📈 economics

From Clerks to Agentic-AI: How will Technology Change Labor Market in Finance?

This paper documents the evolution of labor requirements in the finance industry across three major technological waves by analyzing changes in assets under management, revenue, and operating expenses per employee, aiming to establish stylized facts about how technology scales asset management work rather than identifying causal effects.

Original authors: Lu Yu, Xiang Li

Published 2026-04-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lu Yu, Xiang Li

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: It's Not About Robots Taking Jobs (Yet)

Imagine the financial world as a giant, busy kitchen. For decades, the chefs (financial workers) have been chopping vegetables, stirring pots, and plating food.

This paper asks a simple question: What happens when we introduce new kitchen gadgets?

The authors argue that when new technology arrives in finance (like computers, then index funds, and now AI), it doesn't immediately fire the chefs. Instead, it changes how the kitchen is organized. The chefs stop chopping vegetables by hand and start supervising the food processors. The kitchen gets bigger and faster, but the number of people doesn't drop right away.

The paper looks at three specific "waves" of technology to see how this plays out:

The Computerization Wave (1980s-90s): Like giving everyone a calculator and a typewriter.
The Indexing Wave (2000s-2015): Like creating a recipe book where you just follow strict rules for a salad, rather than guessing the ingredients.
The Agentic AI Wave (Now): Like having a robot assistant that can not only chop but also write the shopping list, check the inventory, and call the supplier.

The Three Waves of Change

1. The Computer Era: "The Super-Tool"

In the past, financial workers used spreadsheets and terminals. This didn't replace the workers; it just made them faster. It was like giving a carpenter a power drill instead of a hand saw. The work got done quicker, but the carpenter was still needed to decide what to build.

2. The Indexing Era: "The Recipe Book"

Then came "passive investing" (like ETFs). This was like automating the salad-making. You didn't need a chef to taste-test every leaf; you just followed a strict recipe. This replaced some specific jobs (the ones doing repetitive salad prep), but it didn't change the whole kitchen.

3. The AI Era: "The Smart Assistant"

Now we have AI. This is different. It's not just a calculator or a recipe book; it's a smart assistant that can read emails, summarize reports, draft contracts, and triage problems.

The Catch: Unlike the salad recipe, this assistant can do a huge variety of tasks, from the boring paperwork to the complex analysis.

What Did They Actually Find?

The authors looked at big banks and financial firms (like JPMorgan, Bank of America, and S&P Global) to see what happens when these companies start talking more about AI in their official reports.

Here are the three main takeaways, explained simply:

1. Productivity Goes Up, But Jobs Don't Vanish Immediately

When companies start using more AI, they get more done per person.

The Analogy: Imagine a team of 10 people managing a lemonade stand. With AI, they can manage 50 stands.
The Result: The paper found that firms using AI are managing more assets (more lemonade stands) per employee. However, they haven't necessarily fired people yet. The "labor cost" per person hasn't dropped sharply.
Why? Because the company is likely reassigning those workers to new tasks (like supervising the AI or talking to clients) rather than firing them. The "reorganization" happens before the "layoffs."

2. The "Middle" is in Trouble

This is the most important part for the future of work.

The Big Guys: Large banks have huge data and money. They use AI to get even bigger and stronger. They are the "supermarkets" of finance.
The Small Teams: Small, agile teams can now use AI to do work that used to require a whole department. They become "specialty shops" that are surprisingly powerful.
The Middle Layer: This is the group in the middle—people whose main job was just coordinating information, checking boxes, or passing messages between departments.
- The Metaphor: Think of the middle layer as the "human internet cables" connecting two computers. If you replace the cables with a direct fiber-optic line (AI), the cables aren't needed anymore.
- The Risk: Jobs that are mostly about "middle management" or "information coordination" are the most at risk because AI is very good at connecting dots and summarizing data.

3. It's a Slow Burn, Not a Flash

The study shows that companies adopt AI at different speeds. Some (like Bank of America) started early; others are just starting.

The Analogy: It's like a neighborhood switching to electric cars. Some people bought them in 2015; others are just getting them in 2024. The change is happening, but it's not a sudden "boom" where everyone switches overnight.

The Bottom Line: Transformation, Not Termination

The paper concludes that we shouldn't panic and think AI will wipe out all finance jobs tomorrow. Instead, think of it as a renovation.

Old Way: Humans do the boring math and data entry; humans also do the big decisions.
New Way: AI does the boring math and data entry. Humans focus entirely on the big decisions, client relationships, and supervising the AI.

The Future Landscape:

Big Banks will get bigger and more efficient.
Small Teams will become super-capable because AI lowers the cost of doing complex work.
The "Middle" (the coordinators and data processors) will face the most pressure because their specific skills are the easiest for AI to copy.

In short: Technology isn't just replacing workers; it's changing the shape of the workforce. It's making the top and bottom stronger, while squeezing the middle. The workers who survive will be the ones who learn to work with the robot assistant, not the ones trying to compete with it on speed.

1. Problem Statement and Research Question

The paper addresses a critical gap in understanding the economic impact of Agentic AI (autonomous systems capable of monitoring, summarizing, and coordinating workflows) on the financial labor market.

The Core Question: Does AI simply eliminate jobs, or does it fundamentally reorganize workflows, scale, and the distribution of capabilities within financial firms?
The Context: Finance is an ideal setting for studying automation due to its mix of standardized workflows, information processing, and judgment-intensive decision-making. The authors argue that previous technological waves (computerization, indexing) did not merely reduce headcount but changed who could operate effectively and how firms scaled.
The Hypothesis: Successive automation waves reorganize labor and scale before labor-cost adjustments become visible. The result is likely a polarized industry where large institutions retain structural advantages, small teams gain capability, and middle-layer coordination functions face the most pressure.

2. Methodology

The study employs a mixed-method approach combining historical analysis with quantitative econometric modeling on firm-level data.

A. Data and Sample

Primary Dataset: A panel of publicly listed U.S. financial firms derived from SEC filings (10-Ks) and accounting data.
Sample Size: The broader design targets ~100 firms (expandable to 300). The current AI-specific analysis focuses on a pilot sample of five major institutions: Bank of America, BNY Mellon, JPMorgan Chase, S&P Global, and State Street.
Time Period: The analysis spans three distinct technological eras defined by calendar cutoffs:
1. Computerization: 1985–2000
2. Indexing: 2000–2015
3. AI/Automation: 2015–Present

B. Variable Construction

Productivity Measures:
- Primary: Revenue per Employee (broad operating productivity).
- Secondary: Assets Under Management (AUM) per Employee (scale in info-processing businesses) and Labor Expense per Employee (proxy for labor-cost compression).
Automation Intensity Measure (The Key Innovation):
- The authors construct a filing-based automation intensity index using Natural Language Processing (NLP) on annual reports.
- Era-Specific Vocabulary:
  - Computerization: "Computers," "terminals," "spreadsheets," "electronic trading."
  - Indexing: "ETFs," "passive management," "benchmark replication," "rules-based allocation."
  - AI Era: "Artificial intelligence," "machine learning," "generative AI," "autonomous agents," "robotic process automation (RPA)."
- Calculation: The count of relevant terms is weighted (direct AI terms weighted higher than adjacent automation terms) and normalized by the total word count of the filing to create an exposure variable ( $ai\_exposure_{it}$ ).

C. Econometric Framework

The study utilizes a Fixed-Effects Panel Specification to test for associations between automation intensity and operating outcomes.

Baseline Model:
$\log(1 + \text{outcome}_{it}) = \beta \cdot \text{ai\_exposure}_{it} + \text{Firm FE} + \text{Year FE} + \epsilon_{it}$
Design Logic: This approach absorbs time-invariant firm characteristics and common macro shocks. The authors explicitly state this is for descriptive association rather than strict causal identification, aiming to observe within-firm changes as automation intensity rises.
Heterogeneity Analysis: The framework allows for comparisons across business models (e.g., Asset Managers vs. Banks, Passive vs. Active firms).

3. Key Results

A. Productivity Trends Across Eras

Historical Progression: Regression evidence shows that both Revenue per Employee and AUM per Employee are substantially higher in later eras compared to the pre-computerization benchmark.
Coefficients: In the Revenue specification, era coefficients rise from 1.4070 (Computerization) to 2.0543 (Indexing) to 2.3688 (AI). Similar upward trends are observed for AUM.
Interpretation: Productivity growth in finance is cumulative, driven by successive waves of technological and organizational change.

B. AI Exposure and Operating Outcomes (The AI Era)

The analysis of the five-firm pilot sample reveals a nuanced pattern:

Revenue per Employee: Negative association ( $\beta = -0.0535$ ). Firms with higher AI disclosure intensity show lower revenue intensity per worker.
AUM per Employee: Strong positive association ( $\beta = 0.5843$ ). Firms with higher AI exposure manage significantly more assets per employee.
Labor Expense per Employee: Statistically indistinguishable from zero ( $\beta = 0.0107$ ).

Synthesis: The combination of higher scale (AUM), lower revenue intensity, and stable labor costs suggests that AI is currently being used for reorganization, reinvestment, and output-mix changes rather than immediate headcount reduction. Firms are expanding their capacity to monitor and manage assets without a proportional increase in labor costs.

C. Labor Cost Dynamics

Lag Effect: Labor costs do not adjust in lockstep with output measures. The data suggests automation initially affects scale and task allocation (augmentation) before translating into visible labor-cost compression or headcount reduction.
Adoption Heterogeneity: Event-study style analysis reveals that AI adoption is staggered across firms (e.g., Bank of America adopted earlier than BNY Mellon), indicating organization-specific mediation rather than a synchronized sector-wide shock.

4. Key Contributions

Historical Framework: The paper successfully embeds the current "Agentic AI" wave within a longer historical sequence of Computerization and Indexing, arguing that AI extends automation from infrastructure to "cognition-adjacent" tasks (monitoring, drafting, triage).
Novel Measurement: It introduces a filing-based automation intensity metric that maps specific technological eras to firm-level disclosures, allowing for a granular analysis of adoption without relying solely on survey data.
Reframing the Narrative: It challenges the "job elimination" narrative, proposing instead that automation in finance leads to workflow reorganization and scale expansion first, with labor displacement being a secondary, slower-moving effect.

5. Significance and Implications

For the Labor Market: The findings predict a polarized industry structure:
- Large Institutions: Will retain structural advantages due to proprietary data, infrastructure, and compliance depth.
- Small Teams: Will become more capable as AI lowers the cost of research support and operational coverage.
- Middle Layer: Faces the greatest risk, as their historical value (coordinating information flows) is the specific domain AI is most likely to standardize and automate.
For Policy and Strategy: The results suggest that the immediate impact of AI is not a "shock" to employment but a gradual shift in the composition of work. Firms are likely to shift human labor toward supervision, exception handling, and client trust, while software absorbs routine tasks.
Theoretical Alignment: The findings align with Autor's task-based framework (substituting routine tasks, complementing abstract ones) and Acemoglu/Restrepo's distinction between automation and augmentation.

6. Limitations

The authors acknowledge several constraints:

Disclosure vs. Implementation: Filing-based measures capture disclosure intensity, which may not perfectly correlate with actual operational adoption.
Sample Size: The AI-specific analysis is limited to five firms, making it a "structured pilot" rather than a definitive sector-wide estimate.
Causality: The fixed-effects design is descriptive; it cannot fully separate causality from reverse causality or omitted organizational variables.
Stylized Eras: The three-era classification is a heuristic framework and not a sharp causal partition of technological change.

Conclusion

The paper concludes that Agentic AI is transforming the financial labor market through task reallocation and workflow redesign rather than uniform labor replacement. The immediate evidence points to increased scalability and productivity (specifically in AUM management) with a lag in labor-cost adjustments, suggesting a future where the financial industry is more polarized and capable, but where the "middle" of the workforce faces significant structural pressure.