LinkedIn

The Echo of the Past: The IT Productivity Paradox

You can see the computer age everywhere but in the productivity statistics.

— Robert Solow, Nobel Laureate in Economics (1987)

In the 1980s and 90s, despite widespread adoption of computers and significant investment in information technology, economic productivity growth remained surprisingly flat.

This "Solow Paradox" highlighted challenges in measuring intangible benefits, the lengthy time needed for organizational restructuring, and the effective integration of new technologies, offering crucial lessons for today's AI revolution.

1

The IT Productivity Paradox: Resolved

  1. Mismeasurement: Traditional metrics failed to capture the true value IT created, such as improved customer service, product variety, and faster delivery times. The benefits were real but invisible to traditional economic accounting.
  1. Lags: Radical technologies require a massive learning curve. It takes years for organizations to build the necessary infrastructure and fundamentally restructure their operations before bottom-line benefits materialize.
  1. Redistribution: Many IT investments merely shifted profits between competitors rather than expanding the total market. A firm might win market share using a new system, but the overall industry saw no net productivity gain.
  1. Mismanagement: Executives often threw technology at broken processes instead of redesigning the work. This poor decision-making created organizational slack and bottlenecks rather than actual efficiency.

2

IT as a Source of Firm Value

Academic consensus from 30+ years of firm-level empirical research

5-10x

Intangible capital complements IT investment (Brynjolfsson & Yang)

50-80%

Margin of return on organizational capital for ordinary capital

3 Pillars

IT + Org practices + Human capital = maximum value

HOW IT CREATES VALUE

Operational Efficiency

Automates routines, reduces transaction costs, lifts TFP and labor productivity (Brynjolfsson & Hitt)

Information & Decisions

Reduces internal/external asymmetries, enabling faster, better-quality managerial decisions (Mithas et al.)

Technology-Content Fit

IT value is usage-dependent and industry-specific; information-intensive sectors gain most (Zhu; Dewan & Kraemer)

Org. Complementarities

Flat hierarchies + empowered teams + IT = outsized gains; IT alone insufficient (Bresnahan, Brynjolfsson, Hitt 2002 )

Customer & Market Value

CRM-linked IT raises satisfaction and retention; value flows through demand-side, not just cost (Mithas)

Platform & Network Effects

IT enables ecosystems where network participation amplifies value beyond internal production (Zhu & Kraemer)

References

3

Making sense of the AI Productivity Paradox

Strategic imperatives for executives navigating the divide between AI's transformative potential and measurable enterprise returns.

4

Table of contents

Table of contents

5

The Early Optimism

GoldmanSachs

Generative AI could raise global GDP by 7%

Our signature newsletter with insights and analysis from across the firm

April 2023

McKinsey & Company

The economic potential of generative AI: The next productivity frontier

Our latest research estimates that generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually across the 63 use cases we analyzed

June 2023

6


7


8

There seems to be evidence of job displacement, suggesting AI productivity gains

Layoffs.fyi

Layoffs.fyi - Tech Layoff Tracker and DOGE Layoff Tracker

[LIVE] Tracking all tech startup layoffs — and lists of employees laid off — since COVID-19 was declared a pandemic. This page is constantly being updated.

CNN

Amazon just cut 14,000 jobs, and it’s not done | CNN Business

Amazon said it would cut 14,000 corporate staffers this year in a mass layoff aimed at readying the company for wide adoption of AI technology.

Reuters

Exclusive: Meta planning sweeping layoffs as AI costs mount

The cuts could affect 20% or more of the company.

9

However The Current Reality

The infamous study that shows 95% of GenAI Projects fail - MIT

MIT NANDA

State of AI in Business 2025

Despite $30–40 billion in enterprise investment into GenAI, 95% of organizations are getting zero return. (300+ public AI initiatives and announcements, and surveys with 153 leaders. )

The core barrier to scaling is not infrastructure, regulation, or talent. It is learning. MostGenAI systems do not retain feedback, adapt to context, or improve over time.

10

Recent Reversals

The Times of India

Ford rehires engineers it laid off as company VP Charles Poon admits that AI is ‘only as good as…’

Ford Motor Company has acknowledged that its push to replace experienced human workers with artificial intelligence (AI) backfired, forcing the American automaker to rehire hundreds of veteran engineers to fix automated quality control issues. According to a recent report by Bloomberg, over the past three years, Ford has brought back more than 350 retired or laid-off senior technical specialists, affectionately called ‘gray beards’ inside the company, to lead physical quality reviews.

Business Insider

McKinsey is doubling down on entry-level hires — even in the AI era

McKinsey & Company plans to boost hiring by 12% in 2026. The firm also plans to hire more entry-level workers, said North America chair Eric Kutcher.

Orgvue

55% of businesses admit wrong decisions in making employees redundant when bringing AI into the workforce

Research from Orgvue shows how businesses face workforce challenges with AI adoption and highlights smarter ways to manage redundancy decisions.

11

67% see no ROI from AI/GenAI- Harvey Nash


Harvey Nash

Harvey Nash Digital Leadership Report 2025

65% of digital leaders would choose an AI-enabled software developer with just 2 years’ experience, over one with a 5-year career but without AI skills.

12

Is this an AI bubble?

13

The AI (Productivity) Paradox

Generative AI promises immense value, yet its real-world impact presents a complex picture of potential alongside significant challenges.

Vast Economic Potential

Generative AI could unlock an estimated $2.6-$4.4 trillion in economic value globally, revolutionizing industries and processes.

Modest TFP Gains

Despite huge potential, long-term Total Factor Productivity (TFP) gains are forecast to be modest, remaining under 0.53% over the next decade.

https://notebooklm.google.com/notebook/9c6a71fd-7992-45d1-80d7-4bc1ff5fdfb4

14

Mismeasurement and Misattribution*

*a better way to measure is here Measuring the Impact of AI


GenAI Divide: 50% of GenAI budgets go to sales and marketing, but back-office automation often yields better ROI.


This bias reflects easier metric attribution, not actual value, and keeps organizations focused on the wrong priorities.



"If I buy a tool to help my team work faster, how do I quantify that impact? How do I justify it to my CEO when it won't directly move revenue or decrease measurable costs? I could argue it helps our scientists get their tools faster, but that's several degrees removed from bottom-lineimpact."

15

Task level: for instance the following show task level productivity of GenAI

Generative AI at Work | NBER

Call-center Operators

Findings from these papers

Jagged Frontier For tasks within AI's current capabilities, consultants using GPT-4 completed 12.2% more tasks, worked 25.1% faster, and produced 40% higher quality results. The technology also acted as a powerful "skill leveler," boosting the performance of bottom-tier consultants by 43%, compared to just a 17% boost for top performers. However, for complex tasks explicitly designed to be outside AI's capabilities, consultants using the tool actually performed worse, providing correct solutions 19 percentage points less often than those working without AI. This error rate spiked because workers tended to blindly trust the machine's output and "fall asleep at the wheel" Cybernetic Teammate AI Equals a Human Teammate: Individuals working alone with AI produced solutions of equal quality to two-person cross-functional teams working without AI AI Erases Functional Silos: Typically, R&D staff pitch technical ideas and commercial staff pitch market ideas. When given AI, individuals from either department produced highly balanced, cross-functional solutions on their own, essentially bridging their own knowledge gaps AI Boosts Morale: Contrary to fears of workplace isolation, professionals using AI reported significantly higher positive emotions (like excitement) and lower negative emotions (like frustration) than those working alone. The AI fulfilled the social and motivational roles usually provided by a human teammate GenAI in Call Centers Productivity Jump: AI assistance increased the number of issues resolved per hour by 14% to 15% on average The "Skill Leveling" Effect: The gains accrued almost entirely to novice and less-skilled workers, whose productivity spiked by roughly 35%. In contrast, highly skilled and experienced workers saw minimal gains, and occasionally slight declines in quality Improved Retention and Morale: Customer sentiment improved (with clients using happier words and fewer angry words), escalation requests to managers dropped by 25%, and employee turnover decreased significantly among newer agents

16

Think Tasks, Not Jobs

hai.stanford.edu

Presenting the 2025 AI Index Report | Stanford HAI

The AI Index, currently in its eighth year, tracks, collates, distills, and visualizes data relating to artificial intelligence.

17

Insights from Coding with AI


18

Insights from Coding with AI

Project Complexity Dictates AI Productivity Gains

AI boosts productivity by 30-40% for low-complexity "greenfield" tasks, but only 0-10% for high-complexity "brownfield" tasks.

Language Popularity Affects AI's Effectiveness

While effective for popular languages, AI can decrease productivity for obscure or niche programming languages.

Net Productivity Gains Are Lower Than Initial Boosts

Initial AI coding productivity boosts of 30-40% are reduced to a net 15-20% due to necessary rework and bug fixes.

Does AI Actually Boost Developer Productivity? (100k Devs Study) - Yegor Denisov-Blanch, Stanford

19

What is likely to work (MIT NANDA)

Thoughts to takeaway

What could we do with GenAI that is Low Complexity, Greenfield, in domains where we do have expertise.

20

Quantifying Human-AI Synergy

Riedl and Weidman

Users better able to infer and adapt to others’ perspectives achieve superior collaborative performance with AI—but not when working alone. Moreover, moment-to-moment fluctuations in perspective taking influence AI response quality, highlighting the role of dynamic user factors in collaboration.

Link to the paper

21

In a more general context…

Harvard Business Review

AI-Generated “Workslop” Is Destroying Productivity

Despite a surge in generative AI use across workplaces, most companies are seeing little measurable ROI. One possible reason is because AI tools are being used to produce “workslop”—content that appears polished but lacks real substance, offloading cognitive labor onto coworkers. Research from BetterUp Labs and Stanford found that 41% of workers have encountered such AI-generated output, costing nearly two hours of rework per instance and creating downstream productivity, trust, and collaboratio

22

Leads one to wonder "Is AI Creating Incompetent Experts?"

IE Insights

Is AI Creating Incompetent Experts? | IE Insights

Generative AI is short-circuiting the learning process that builds real expertise, writes Kiron Ravindran.

When a GPS fails, you know you’re lost. When a calculator malfunctions, the errors are likely quite obvious. But when ChatGPT fabricates plausible-sounding analysis in domains you personally don’t understand, the failure goes undetected.

Clark acknowledges we need new “metacognitive skills” to evaluate AI outputs – but that’s precisely what novices in the danger zone lack. You can’t develop judgment about what you don’t know.

23

Complementary Investments in Humans and Workplace Organization

www.dwarkesh.com

Tyler Cowen — The #1 bottleneck to AI progress is humans

Why he thinks AI won't drive explosive economic growth

- Team dynamics

- Incentives

- Task design
- Misguided Objectives



24

Yeah Right! more…


Did Claude really write all of Claude Cowork

25

Thoughts to takeaway

What could we do with GenAI that is Low Complexity, Greenfield, in domains where we do have expertise?

26

Augmentation vs. Substitution: why AI may not reduce costs

1

Augmentation Strategy

Human-AI complementarity drives employment gains, wage growth, sustained value creation

  • Airplane Pilots
  • Radiologists
  • Jevon's Paradox
  • Knowledge Economy
2

Substitution Strategy

Labor replacement approach correlates with declining headcount, limited productivity gains

27

28

Before we compare the AI "revolution" to the Industrial Revolution, consider this…

It took 5 decades after the Industrial Revolution began for working-class wages to rise, a phenomenon known as "Engels’ Pause"

29

Labor Market Disruption: Who's Actually Affected

Seniority-Biased Impact

Junior workers (ages 22-25) in exposed roles face structural hiring slowdowns

  • Software development
  • Customer service
  • Content creation

The Mechanism

Displacement driven by reduced hiring, not mass layoffs—creating invisible talent pipeline crisis

Strategic Implication

Organizations must redesign entry-level pathways to preserve institutional knowledge transfer

30

Seniority-biased Technical Change

Using U.S. resume and job posting data covering nearly 62 million workers in 285,000 firms (2015–2025), we track within-firm employment dynamics by seniority. We identify AI adoption through a text-analysis approach that flags postings for dedicated “AI integrator” roles, signaling active implementation of generative AI. Difference-in-differences and triple-difference estimates show that, beginning in 2023Q1, junior employment in adopting firms declined sharply relative to nonadopters, while senior employment continued to rise. The junior decline is driven primarily by slower hiring rather than increased separations, with the largest effects in wholesale and retail trade. Heterogeneity by education reveals a U-shaped pattern: mid-tier graduates see the largest declines, while elite and low-tier graduates are less affected. Overall, the results provide early evidence of a seniority-biased impact of AI adoption and its mechanisms. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5425555

This is very different from Skill-biased Technical Change

Computers replaced typists and favored young college educated candidates in white collar jobs

GenAI is hollowing out the middle with the young college educated finding it hardest to find jobs.

31

Recent grads are more unemployed on average

J.P. Morgan

AI’s Impact on Job Growth | J.P. Morgan Global Research

AI is poised to displace jobs, with some industries more at risk than others. Is the paradigm shift already underway?

32

Staying Relevant: 4 strategies from science

  1. Develop an "AI Theory of Mind" —the capacity to infer the AI's "mental state," knowledge gaps, and limitations—as the primary driver of successful collaboration
  1. The Insight: High "collaborative ability" with AI is mathematically distinct from your "solo ability",. Being smart on your own does not guarantee you will be smart with AI; you need the specific social-cognitive skill to navigate the AI's "alien" psychology,.
  1. Action: Treat the AI as a teammate rather than a search engine. Actively diagnose what context the AI is missing, anticipate its "hallucinations" (false beliefs), and repair communication breakdowns in real-time,.
  1. Target the "Hard" Tasks for Maximum Synergy Contrary to the belief that AI is only for rote automation, new benchmarks show that human-AI synergy is highest on the most difficult tasks.
  1. The Insight: While AI helps lower-skilled workers catch up (the "leveling" effect), it acts as a "cognitive amplifier" for complex problems where humans struggle alone.
  1. Action: Use AI for "hard" reasoning problems where humans need a cognitive boost (Riedl), but avoid relying on it for "complex" systemic integrations where the AI loses context (Denisov-Blanch).
  1. Become "Bilingual" (Domain + Algorithmic Literacy) You must combine deep domain expertise with algorithmic literacy.
  1. The Insight: Financial markets assign higher value to firms that decentralize algorithmic skills among domain experts rather than sequestering them in IT departments,
  1. Action: You do not need to be a coder, but you must understand how to apply AI tools to your specific industry context (e.g., marketing, law, biology) to bridge the gap between technical capability and business value,.
  1. Master "Centaur" and "Cyborg" Workflows To navigate the "Jagged Frontier" of AI capabilities, you must adopt specific modes of interaction.
  1. Centaur: Strategic delegation. You handle the tasks you are best at and hand off distinct sub-tasks to the AI. •
  1. Cyborg: Deep integration. You intertwine your thought process with the AI, moving back and forth rapidly to generate output that neither could produce alone.

33

Measuring the Impact of AI

34

The Hidden Value Story

$2.5T

Annual Consumer Surplus

Hidden AI welfare dividend invisible in GDP

6%

Equivalent GDP Impact

Of surveyed countries—uncaptured by traditional metrics

GDP-B: Digital Welfare

True economic benefit accrues to consumers as unmeasured surplus—reshaping how we assess AI's societal contribution

Stanford Digital Economy Lab

GDP-B: A New Way to Measure Growth and Well-Being in the Economy

This course will explore how the advances in AI can and will transform our economy and society in the coming years. Each week, we will hear from frontier researchers and industry leaders in technology, economics, and business, read the relevant research, and discuss the implications. Students will also have the opportunity to participate in one of eight optional dinners with the speakers.

35

Creating Intermediate Process Metrics to Manage Innovation Investments

36

Managing Innovation Investments by...

Developing Intermediate Process Metrics

  1. Identify the outcome goal: to maximize (Rev/Inv or NPS/Inv or Environmental Impact/ Inv... whatever)
  1. Break down the steps from investment to end target (from step 1)
  1. Create Process Goals: Output/Input metrics for each step
  1. Make sure that if all Numerators and Denominators cancel out leaving only the target to maximize (step 1)
  1. Set Performance Goal to track performance over time

Harvard Business Review

How To Really Measure a Company’s Innovation Prowess

In search of a better metric.

37

Similar to DuPont Analysis of Return on Equity (ROE)

The DuPont Analysis provides a powerful framework for dissecting a company's Return on Equity (ROE) into three core performance drivers:

Net Profit Margin

Measures profitability: how much profit is generated per dollar of sales. (Net Income / Sales)

Asset Turnover

Evaluates efficiency: how effectively assets are utilized to generate sales. (Sales / Assets)

Financial Leverage

Assesses solvency: the extent to which assets are financed by equity. (Assets / Equity)

38

The AI Productivity Paradox Notebook

notebooklm.google.com

AI Productivity Paradox

I am puzzled by the *AI Productivity Paradox*: While every one seems to be using (Gen)AI to do new things and old things faster, firms claim to not see any benefits in terms of ROI and at a macroeconomic level too the evidence seems to be scarce. This notebook has curated some of the most reliable sources so far to try an make sense of this paradox.

39

New information that have not yet been incorporated fully

  1. Microsoft's New Future of Work 2025
  1. https://arxiv.org/abs/2510.22780
  1. https://arxiv.org/html/2506.06576v2
  1. https://www.gartner.com/en/newsroom/press-releases/2026-01-15-gartner-predicts-60-percent-of-brands-will-use-agentic-ai-to-deliver-streamlined-one-to-one-interactions-by-2028
  1. https://www.pwc.com/m1/en/publications/agentic-ai-the-new-frontier-in-genai.html
  1. https://www.weforum.org/stories/2025/05/ai-agents-select-the-right-agent/
  1. https://hbr.org/2026/01/companies-are-laying-off-workers-because-of-ais-potential-not-its-performance
  1. AI-driven Productivity Growth Is Here. And It's Just Getting Started
  1. Reddit thread on Claude writing Claude Cowork https://www.reddit.com/r/ClaudeAI/comments/1qbscy5/claude_code_creator_boris_100_of_new_cowork_wrote/
  1. https://www.thestateofbrand.com/news/ai-job-cuts-not-improving-returns

40

References

  1. "Paradox Lost? Firm-Level Evidence on the Returns to Information Systems Spending" — Brynjolfsson, E. & Hitt, L. (1996). Management Science, 42(4), 541–558.
  1. "Beyond Computation: Information Technology, Organizational Transformation and Business Performance" — Brynjolfsson, E. & Hitt, L. M. (2000). Journal of Economic Perspectives, 14(4), 23–48.
  1. Bresnahan, T., Brynjolfsson, E. & Hitt, L.M. (2002). "Information Technology, Workplace Organization, and the Demand for Skilled Labor: Firm-Level Evidence." Quarterly Journal of Economics, 117(1), 339–376.
  1. Brynjolfsson, E., Hitt, L.M. & Yang, S. (2002). "Intangible Assets: Computers and Organizational Capital." Brookings Papers on Economic Activity, 2002(1), 137–198.
  1. Dewan, S. & Kraemer, K.L. (2000). "Information Technology and Productivity: Evidence from Country-Level Data." Management Science, 46(4), 548–562.
  1. Mithas, S., Krishnan, M.S. & Fornell, C. (2005). "Why Do Customer Relationship Management Applications Affect Customer Satisfaction?" Journal of Marketing, 69(4), 201–209.
  1. Mithas, S., Ramasubbu, N. & Sambamurthy, V. (2011). "How Information Management Capability Influences Firm Performance." MIS Quarterly, 237–256.
  1. Zhu, K. & Kraemer, K.L. (2002). "e-Commerce Metrics for Net-Enhanced Organizations." Information Systems Research, 13(3), 275–295.

41