Data on the technology frontier at the industry level
There is a commonly held view that construction is a digital laggard. A widely cited McKinsey report in 2016 argued: “the construction sector has been slow to adopt process and technology innovations …. The industry has not yet embraced new digital technologies that need up-front investment, even if the long-term benefits are significant. R&D spending in construction runs well behind that of other industries: less than 1 percent of revenues, versus 3.5 to 4.5 percent for the auto and aerospace sectors. This is also true for spending on information technology, which accounts for less than 1 percent of revenues for construction, even though a number of new software solutions have been developed for the industry.”
Technology use and diffusion is an important dynamic in industry development, but good data is rare. Most surveys are of specific industries or selected firms (typically large ones), and surveys with large samples that would be more representative of firms across the economy by including small and medium size firms are scarce. This particularly affects built environment industries like construction and professional services because of the large number of small firms in those industries.
The United States Census Bureau conducts an Annual Business Survey (ABS), and in 2018 the ABS included a technology module with three questions about the extent of technology use between 2015 and 2017: the availability of information in digital format (digitization), expenditure on cloud computing services, and use of a range of advanced business technologies. The survey was a partnership between the Census Bureau and the National Center for Science and Engineering Statistics, and the first results have been released in a working paper from the National Bureau of Economic Research. The results are summarized below.
The survey data is at a high level of generality, to make the questions relevant across the variety of firms and industries included. There are also issues around how the data has been modelled and analysed, and the adjustments for high counts for firms that minimally use or are not using advanced business technologies at all. Importantly, the survey shows construction is not significantly lagging other industries in the US in digitization and use of cloud services, however it is doing less testing and development of advanced business technologies.
The main finding of the survey was “Despite increasingly widespread discussion in the press of machine learning, robotics, automated vehicles, natural language processing, machine vision, voice recognition and other advanced technologies, we find that their adoption rates are relatively low. Furthermore, adoption is quite skewed, with heaviest concentration among a small subset of older and larger firms. We also find that technology adoption exhibits a hierarchical pattern, with the most sophisticated technologies being present most often only when more-basic applications are as well.”
There were 583.000 responses to the survey. Two thirds of the firms employed under 10 people and were less than 20 years old, making this the most comprehensive survey of diffusion of advanced business technologies done so far. The industry breakdown of firms is in Table 1, with the built environment industries of construction, real estate and professional services well represented. This makes the data particularly interesting.
Table 1. Industry breakdown of firms
Sector | Distribution % |
Agriculture, Mining, Utilities | 2 |
Construction | 10 |
Education | 1 |
Finance, Insurance, Real Estate | 10 |
Health Care | 9 |
Information | 2 |
Management & Administrative | 5 |
Manufacturing | 8 |
Other (Arts, Food, Other Services) | 14 |
Professional Services | 17 |
Retail Trade | 13 |
Transportation & Warehousing | 4 |
Wholesale Trade | 5 |
Following are extracts from the NBER paper on the results of the survey on the three technology questions: the availability of information in digital format, expenditure on cloud computing services, and use of advanced business technologies with a focus on the use of AI. The next post will discuss the survey and its implications for construction.
Digital Share of Information by Business Activity
The first question in the 2018 ABS technology module queried firms on the type of information stored digitally. In all sectors, financial and personnel information are the most likely to be digitized, followed by customer feedback and marketing. This is the case for Construction, with financial, personnel and marketing digitized. The lowest rates of adoption are in production and supply chain activities.
Figure 2 is a butterfly chart of adoption and use rates for digital information by sector, where the ranking of sectors by adoption and intensity of use rates parallel each other. The right panel of the chart represents, by sector, the adoption rates of digital information across all surveyed information types. The segments within each bar in the chart capture adoption rates by the number of information types in digital format. In all sectors, a large share of adopters report having three or more types of information digitized.
The left panel of Figure 2 represents intense use of digitization. Most firms report digitizing at least two types of information, regardless of sector, the fraction of firms digitizing only one type of information intensively is relatively small in each sector. Overall, digitization appears to be highly prevalent across sectors. Manufacturing, Information and Professional Services are among the highest adopters of digitization, with size being a primary correlate of adoption.
Figure 2: Extensive and Intensive Margin Measures of Digitized Information by Sector
Cloud Service Purchases by IT Function
This section describes the adoption patterns for cloud service purchases across size, age and sector. Like digitization, the highest adoption and intensive-use rates are in Information, followed closely by Professional Services and Education. The lowest rates are in Agriculture, Mining, Utilities, Retail Trade, and Transportation and Warehousing, and the Other category. Figure 4 reveals that cloud services purchases have much lower diffusion rates compared to those for digital information in any given sector.
Billing and Security are the most common IT functions for most sectors, with certain sectors, including Construction, predominantly relying on the cloud to perform collaborative or synchronized tasks. The Data Analysis function has the lowest number of firms reporting some cloud purchase, Billing and Account Management has the highest number of firms, closely followed by Security or Firewall and Collaboration and Synchronization functions.
Although the adoption rates for business IT functions in the cloud is significantly lower than the adoption rates of storing information digitally, this technology is widespread across various applications, with nearly a third of each different type of IT function being performed in the cloud and being used intensively.
Figure 4: Extensive and Intensive Margin Measures of Use Rates for Cloud Service Purchases by Sector - Conditional
Advanced Business Technologies
In this section we analyze firm responses to the business technologies question. Due to their wide technological scope, we link the responses here with the previous technology adoption questions and perform a deeper set of analyses assessing the range of response categories. Very few firms use the business technologies included in the module, and many answered, “Don’t know”. Based on our tabulation weights, only 10.3% (8.5% non-imputed) of firms adopt at least one of the listed advanced business technologies.
The highest use frequencies are in touchscreens and machine learning. For touchscreens the adoption rate is 6.1% of firms. Machine learning comes second but the rate is low at 2.9%. Voice Recognition and Machine Vision, which are can be considered examples of Machine Learning applications, have the next two highest use rates.
The overall diffusion of robotics is very low across firms in the U.S. The use rate is only 1.3%, concentrated in large, manufacturing firms. The distribution of robots among firms is highly skewed toward larger firms. The least-used technologies are RFID (1.1%), Augmented Reality (0.8%), and Automated Vehicles (0.8%).
Looking at the most common types of business technologies adopted by sector in Table 2 there is substantial variation. All sectors (except Manufacturing adopt Touchscreens followed by Machine Learning or Voice Recognition. Manufacturing is most likely to adopt Machine Learning followed by Touchscreens and Robotics. RFID technology is most commonly used in the Retail, Wholesale, and Transportation and Warehousing sectors, consistent with these industries tracking physical goods through supply chains.
Table 2. Top Use Sub-Categories for Business Technologies by Sector (p. 60).
Sector | 1st | 2nd | 3rd |
Agriculture, Mining, Utilities | Touchscreens | Machine Learning | Automated vehicles |
Construction | Touchscreens | Machine Learning | Voice Recognition |
Education | Touchscreens | Machine Learning | Voice Recognition |
Finance, Insurance, Real Estate | Touchscreens | Voice Recognition | Machine Learning |
Health Care | Touchscreens | Voice Recognition | Machine Learning |
Information | Touchscreens | Machine Learning | Voice Recognition |
Management & Administrative | Touchscreens | Machine Learning | Voice Recognition |
Manufacturing | Machine Learning | Robotics | Touchscreens |
Other Arts, Food, Other Services | Touchscreens | Machine Learning | Machine Vision |
Professional Services | Touchscreens | Voice Recognition | Machine Learning |
Retail Trade | Touchscreens | Machine Learning | RFID |
Transportation & Warehousing | Touchscreens | Machine Learning | RFID |
Wholesale Trade | Touchscreens | Machine Learning | RFID |
The testing-versus-use rates across different technologies are used to assess which technologies are in earlier phase of diffusion, that is, where testing is high relative to use. In Figure 6, the vertical axis represents the ratio of the fraction of firms testing to the fraction of firms using. The technologies are represented by the circles. The size of each circle corresponds to the use rate for that technology with larger circles representing higher rates of use. Technologies are ordered in the figure by usage rate, low to high.
As shown in panel a, the technology with the highest testing-to-use ratio is Augmented Reality, where nearly half as many firms as those using the technology report testing it. The next highest ratios are observed in RFID and Natural Language Processing and the lowest ratios are in technologies that are relatively more diffused (and hence, used), such as Touchscreens, Machine Learning and Machine Vision. For Touchscreens, for instance, only about 15 firms report testing the technology for every 100 that use it. It is notable that most testing-to-use ratios are below 0.3, indicating that there are fewer than 30 firms testing the technology for every 100 using it.
The remaining panels of Figure 6 plot the testing-to-use ratio for technologies by firm size, age, and manufacturing status. Panel b displays ratios by firm size, where small firms are defined as those with 1-9 employees and large firms are those with at least 250 employees. The blue circles capture usage among large firms and the orange circles represent usage among small firms. The sizes of the circles are smaller for small firms for each technology, consistent with the earlier finding that larger firms tend to use the business technologies at a higher rate, in general.
Figure 6: Testing-to-Use Ratios
a. Testing-to-Use Ratios for all Business Technologies (All Firms)
b. Testing-to-Use Ratios for all Business Technologies (By Size)
c. Testing-to-Use Ratios for all Business Technologies (By Age)
d. Testing-to-Use Ratios for all Business Technologies (By Manufacturing Status)
The butterfly chart in Figure 7 provides sectoral diffusion rates for all business technologies considered together. Manufacturing leads with about 15% of firms indicating use of at least one business technology, followed by Health Care (14%), Information (12%), Education (11%) and Professional Services (10%). The lowest diffusion rates for the technologies are in Construction, Agriculture, Mining and Utilities, Management and Administrative, and Finance, Insurance and Real Estate sectors.
Figure 7: Extensive and Intensive Margin Measures of Use and Testing Rates for Business Technologies by Sector
The third question asked directly about the use of advanced “business technologies,” including those typically categorized as “AI.” These technologies include automated guided vehicles, machine learning, machine vision, natural language processing, and voice recognition software. Respondents are presented with a list that covers robotics (i.e., “automatically controlled, reprogrammable, and multipurpose machines”), various cognitive technologies (i.e., applications that help machines to “perceive, analyze, determine response and act appropriately in [their] environment”, a standard definition of AI), radio frequency identification, touchscreens/kiosks for customer interface, automated storage and retrieval systems, and automated guided vehicles.
Across all AI-related technologies, the aggregate adoption rate for all firms in the economy is 6.6% meaning that approximately 1 in 16 firms in the US are utilizing some form of AI in the workplace. This adoption rate is significantly lower than the adoption rate highlighted in the AI survey by the European Commission and other private surveys by McKinsey, Deloitte, and PwC. However, it is important to consider the sampling methods of those surveys. None of the other surveys claim to be nationally representative and tend to focus on larger, publicly traded companies. In contrast, the ABS sample includes many small firms where AI adoption is very low. This is important because AI adoption rate varies greatly by firm size. Adoption rates (defined as usage or testing) increase from 5.3% for the group of firms with the smallest number of employees to 62.5% for firms with 10,000+ employees.
In other words, scale appears to be a primary correlate of AI usage, likely due to both the large quantities of data and computing power required to fully realize the most popular types of AI currently available. This may potentially have far-reaching implications on topics such as inequality, competition and the rise of “superstar” firms, especially if AI is shown to have widespread productivity benefits. If only a select group of firms are able to fully realize the benefits of AI, we can expect further divergence for the “frontier” and most productive set of firms.
Our data and explanatory variables are simply too crude to provide a reliable predictor for the precise types of firms that adopt certain technologies and those that do not. While we can claim that size is a reliable predictor of adoption, even amongst large firms, we see heterogeneous patterns of adoption depending on the technology type. In other words, there are simply too many unknown factors that cannot be measured by traditional metrics (such as firm size, age and industry) that appear to drive technology adoption.
Conclusion
We have provided an introduction to the technology module in the 2018 ABS and placed it in the larger context of related work at the Census Bureau to collect comprehensive data on technology adoption and use by U.S. firms in order to provide a more accurate picture of the state of advanced technology use in the U.S. economy. Because of the large pool of respondents (about 850,000 firms) in the 2018 ABS, the module represents a unique opportunity to offer insights on technology adoption and use across all sectors of the economy and across a variety of key firm characteristics. The same technology module is expected to be a part of the 2021 Annual Business Survey.
A primary contribution for the paper is to develop a nationally representative set of technology adoption and use measures based on the survey results, which in public use tabulations report aggregate response counts for each technology question (see public use tabulations at: https://www.census.gov/data/tables/2018/econ/abs/2018-abs-digital-technology-module.html ).
Advanced Technologies Adoption And Use By U.S. Firms: Evidence From The Annual Business Survey, by Nikolas Zolas, Zachary Kroff, Erik Brynjolfsson, Kristina McElheran, David N. Beede, Cathy Buffington, Nathan Goldschlag, Lucia Foster and Emin Dinlersoz. 2020. National Bureau of Economic Research, Cambridge, MA Working Paper 28290 http://www.nber.org/papers/w28290
