Three decades of workforce data shows how AI reduces organizational layers and broadens teams
Listen to this article:
Conversations about AI and the shape of companies typically happen in the future tense. Will organizations need fewer employees? How will departments be restructured? What will happen to managers? Questions like these sound new because modern AI tools are so novel that their effects remain uncertain. But the AI organizational story already has a prehistory: firms have been incorporating AI-related capabilities long before ChatGPT broke onto the scene. This important point doesn’t go overlooked in the new working paper “AI Exposure and Organizational Structure,” co-written by Feng Zhu, co-lead of the Platform Lab at the HBS AI Institute. Using data from 1990 through 2020, Zhu and his co-author Guohou Shan ask whether exposure to AI reshapes two foundational features of organizational design: the number of hierarchical ranks a firm carries, and how many people each manager oversees. Executives have predicted flattening for years. Does it actually show up in large-scale organizational data?
Key Insight: Counting AI’s Entry into the Firm
“This measure captures the point at which AI-related human capital becomes embedded within the firm’s workforce, reflecting meaningful organizational exposure to AI rather than indirect use through vendors or external consultants.” [1]
The researchers examined 7,681 U.S. firms primarily through workforce data from Revelio Labs. Revelio’s records capture employees’ roles and career histories alongside information such as education, professional skills, and estimated seniority. Aggregating those records to the firm-year level allows the researchers to reconstruct organizational structures over time, producing 171,836 observations. The study defines “AI exposure” specifically as when a firm employs at least one worker whose listed skills include artificial intelligence or related computational techniques. They then track two structural outcomes: “hierarchical ranks” capture how many seniority levels separate the organization vertically (more ranks mean a taller organization), while “span of control” captures the typical size of groups reporting to a manager (wider spans are consistent with flatter organizations). The authors acknowledge that both are inferred from titles and predicted seniority rather than directly based on internal organization charts. Because firms cross the AI-exposure threshold at different moments, the authors use a staggered “difference-in-differences” design: comparing how a firm’s structure moves around its own exposure date against firms not yet exposed, holding constant fixed firm characteristics, common time trends, and size. That carries more weight than comparing AI firms with non-AI firms at one moment.
Key Insight: The AI Structural Shift
“These results suggest that AI acts as a managerial augmentation technology, enabling firms to compress vertical layers while expanding the scope of oversight for remaining managers.” [2]
The primary result is a flattening pattern. AI exposure is associated with 0.203 fewer hierarchical ranks, about a 3.21% decline relative to the sample mean, and a 0.0278 increase in span of control, about 0.83% relative to the mean. Together, the estimates point to fewer vertical levels alongside somewhat broader managerial oversight. The span of control effect appears across the hierarchy, but is much larger toward the top, and the effects are also stronger where firms have more technically trained employees. The authors interpret this through the theory of managerial augmentation: AI-related capabilities may produce more structural change when people have the expertise to understand and incorporate them. Industry context matters too. The effects are stronger in high-tech industries and in industries with more routine tasks. But firm size adds a wrinkle: larger organizations show a stronger widening of spans but a less pronounced reduction in ranks. AI may therefore help large companies extend managers’ reach without making the organizational layers required to run a complex enterprise disappear altogether.
Why This Matters
One of the paper’s most useful implications is that organizational flattening is not an automatic consequence of AI. Structure moved where technical capability, industry conditions, and scale made movement possible, and moved less where they didn’t. But technology doesn’t redesign an organization, leaders do. Leaders still have to decide where and how information should flow, who owns decisions, how accountability works, and where human coordination is still important. For business leaders and executives, the practical move is sequencing: watch where AI actually expands managerial capacity before widening spans or removing layers, and invest in the technical fluency that lets managers use what they’ve been given.
References & Notes
[1] Shan, Guohou and Feng Zhu, “AI Exposure and Organizational Structure” (May, 2026): 9.
[2] Shan and Zhu, “AI Exposure and Organizational Structure,” 19.
This article was created through a human-led, AI-assisted process.
Meet the Authors
Guohou Shan is Assistant Professor, Supply Chain & Information Management, at the D’Amore-McKin School of Business of Northeastern University.
Feng Zhu is Harry E. Figgie, Jr. Professor of Business Administration at Harvard Business School and lead of the Platform Lab at the HBS AI Institute.