The Microstructure of AI Diffusion: Evidence From Firms, Business Functions, and Worker Tasks
April 2026
Working Paper Number:
CES-26-25
Abstract
Using novel, nationally representative data from the 2026 AI supplement to the U.S. Census Bureau's Business Trends and Outlook Survey (BTOS), we characterize AI diffusion across three interconnected layers: overall firm use, deployment across business functions, and worker-task use. This multi-layered approach provides a nuanced picture of business AI adoption. During the supplement reference period (Nov 2025-Jan 2026), 18% of firms used AI in a business function, rising to 32% on an employment-weighted basis; adoption is expected to reach 22% within six months. AI use is substantially higher in large firms and knowledge-intensive sectors, with use rates reaching 50%-60% (60%-70%, employment-weighted) for very large firms in the Information, Professional Services, and Finance sectors. Among adopting firms, the scope of use remains limited: 57% of users integrate AI in three or fewer business functions, most commonly Sales and Marketing (52%), Strategy and Business Development (45%), and IT (41%). In 23% (41%, employment-weighted) of firms, workers use AI in work-related tasks. Writing, document analysis, and information search are the leading Generative AI use in tasks, though 65% of firms limit use to three or fewer tasks. The evidence points to both top-down and bottom-up diffusion channels: worker task use sometimes occurs without formal firm-level adoption, and firm-level adoption sometimes occurs without worker task use. Most users (66%) rely on AI solely to augment tasks, while AI-related employment decreases are rare, occurring in only 2% of firms. Regression analysis shows a robust positive correlation between firm commercial performance and the breadth of AI integration, including functional deployment, task-level use, and operational investment. A distinct divergence emerges, however, with respect to labor outcomes. Functional breadth and operational investment are positively associated with employment decreases, whereas worker-task integration shows no significant link to headcount reduction once functional integration and operational investment are taken into account.
Document Tags and Keywords
Keywords
Keywords are automatically generated using KeyBERT, a powerful and innovative
keyword extraction tool that utilizes BERT embeddings to ensure high-quality and contextually relevant
keywords.
By analyzing the content of working papers, KeyBERT identifies terms and phrases that capture the essence of the
text, highlighting the most significant topics and trends. This approach not only enhances searchability but
provides connections that go beyond potentially domain-specific author-defined keywords.
:
enterprise,
agency,
technological,
organizational,
corporate,
trend,
incorporated,
strategic,
innovator,
workforce,
warehousing,
marketing
Tags
Tags are automatically generated using a pretrained language model from spaCy, which excels at
several tasks, including entity tagging.
The model is able to label words and phrases by part-of-speech,
including "organizations." By filtering for frequent words and phrases labeled as "organizations", papers are
identified to contain references to specific institutions, datasets, and other organizations.
:
Bureau of Labor Statistics,
Ordinary Least Squares,
Office of Management and Budget,
University of Maryland,
American Immigration Council,
Federal Reserve Bank,
Current Population Survey,
Economic Census,
Educational Services,
North American Industry Classification System,
Technical Services,
Duke University,
Disclosure Review Board,
Linear Probability Models,
Arts, Entertainment,
Annual Business Survey,
Professional Services,
Meyer et al
Similar Working Papers
Similarity between working papers are determined by an unsupervised neural
network model
know as Doc2Vec.
Doc2Vec is a model that represents entire documents as fixed-length vectors, allowing for the
capture of semantic meaning in a way that relates to the context of words within the document. The model learns to
associate a unique vector with each document while simultaneously learning word vectors, enabling tasks such as
document classification, clustering, and similarity detection by preserving the order and structure of words. The
document vectors are compared using cosine similarity/distance to determine the most similar working papers.
Papers identified with 🔥 are in the top 20% of similarity.
The 10 most similar working papers to the working paper 'The Microstructure of AI Diffusion: Evidence From Firms, Business Functions, and Worker Tasks' are listed below in order of similarity.
-
Working PaperTracking Firm Use of AI in Real Time: A Snapshot from the Business Trends and Outlook Survey🔥
March 2024
Working Paper Number:
CES-24-16R
Timely and accurate measurement of AI use by firms is both challenging and crucial for understanding the impacts of AI on the U.S. economy. We provide new, real-time estimates of current and expected future use of AI for business purposes based on the Business Trends and Outlook Survey for September 2023 to February 2024. During this period, bi-weekly estimates of AI use rate rose from 3.7% to 5.4%, with an expected rate of about 6.6% by early Fall 2024. The fraction of workers at businesses that use AI is higher, especially for large businesses and in the Information sector. AI use is higher in large firms but the relationship between AI use and firm size is non-monotonic. In contrast, AI use is higher in young firms. Common uses of AI include marketing automation, virtual agents, and data/text analytics. AI users often utilize AI to substitute for worker tasks and equipment/software, but few report reductions in employment due to AI use. Many firms undergo organizational changes to accommodate AI, particularly by training staff, developing new workflows, and purchasing cloud services/storage. AI users also exhibit better overall performance and higher incidence of employment expansion compared to other businesses. The most common reason for non-adoption is the inapplicability of AI to the business.View Full Paper PDF
-
Working PaperAutomation and the Workforce: A Firm-Level View from the 2019 Annual Business Survey
April 2022
Working Paper Number:
CES-22-12R
This paper describes the adoption of automation technologies by US firms across all economic sectors by leveraging a new module introduced in the 2019 Annual Business Survey, conducted by the US Census Bureau in partnership with the National Center for Science and Engineering Statistics (NCSES). The module collects data from over 300,000 firms on the use of five advanced technologies: AI, robotics, dedicated equipment, specialized software, and cloud computing. The adoption of these technologies remains low (especially for AI and robotics), varies substantially across industries, and concentrates on large and young firms. However, because larger firms are much more likely to adopt them, 12-64% of US workers and 22-72% of manufacturing workers are exposed to these technologies. Firms report a variety of motivations for adoption, including automating tasks previously performed by labor. Consistent with the use of these technologies for automation, adopters have higher labor productivity and lower labor shares. In particular, the use of these technologies is associated with a 11.4% higher labor productivity, which accounts for 20'30% of the difference in labor productivity between large firms and the median firm in an industry. Adopters report that these technologies raised skill requirements and led to greater demand for skilled labor, but brought limited or ambiguous effects to their employment levels.View Full Paper PDF
-
Working PaperAdvanced Technologies Adoption and Use by U.S. Firms: Evidence from the Annual Business Survey
December 2020
Working Paper Number:
CES-20-40
We introduce a new survey module intended to complement and expand research on the causes and consequences of advanced technology adoption. The 2018 Annual Business Survey (ABS), conducted by the Census Bureau in partnership with the National Center for Science and Engineering Statistics (NCSES), provides comprehensive and timely information on the diffusion among U.S. firms of advanced technologies including artificial intelligence (AI), cloud computing, robotics, and the digitization of business information. The 2018 ABS is a large, nationally representative sample of over 850,000 firms covering all private, nonfarm sectors of the economy. We describe the motivation for and development of the technology module in the ABS, as well as provide a first look at technology adoption and use patterns across firms and sectors. We find that digitization is quite widespread, as is some use of cloud computing. In contrast, advanced technology adoption is rare and generally skewed towards larger and older firms. Adoption patterns are consistent with a hierarchy of increasing technological sophistication, in which most firms that adopt AI or other advanced business technologies also use the other, more widely diffused technologies. Finally, while few firms are at the technology frontier, they tend to be large so technology exposure of the average worker is significantly higher. This new data will be available to qualified researchers on approved projects in the Federal Statistical Research Data Center network.View Full Paper PDF
-
Working PaperStarting Up AI
March 2024
Working Paper Number:
CES-24-09R
Using comprehensive administrative data on business applications over the period 2004- 2023, we study business applications (ideas) and the resulting startups that aim to develop AI technologies or produce goods or services that use, integrate, or rely on AI. The annual number of new AI-related business applications is stable between 2004 and 2011, but begins to rise in 2012 with further increases from 2016 onward into the Covid-19 pandemic and beyond, with a large, discrete jump in 2023. The distribution of these applications is highly uneven across states and sectors. AI business applications have a higher likelihood of becoming employer startups compared to other applications. Moreover, businesses originating from these applications exhibit higher revenue, average wage, and labor share, but similar labor productivity and lower survival rate, compared to other businesses. While it is still early in the diffusion of AI, the rapid rise in AI business applications, combined with the better performance of resulting businesses in several key outcomes, suggests a growing contribution from AI-related business formation to business dynamism.View Full Paper PDF
-
Working PaperAI Adoption in America: Who, What, and Where
September 2023
Working Paper Number:
CES-23-48R
We study the early adoption and diffusion of five AI-related technologies (automated-guided vehicles, machine learning, machine vision, natural language processing, and voice recognition) as documented in the 2018 Annual Business Survey of 850,000 firms across the United States. We find that fewer than 6% of firms used any of the AI-related technologies we measure, though most very large firms reported at least some AI use. Weighted by employment, average adoption was just over 18%. AI use in production, while varying considerably by industry, nevertheless was found in every sector of the economy and clustered with emerging technologies such as cloud computing and robotics. Among dynamic young firms, AI use was highest alongside more educated, more-experienced, and younger owners, including owners motivated by bringing new ideas to market or helping the community. AI adoption was also more common alongside indicators of high-growth entrepreneurship, including venture capital funding, recent product and process innovation, and growth-oriented business strategies. Early adoption was far from evenly distributed: a handful of 'superstar' cities and emerging hubs led startups' adoption of AI. These patterns of early AI use foreshadow economic and social impacts far beyond this limited initial diffusion, with the possibility of a growing 'AI divide' if early patterns persist.View Full Paper PDF
-
Working PaperYou're (not) Hired: Artificial Intelligence and Early Career Hiring in the Quarterly Workforce Indicators
April 2026
Working Paper Number:
CES-26-27
Using detailed tabulations from matched employer-employee administrative data, I document evidence of an immediate, sizable, and persistent decrease in the level of early career (22-24 year old) hires following introduction of ChatGPT within the industry-state cells that are most exposed to AI. The decline in hires is the primary cause of large observed declines in employment over the subsequent period. Regressionadjusted employment of early career workers in the most AI-exposed quintile of industry-state cells declined by 12% over the 10 quarters following the introduction of ChatGPT, even as employment in lessexposed industries has remained stable. The rate of hiring largely recovered by early 2025, attributable to a smaller employment base. Earnings growth of early career workers in the most exposed industries slowed slightly relative to those in less exposed industries. Although the most AI-exposed quintile of detailed industries is dominated by a handful of industry sectors, I find that the association of higher AI exposure with reduced early career employment and fewer hires is observed across most sectors of the economy. Timing of effects in event studies is consistent with an immediate effect on hiring following introduction of ChatGPT. However, triple difference estimates provide some evidence of earlier trend shifts on employment, hiring, and separations around the onset of the COVID pandemic. I discuss potential explanations, including the increase in remote work and increased educational attainment among workers in AI-exposed occupations. Nonetheless, job gains to early career workers and backfill hires show evidence of discontinuous decline at the time of ChatGPT's release in comparison to older workers in the same industries. A local projections analysis at the NAICS industry group level shows that industries with high AI exposure are not particularly sensitive to unexpected fluctuations in monetary policy on average relative to other industries in employment, hiring, or separations. A historical decomposition suggests that up to one quarter of relative early career employment declines through 2025q2 may be attributable to monetary policy shocks through 2023, but the analysis does not find evidence that these shocks can explain the rapid decline in hires at the most AI-exposed firms in comparison to others.View Full Paper PDF
-
Working PaperTapping Business and Household Surveys to Sharpen Our View of Work from Home
June 2025
Working Paper Number:
CES-25-36
Timely business-level measures of work from home (WFH) are scarce for the U.S. economy. We review prior survey-based efforts to quantify the incidence and character of WFH and describe new questions that we developed and fielded for the Business Trends and Outlook Survey (BTOS). Drawing on more than 150,000 firm-level responses to the BTOS, we obtain four main findings. First, nearly a third of businesses have employees who work from home, with tremendous variation across sectors. The share of businesses with WFH employees is nearly ten times larger in the Information sector than in Accommodation and Food Services. Second, employees work from home about 1 day per week, on average, and businesses expect similar WFH levels in five years. Third, feasibility aside, businesses' largest concern with WFH relates to productivity. Seven percent of businesses find that onsite work is more productive, while two percent find that WFH is more productive. Fourth, there is a low level of tracking and monitoring of WFH activities, with 70% of firms reporting they do not track employee days in the office and 75% reporting they do not monitor employees when they work from home. These lessons serve as a starting point for enhancing WFH-related content in the American Community Survey and other household surveys.View Full Paper PDF
-
Working PaperThe Rise of Industrial AI in America: Microfoundations of the Productivity J-curve(s)
April 2025
Working Paper Number:
CES-25-27
We examine the prevalence and productivity dynamics of artificial intelligence (AI) in American manufacturing. Working with the Census Bureau to collect detailed large-scale data for 2017 and 2021, we focus on AI-related technologies with industrial applications. We find causal evidence of J-curve-shaped returns, where short-term performance losses precede longer-term gains. Consistent with costly adjustment taking place within core production processes, industrial AI use increases work-in-progress inventory, investment in industrial robots, and labor shedding, while harming productivity and profitability in the short run. These losses are unevenly distributed, concentrating among older businesses while being mitigated by growth-oriented business strategies and within-firm spillovers. Dynamics, however, matter: earlier (pre-2017) adopters exhibit stronger growth over time, conditional on survival. Notably, among older establishments, abandonment of structured production-management practices accounts for roughly one-third of these losses, revealing a specific channel through which intangible factors shape AI's impact. Taken together, these results provide novel evidence on the microfoundations of technology J-curves, identifying mechanisms and illuminating how and why they differ across firm types. These findings extend our understanding of modern General Purpose Technologies, explaining why their economic impact'exemplified here by AI'may initially disappoint, particularly in contexts dominated by older, established firms.View Full Paper PDF
-
Working PaperHigh Frequency Business Dynamics in the United States During the COVID-19 Pandemic
March 2021
Working Paper Number:
CES-21-06
Existing small businesses experienced very sharp declines in activity, business sentiment, and expectations early in the pandemic. While there has been some recovery since the early days of the pandemic, small businesses continued to exhibit indicators of negative growth, business sentiment, and expectations through the first week of January 2021. These findings are from a unique high frequency, real time survey of small employer businesses, the Census Bureau's Small Business Pulse Survey (SBPS). Findings from the SBPS show substantial variation across sectors in the outcomes for small businesses. Small businesses in Accommodation and Food Services have been hit especially hard relative to those Finance and Insurance. However, even in Finance and Insurance small businesses exhibit indicators of negative growth, business sentiment, and expectations for all weeks from late April 2020 through the first week of 2021. While existing small businesses have fared poorly, after an initial decline, there has been a surge in new business applications based on the high frequency, real time Business Formation Statistics (BFS). Most of these applications are for likely nonemployers that are out of scope for the SBPS. However, there has also been a surge in new applications for likely employers. The surge in applications has been especially apparent in Retail Trade (and especially Non-store Retailers). We compare and contrast the patterns from these two new high frequency data products that provide novel insights into the distinct patterns of dynamics for existing small businesses relative to new business formations.View Full Paper PDF
-
Working PaperU.S. Productivity and Electronic Processes in Manufacturing
October 2001
Working Paper Number:
CES-01-11
Recent studies argue that the use of information technology is a significant source of U.S. productivity growth. Official U.S. data on this use have been scarce. New official data on the use of electronic business processes (business processes such as procurement, payroll, inventory, etc.,conducted over computer networks) in the manufacturing sector of the United States were recently released. Preliminary estimates based on these data are consistent with some results in the literature. However, they also raise questions requiring additional detailed micro data analysis.View Full Paper PDF
