statistically significant across all three performance outcomes; * denotes statistically significant variables across only sales and jobs. Figure 3 summarizes the main findings for productivity, total sales, and jobs. A greater range of more sophisticated DT uses by microenterprises based on internet-enabled computers or smartphones (green bars) relative to DT uses based on only 2G phones (blue bars) are conditionally associated with higher job levels. Importantly, there is a positive progression in the number of more sophisticated DTs (green bars) associated in turn with higher productivity, sales, and job levels. Six internet-enabled and three non-internet-enabled DT uses are the only significant conditional correlates of higher job levels. Using a smartphone and using a computer are statistically significantly conditionally associated with some performance outcomes, as well as using internet to find suppliers, to understand customers, and for online banking, as well as using accounting and inventory control/POS software. Importantly, use of DTs for these simple management functions is sales ln productivity strongly associated with jobs: firms using accounting and inventory control/POS software are associated with a roughly 1.6-person larger firm size (0.47 and 0.44 log points, respectively) than those not using them.

Informality
This section explores differences and similarities in performance outcomes across formal and informal enterprises to better understand the heterogeneity across microenterprises in SSA. It seeks to understand the extent to which some informal microenterprises are similar to "good" formal firms, especially with regards to the use of DTs, where "good" firms are defined as those that have higher productivity and sales levels including sales abroad and generate more full-time jobs as well as higher earnings. It explores whether there are observable characteristics associated with these desirable performance outcomes, and whether correlates of good performance apply equally to formal and informal firms. The first sub-section characterizes the performance gaps between the average formal and informal enterprise, as well as gaps in terms of assets available to them and use of DTs. The second sub-section explores the use of DTs by top-performing formal and informal firms. It begins by exploring the differential use by formal and informal firms in the top productivity and sales deciles. The exploration is broadened by creating four clusters of firms, namely highproductivity (high productivity and sales), scale (average productivity with high sales and jobs), under-performing (average productivity and sales, low jobs), and laggard (low productivity, sales, and jobs) firms, based on a data-driven mix of four performance variables (productivity, total sales, exporting status, and jobs). It examines the use of DTs among informal firms relative to formal firms across these clusters.

Performance gaps
There are significant differences between the performance indicators of formal and informal firms. This can be seen in Figure 4 that presents Kernel densities of performance indicators of formal and informal firms in the data set. The indicators used are log of labor productivity in USD, log of sales and log of total people employed in the enterprise including the owners. As there are large differences in the means of the variables across countries, the variables are standardized at the level of countries. The figures show that for all the performance variables, the distribution of formal firms is the right of that for informal firms, reflecting overall higher performance. The figures also show that there are large areas of overlap, reflecting the fact that both sets of firms are quite heterogeneous and there are informal firms that "look like" formal firms in terms of performance. 13

Figure 4. Kernel densities of productivity, sales and employment
Note: kernel densities of log labor productivity, log sales and log number of employees (including the owners). All variables have been standardized at the level of countries.
There are large gaps in the overall performance of formal versus informal firms. Formal firms, constituting 27 percent of all firms in the data set, on average have higher productivity, higher sales, pay more to their workers, and have higher profits per owner (Table 7). Also, while about 9 percent of formal firms sell to some customers abroad, only 3.4 percent of informal firms do so. Interestingly, the difference between the median value of wages per worker for informal versus formal firms is smaller. The difference in means for wages reflects the skewed nature of wages per worker among formal firms, as there are a number of formal firms that pay much higher wages relative to the rest of formal firms.

Table 7. Performance gaps between formal and informal firms
Note: The overall total sub-sample is smaller as there are x firms with missing values of labor productivity.
The gap in the performance indicators of formal versus informal firms is smaller among the high performing firms in terms of productivity and sales. This indicates that the top performing informal firms are on average relatively more like the top performing formal firms and reflects the large tail of low performing informal firms. Furthermore, the share of informal firms among the top decile firms is relatively high (about 50 percent among the top decile by productivity and 52 percent among top decile by sales). While for the overall sample the labor productivity of formal firms is about twice as large as that of informal firms, for firms in the top decile this ratio is only 1.3 (1.2 for firms in the top decile by volume of sales). There are similar reductions in the gaps for sales and profit per owner when moving from the overall sample to the top decile firms, but not for mean wages per workers or export orientation.
Formal firms generally benefit from higher levels of various types of assets relative to informal firms (Table 8). Formal firms are more likely to be owned by transformational owners, managed by more educated people, are more likely to have electricity, and better access to external finance. Formal firms also are more likely to have suppliers that are formal or are situated abroad, and to have large enterprises and non-local buyers as customers.

Table 8. Assets of formal vs informal firms
The gap in the assets available to formal and informal firms is lower for firms in the top decile with respect to productivity than for the overall sample. For example, while the likelihood ratio of having transformational owners (that is, the ratio of the share of formal firms with transformational owners to that of informal firms with transformational owners) is 1.74 (=0.41/0.24) for the overall sample, this ratio is only 1.24 for firms in the top decile of productivity. Similar reductions in the likelihood ratios occur in all assets (except for median schooling of manager, which remains the same) Interestingly, a different pattern emerges when one compares the gaps between formal and informal firms in the overall sample with those in the top decile by sales. For many of the assets, the gap increases. For example, the likelihood ratio of having transformational owners increases from 1.7 in the overall sample to 2.1 in the top decile by sales. The ratio increases from 2.3 to 4.2 for the share of firms whose managers have had vocational training, from 3.4 to 6.2 for the share of firms with loans, and from 3.5 to 6.4 for the share of firms with suppliers abroad.

Use of DTs by the average formal firm and performance outcomes
Annex Tables 2a-f present performance regressions with formality status interacted with use of DTs for productivity, sales, export status, and employment, profits per owner, and wages per worker, controlling for other enterprise characteristics. This analysis builds on the analysis presented in section 4.2.
A first finding is that the average formal firm-relative to the average informal firm--is statistically significantly and positively associated with higher productivity, sales, and jobs outcomes, as well as with higher earnings, both profits per owner and wages per worker. This conditional finding is aligned with the unconditional findings reported in Table 7.
A second finding is that, even when including an interactive formality term, the average firm using key DTs-relative to the average non-user firm-is statistically significantly and positively associated with higher productivity, sales, and profits per owner. Key DT uses with these positive associations include non-internet-enabled 2G/2.5G phones, mobile phones to communicate with suppliers and customers, SMS to advertise with customers, and mobile money to pay suppliers and receive payments from customers, as well as internet-enabled smartphones and computers, and internet to understand customers. For the employment outcome variable, the association with uses of DTs is significantly positive for using mobile phones to communicate with suppliers and customers, using SMS to advertise with customers, using mobile money to pay workers, and using internet to recruit workers.
A third finding linked to the interactive formality term is a significant difference between formal and informal firms in the association between their use of either internet-enabled or non-internetenabled DTs and key performance outcomes. Regarding the use of mobile 2G/2.5G phones as an access technology, mobile phones to communicate with suppliers and customers, and mobile money (to pay suppliers and receive payment from customers), the association between these noninternet-enabled DTs and productivity and sales outcomes is significantly weaker for formal firms, relative to informal firms (the sign of the interactive term coefficients is significantly negative). This difference between formal and informal firms also holds for the use of mobile 2G/2.5G phones for jobs and profits per owner outcomes. While the association between these non-internet-enabled DTs and these performance variables is weaker for formal firms, the association between key internet-enabled DT uses and jobs is stronger for formal firms (the sign of the interactive term coefficients is significantly positive). These DT uses include using a smartphone and computer, email and VOIP to communicate, internet to understand customers, ecommerce, inventory control software, and internet to recruit workers and for online banking. While this finding does not imply causality, it could suggest that the job benefits from the use of specific internet-enabled DTs are more readily appropriable by the average formal firm relative to the average informal firm. Conversely, it could suggest that the productivity, sales, and jobs benefits from the use of specific non-internet-enabled DTs are more readily appropriable by the average informal firm relative to the average formal firm, or that they are used in a more extensive manner by the average informal firm and thereby yielding a stronger positive association with outcome variables.

Use of DTs by top-performing firms
Relative to informal firms, formal firms are more likely to adopt various types of DTs (Table 9). Regarding access technologies, the difference in adoption between formal and informal firms is very small for 2G phones but sizable for the use of smart phones, computers and websites. For most DTs presented in Table 3, the gap in use is smaller for firms both in the top decile with respect to productivity and sales than for the overall sample. In the case of having computers, for example, while in the overall sample the likelihood of having a computer for formal firms is almost 7 times that of informal firms, this ratio is reduced to a factor of 2(3) among firms in the top decile of productivity (sales). Similar reductions in use gaps are evident for DTs that are employed for GBFs external to the firm. For the overall sample, the gap is especially large for using internet for online banking (with formal firms 11 times more likely to use internet for online banking than informal firms), to look for suppliers (ratio of 5.5), for e-commerce (ratio of 4.7), and to understand customers better (ratio of 4.3). For all the GBFs listed in the table except for two, the likelihood ratio of formal to informal firms using that DT is larger for the overall sample relative to the subsample of firms in the top deciles of productivity and sales. The only exceptions are e-commerce and using internet to recruit workers. There are no informal firms in the top decile of sales using e-commerce. Interestingly the gap in the use of e-commerce almost disappears for firms in the top decile of productivity. None of the informal firms in the top deciles use internet to recruit workers. Finally with respect to DTs used for GBFs internal to the firm, the gap in the overall sample is very high for use of accounting and inventory control or point of sales software (ratios of 13 and 22, respectively). The gaps are significantly reduced in the case of firms in the top deciles (to 4.1 and 5.0, respectively, in the case of using accounting software and to 5.9 and 9.0 in the case of inventory control/POS software).
The discussion so far reveals an interesting comparison between the gaps between formal and informal firms on average on the one hand and being "good" (in the top decile) in terms of productivity and sales. The data reveal that in terms of using various types of DTs relative to the overall sample, formal and informal firms become more "similar" once they are top performers, irrespective of whether top performance is defined in terms of productivity or sales. However, this is not true for access to useful assets. For many assets, the gap between formal and informal firms increases for the group of top sales performers (whereas they mostly decline among top productivity performers). Gaps increase with scale and decrease with productivity. Note: All responses are shares (%) of firms based on weighted data. Use 2G/2.5G mobile phone is based on responses to "Does the business manager have a mobile phone?" and subtracting those reporting using a 3G/4G phone. Smartphone users answered "yes" to "How does the business access the internet: Mobile broadband (3G/4G, wireless)." Use computer is a non-zero response to "How many computers does your business have?" Have website have answered "Yes" to the question "Does your business have a website?". Use mobile to communicate with suppliers and customers are "mobile phone" answers (so could be using 2G, 2.5G, 3G or 4G phones) to the question "How does the business usually communicate with its suppliers/customers?" Advertise SMS are "mobile/SMS" answers to the question "How does the business advertise?" Mobile banking is in response to "Have you used mobile phone banking for business?" Reported answers to "What do you use the internet for?" include "looking for suppliers online" (here "find supplier"), and "e-commerce (selling products and services online)". Reported answers to "Does the business use mobile money for…" include "paying suppliers", "receiving payments from customers", and "paying employees". Use internet to understand customers is an "agree" (as opposed to "not sure" or "disagree") response to the question "Regarding the internet/social media use, it helps to understand our customers better". The management-related questions are "Does your company use accounting software?" and "Does your company make use of inventory control/point of sale (POS) software?" (both asked in the computer section of the questionnaire).
These similarities and divergences among top-performing firms suggest a more detailed look at heterogeneity may be warranted. There is significant heterogeneity across formal and informal firms in terms of performance indicators, assets, and use of DTs. Furthermore, there are multiple indicators that can be used to evaluate performance. Some firms excel in productivity but generate few or no jobs, while others generate scale in terms of sales and/or jobs while still others may excel in productivity, sales, and jobs. A clustering approach is used to classify firms in terms of four indicators: productivity, total sales, export status, and jobs. As the inclusion of export status does not make much difference in the clusters created, it was left out in the reported cluster analyses. The procedure resulted in 4 clusters with the following characteristics (Table 10): 14 1) High-productivity firms: high productivity and sales, average employment 2) Scale firms: average productivity, high sales and employment 3) Under-performing firms: average productivity and sales, low employment 4) Laggards: low productivity, sales and employment. Table 10 also shows that both firms in the high-productivity and scale clusters are more export oriented than average.

Table 10. Performance characteristics of firm clusters
Note: log labor productivity, log sales and log employment are standardized (i.e., mean zero and standard deviation equal to 1). Export status is not standardized. Values within the ad-hoc -0.15 to +0.15 standard deviation range are considered close enough to 0 to be labelled "average", values above "higher than average", values below "lower than average". Table 11 provides information on DT use by formal and informal firms in the different clusters. Focusing first on basic access technologies (the panel on the left), frequency of DT use is higher among both high-productivity and scale firms relative to under-performing firms. Further, among high-productivity firms and scale firms, DT use is more widespread among formal relative to informal firms, except for 2-2.5G mobile phones. Third, formal firms in the scale cluster are more likely to use DTs than formal firms in the high-productivity cluster. Mainly as a result of this, the gap between the likelihood of using access technologies of formal and informal firms is larger among scale firms relative to high-productivity firms. These patterns are broadly (but not always) repeated for DTs that are used for business transactions that are external to the firms and those that are used for operations within the firm. For all types of DTs, high-productivity and scale firms are more frequent users than under-performers. Within both groups of GBFs, formal firms are more frequent users of DT relative to informal firms. For some DTs, the gaps between formal and informal firms are largest among scale firms. For example, within the scale cluster, the share of firms using online banking is almost 28 times larger than the share of informal firms using this technology. This ratio is about 14 for using internet for e-commerce, 7.8 for using internet to find suppliers, 8.3 for using accounting software and 5.6 for using internet to understand customers. These large gaps are generally associated with technologies that rely on internet access rather than simply 2/2.5G. It is noteworthy that the corresponding ratios of formal to informal firms within the high productivity cluster are smaller (5.0, 1.1, 3.0, 4.4 and 2.2, respectively). Hence regarding use of internet-reliant DTs, the gaps between formal and informal firms are larger among scale firms relative to high-productivity firms. This, in turn, reflects relatively high use among formal scale firms. Put differently, among formal firms, scale seems to be more closely associated with DT use than productivity. Finally, the gap between formal and informal firms in the under-performers cluster is also smaller than that in scale firms.
The clustering exercise reveals two interesting patterns. First, it points to a tension between higher productivity and increased employment. Firms in the cluster characterized by high productivity are not generators of high employment, and vice versa. While a tension between automation and jobs has long been emphasized in the literature (Acemoglu and Restrepo 2019) and a tension between productivity and employment has been observed among SMEs and large enterprises in Ethiopia and Tanzania (Diao et. al. 2021), in the present context the apparent tension is observed among micro enterprises. Better data is required to evaluate whether this tension observed in the data reflects a true tradeoff between employment and productivity: high labor productivity does not necessarily mean high total factor productivity (TFP), and the latter is a better indicator of productivity. It could be that high labor productivity is associated with high capital intensity but low TFP. Moreover, it could be that high productivity firms will employ more people in the future, a dynamic aspect that cannot be captured in a single cross section data set.
The second interesting pattern is that among formal firms, scale is more strongly associated with use of internet-reliant DTs. Why this may be the case is an important question for future research. A related question is whether alleviating constraints in greater use of DTs among high-productivity firms could enable them to expand their scale in line with their higher productivity.

Adoption: Why is productive use so low?
An important outstanding puzzle is why so few microenterprises, including high-productivity informal ones, use DTs for productive use or even have a smartphone or computer. Of total people with available internet service, SSA countries have the highest share of unconnected users in the world, with almost three-quarters (74%) of those living in areas with 3G-4G availability not connected and not using these services at all in 2021. This figure has virtually not changed since 2010. And while 3G-4G availability (coverage) has increased from an average of 25 percent of people in 2010 to 71 percent in 2018 and 84 percent by 2021, mobile internet adoption and use has only increased from 6 percent in 2010 to 18 percent in 2018 and less than 22 percent by 2021. This 18 percent use figure across SSA in 2018 includes multiple individuals in high and middle-income households for entertainment and other consumption uses, as well as multiple users within larger enterprises employing 10 or more workers. Against this figure, that only 6.5 percent of microenterprises use smartphones remains surprising. It is even more startling given that, as reported in sections 4 and 5, the use of DTs, especially DTs with computers and internet-enabled smartphones, is on average associated with higher productivity, higher sales, more jobs, and higher profits for owners. This section presents novel data to identify main correlates of enterprise use. It is important to understand what the main associates of enterprise adoption of DTs are, starting with access technologies such as smartphones and computers, to provide insights into the formulation of effective public policies to reduce the barriers and enhance the positive correlates of adoption.

Table 12. Self-reported internet access constraints
Note: Respondents that replied No to "Are you using internet/social media for business purposes" were asked "Why does your business not have internet access?" Multiple replies are allowed among the reported choices.
The main drivers of adoption are affordability, which includes ability to pay for DTs as well as complementary infrastructure, and willingness to use, which includes attractiveness and capabilities, as well as business environment and socioeconomic factors (section 2). A first set of data to explore main constraints to internet-enabled DT uses comes from directly asking microenterprises why they do not have internet access. Based on self-reported reasons for not using internet, lack of attractiveness is the most important constraint. As reported in Table 12, availability of internet service is not the main constraint, with less than 20 percent of respondents replying lack of availability as a constraint. This corresponds roughly with the average of 29 percent of people who had no 3G+ coverage in 2018 across SSA countries, with a lower figure expected here given that sampled Enumeration Areas for microenterprises have largely focused on residential areas where coverage is expected to be higher. Roughly one-third (34.8%) of respondents indicate that affordability is a constraint. This is likely linked to being able to purchase a smartphone or computer, pay for internet use, as well as having availability and being able to pay for electricity. However, affordability alone does not explain the majority of the more 30 than 93 percent of microenterprises that do not have a smartphone or a computer to access internet. By far the largest number of respondents, over seven out of ten non-users, indicate that lack of attractiveness is the main constraint ("no need"), presumably either because there are no apps available that are useful to them in their local language that meet their productive needs, because their general skill level does not enable them to understand how they could productively use available apps, or because the available quality of service is so poor (with no or limited download availability of useful information at times when it is needed) that it is not useful to them. Finally, another roughly one-third (33.7%) indicate that there is capability or skill gap, likely because the apps are not designed with the level of digital skills of the users in mind and are not sufficiently intuitive and easy to use. 15 Table 13 reports marginal effects based on multinomial logit/probit regressions on available enterprise characteristics controlling for country fixed effects. Smartphone adoption is modeled as a multinomial logit regression with enterprise choice of either smartphone, 2G or 2.5G phone (without full internet functionality) or no phone, while computer adoption is modeled as a simple yes/no choice. The reported findings are restricted to variables that have at least one statistically significant association (at least at the 5 percent level of significance). Although reported findings do not allow inferences to be made about causality, an assessment of statistically significant associations between DT uptake and firm characteristics provides valuable inputs to enrich the existing understanding of DT adoption across SSA countries.  complementary infrastructure, including access to cheaper and better transportation and logistics services). Having a loan is the largest correlate of both adopting a smartphone and a computer, in terms of being highly statistically significant (at the 1% level) and with the largest coefficient (relative to other coefficients that are all related to 0/1 independent variables, namely all covariates listed in the table except for schooling of manager and firm age). Firms that have a loan are 18 percentage points more likely to adopt a smartphone and 9 percentage points more likely to adopt a computer. Having a loan is also associated negatively with having a more basic 2G-2.5G phone: firms with a loan are almost 15 percentage points less likely to have a basic phone. Not having a credit line is significantly negatively associated with not having a mobile, hence with having at least a 2G phone or a smartphone or computer. Having electricity is also a highly statistically significant correlate of having a smartphone or computer, though it is also significantly associated with having a more basic 2G-2.5G phone. As highlighted by the large, significant negative coefficient for no mobile, not having electricity is the largest correlate of not having a mobile: firms that have electricity are 13 percentage points less likely to have no mobile. Finally, being in an urban location is a strong positive correlate of having a smartphone: urban firms are 4 percentage points more likely to have a smartphone.
Attractiveness of DTs as a determinant of adoption is likely at least in part driven by requirements to use specific DTs when microenterprises have important business linkages with larger or non-local firms either as upstream suppliers or as downstream customers. Firms that have large formal businesses as main suppliers are almost 7 percentage points less likely to have no mobile. Conversely, these firms are 7 percentage points more likely to have a basic mobile phone. Similarly, firms that have non-local customers as their most important customers are over 10 percentage points less likely to have no mobile. Importantly, it is having big enterprises as main customers that enhances the likelihood of adopting a smartphone or computer. Firms with big firms as customers are 4 percentage points more likely to have a smartphone and 5 percentage points more likely to have a computer. This finding supports the view that microenterprises can gain significantly from adopting better technologies and the associated learning from participation in value chains with larger, more sophisticated downstream buyers.
People's capabilities, especially vocational training, are also strongly associated with both smartphone and computer use. Capabilities affect willingness to use largely by enabling understanding of how to use and extract productive value from DTs. The skills dimension of capabilities is proxied by the highest year of schooling attained and by vocational training certificates held by the manager. The schooling of the manager is statistically significantly associated with both smartphone and computer uptake, as well as the use of basic mobiles: one more year of a firm manager's schooling is associated with a 0.2 percent higher likelihood of adopting a smartphone, a 0.5 percent higher likelihood of adopting a computer, and an almost 0.8 percentage point higher likelihood of adopting a basic mobile. It is also associated with a 1 percent lower likelihood of not adopting any mobile. These effects should be scaled up in magnitude to the extent that individuals typically have several additional years of schooling if they decide to continue beyond primary to finish secondary schooling. Vocational training is associated with a 4 and 3 percentage points higher likelihood to adopt a smartphone and computer, respectively. Importantly, vocational training is not significantly associated with basic phone adoption, while firms with vocational training are 7 percent less likely to not adopt any mobile phone.
Enterprise technological capabilities, namely the need and ability of firms to use more sophisticated DTs, are linked to the complexity of inputs, production, and quality of outputs required by markets. They are proxied by firm size and age, formality status, and sector of industrial activity, with the presumption that larger, younger, and formal firms, as well as firms in manufacturing and services are likely to have accumulated more complementary know-how assets and therefore have a greater need to use smartphones and computers. 16 Firms that are larger are more likely to use computers: firms that have 3 or more employees are 3.5 percentage points more likely to us a computer. Younger firms are more likely to use smartphones, while older firms are more likely to use basic phones. 17 Formal firms are 2 percentage points more likely to use smartphones and over 3 percentage points more likely to use computers. Regarding sectors of enterprise activity, it is firms in other services activities that are 4 and 3 percentage points more likely to use a smartphone and a computer, respectively, relative to agriculture. Trading firms, on the other hand, are more likely to still use basic mobile phones.
Finally, the only available socioeconomic variable is whether the enterprise is owned by women. Women-owned firms are roughly 2 and 3 percentage points less likely to use a smartphone and computer, respectively, than male-owned firms. 18 Since a majority of these firms are self-employed enterprises with no full-time paid employees where the owner is also the manager, this digital divide may reflect prevailing social norms and rules that make ownership of these access devices relatively more difficult for women.

Conclusions and further work
The main findings of this paper can be summarized as follows. First, a surprisingly small number of microenterprises across SSA countries use internet-enabled access technologies. Less than 7 percent use a smartphone, less than 6 percent a computer, and roughly 20 percent still do not use any mobile phone. This suggests that more policy research work is needed to understand why productive use of DTs is so low across SSA countries. Second, even fewer firms use digital tools enabled by these access technologies. Only two in five firms use a mobile phone to communicate with suppliers or customers, and less than one in five use them to pay suppliers or receive payments with mobile money. And among those with smartphones, less than half use internet to find lower-cost or more favorable suppliers, and less than a third use internet for ecommerce. Only half of firms with a computer use accounting software or inventory control/POS software. Third, across all DTs, women are less likely to use them than men. The largest divides between men and women are in the use of computers (with men 3.3 times more likely to use them), using internet to find suppliers (2.4 times more likely), and using accounting software, inventory control/POS software, and the general use of smartphones (each 2.3 16 There is nothing inherent about enterprises in the agriculture sector that necessarily predispose them to be using less sophisticated technologies, but they may not have as many formalized systems in place to build capabilities and accumulate know-how assets in lower-income countries. Based on the larger-firm FAT findings, technological sophistication is higher for higher-income countries (the Republic of Korea, Poland and Brazil) for agriculture relative to manufacturing and services, as summarized by the average business sophistication index of the firm across all general business functions and the sector-specific business functions of a particular sector. See Figure 2.2, Cirera, Comin and Cruz (2022). 17 This conditional finding is aligned with a similar unconditional finding reported in Table 5. 18 This conditional finding is also aligned with a similar unconditional finding reported in Table 5. 33 times more likely). These stark differences suggest that more policy research work is needed to understand the constraints facing women in their productive use of DTs, and what can be done to overcome these constraints.
Regarding corelates of performance, a greater range of more sophisticated DT uses based on internetenabled computers or smartphones relative to DT uses based on only 2G phones are conditionally associated with higher job levels. There is a positive progression in the number of more internetenabled DTs associated in turn with higher productivity, sales, and job levels. Six internet-enabled and three non-internet-enabled DT uses are the only significant conditional correlates of higher job levels. Firms using accounting and inventory control/POS software are associated with a roughly 1.6-person larger firm size (0.47 and 0.44 log points, respectively) than those not using them. There is a tension between higher productivity and increased employment. Firms characterized by the highest productivity are not generators of high employment, and vice versa. While this tension between high-productivity, skill-intensive technologies (automation) and jobs has been emphasized among SMEs and large enterprises, this is the first evidence of such tension among microenterprises. Formal "scale" (high sales and jobs) firms are more strongly associated with the use of internet-enabled DTs than formal and informal high-productivity firms. A relevant policy research question is whether alleviating constraints in greater use of DTs among high-productivity firms could enable them to expand their scale in line with their higher productivity.
Regarding determinants of adoption, over seven of ten non-users indicate that lack of attractiveness ("no need") is the main constraint to productive use of DTs. A relevant policy question is whether this is mainly because there are no apps available that are useful to them in their local language that meet their productive needs, because their general skill level does not enable them to understand how they could productively use available apps, because the available quality of service is so poor that access to internet is not useful to them, or due to some other reason. The most important conditional correlates of smartphone and computer adoption among available variables are related to having a loan, having electricity, having business linkages with big firms as customers (being embedded in value chains that require upgrading), and relevant skills (managers having vocational training). More policy research work is needed to better understand what policies may be effective and appropriate to increase adoption and more intensive productive use of DTs across SSA countries.
As was emphasized in the introduction, while the data set used in this paper provides a wealth of information difficult to find elsewhere on microenterprises, the fact that it consists of a single cross section prevents causality from being inferred from the presented results. Importantly, it precludes any analysis of firm dynamics. Repeated implementation of the same survey, by following the same enterprises over time, would increase the amount of information for causal inference of the adoption decision, the impact of DTs on performance and firm growth, as well as on determinants of firm exit, and the differential impacts for formal versus informal enterprises. Given that one of the major problems faced by microenterprises in SSA is that few of them grow into competitive mid-sized or large companies, such a panel data set would provide valuable information to understand why this is so and whether the more widespread use of DTs for productive purposes can be effectively used to circumvent the associated barriers. Providing support for such future additional surveys would constitute an invaluable investment to increase the World Bank's knowledge capital, and most importantly the knowledge capital available to SSA countries and their policy makers.