White Paper

WBN RESEARCH DESK AI multiplier effect could amplify job loss forecasts by 1.5–2× | WEF: 92M jobs displaced globally by 2030 | Wage income velocity 3–4× higher than capital income | Pigouvian automation tax: the only corrective instrument that works | Block cuts 4,000 workers; Salesforce replaces 4,000 support agents with AI | Red Queen effect: better AI widens the distortion, not resolves it | Over-automation harms both workers AND firm owners — a deadweight loss, not a transfer |

⬤ WHITE PAPER · CLIENT RESEARCH SERIES · ECONOMICS & LABOUR MARKETS

The AI Multiplier Effect

How AI-Driven Job Losses Cascade Through the Economy

WBN Research Desk May 31, 2026 HIGH IMPACT Based on: Hemenway Falk & Tsoukalas (2026)

Overview

The Cascade

Competitive Trap

Policy

Implications

Conclusion

Direct Jobs Displaced (Global, by 2030)

92M

World Economic Forum projection across 55 economies

True Impact with Multiplier

138–184M

Applying standard 1.5–2× Keynesian employment multiplier

Wage vs. Capital Velocity

3–4×

Each $1 of wage income recirculates 3–4× more than $1 of capital income

U.S. Workers Exposed to LLM Automation

80%

Eloundou et al. (2024) — tasks susceptible to large language model displacement

The Core Problem

What Standard Forecasts Miss

FOUNDATIONAL

Standard economic forecasts count direct job losses from AI automation. They stop there. What they systematically miss is the velocity effect — the cascade of lost spending that propagates through the entire economy each time a worker is displaced.

This is not a theoretical risk. It is a structural feature of how modern economies work, and it causes conventional job-loss forecasts to understate the true economic damage — in some scenarios by a factor of two or more.

⬤ Key Insight

The Displacement Cascade

When AI displaces a worker, it does not eliminate one job. It eliminates that worker's spending — which was someone else's revenue. That revenue loss then threatens further jobs, which threatens further spending, in a self-reinforcing cycle that headline forecasts do not capture.

Why Headlines Undercount the Damage

Three structural differences make the current AI wave more dangerous than past displacement episodes:

Simultaneity. AI displacement is occurring across multiple sectors at once. When Block, Salesforce, Amazon, and Goldman Sachs cut in the same 12-month window, multiplier effects overlap and reinforce rather than being absorbed by unaffected parts of the economy.
Reabsorption lag. Previous layoff cycles saw workers reabsorbed within months. AI displacement may require years of retraining. During that lag, displaced workers do not spend — and the multiplier compounds each quarter.
Capital vs. wage velocity. When a firm saves $50M by replacing workers with AI, that money flows to equity holders who spend far less of it back into the economy. A dollar earned by a worker turns over 5–7 times. A dollar captured as corporate profit turns over far less.

The spending velocity of wage income is 3–4× higher than capital income.
Every dollar shifted from workers to owners destroys more demand than it creates.

The Five Layers of Demand Destruction

How Each Wave of Displacement Travels

STRUCTURAL ANALYSIS

Each AI-driven displacement event travels through at least five reinforcing layers of economic impact:

Layer	Mechanism	Illustrative Example
1 — Direct Displacement	Automated worker loses wage income entirely	Customer service agent loses $52,000/yr income
2 — Consumer Spending Stops	Displaced worker curtails rent, groceries, restaurants, services	Local businesses see 3–5% revenue decline per cluster of displaced workers
3 — Supplier Jobs Threatened	Businesses losing revenue reduce staff or hours	Restaurant cuts shifts; landlord faces vacancy; childcare loses client
4 — Tax Base Erodes	Municipalities collect less income and sales tax	Public services cut or public-sector employment reduced
5 — Corporate Revenue Shrinks	Automating firms' own customer base has less to spend	B2B spending compresses as client companies also lose workforce income

Quantifying the Cascade

Multiplier-Adjusted Displacement Estimates

DATA

Applying the standard Keynesian employment multiplier (1.5–2.0 for developed economies) to current projections produces materially larger true impact estimates:

Forecast Source	Direct Jobs Displaced by 2030	True Impact (with Multiplier)	Demand Shock Equivalent
World Economic Forum (global)	92 million	138–184 million	Comparable to 2008–2009 GFC
Goldman Sachs (global tasks)	300M tasks affected	Significant; highly variable	Depends on reabsorption speed
U.S. Tech Sector (2025 actual)	100,000+ per year	150,000–200,000 total affected	Concentrated in major metros
U.S. Customer Service	80% automation risk	120–160% of direct job count	Disproportionate for low-income workers

⬤ WBN Research Note

These Are Central Estimates, Not Worst Case

These figures represent the central estimate when standard multiplier methodology is applied to mainstream forecasts. The true range is wider in both directions depending on the pace of new job creation and the income replacement rate for displaced workers.

The Competitive Trap: Why Firms Cannot Stop

The Demand Externality

HEMENWAY FALK & TSOUKALAS (2026)

Research by Hemenway Falk and Tsoukalas (2026) demonstrates formally what makes this problem so intractable: rational, forward-looking firms cannot stop the arms race even when they can see the cliff ahead.

The mechanism is a demand externality. When a firm automates, it captures 100% of the cost savings but bears only 1/N of the demand destruction it causes — where N is the number of competing firms. The rest falls on rivals.

Dominant strategy. Automation is individually rational for every firm regardless of what competitors do — it is a strictly dominant strategy, not merely a best response.
No voluntary restraint. No self-enforcing agreement among firms can hold. Any firm that holds back loses market share while bearing the same demand destruction as if it had automated.
Competition amplifies the distortion. More fragmented industries exhibit the widest gap between competitive and socially optimal automation rates. A monopolist fully internalises the externality; competing markets cannot.
Better AI makes it worse — the Red Queen effect. Each productivity improvement raises equilibrium automation rates without changing the efficient benchmark, widening the distortion rather than resolving it.

⬤ Research Finding

A Structural Prisoner's Dilemma

In the frictionless limit, firms face a classic Prisoner's Dilemma: mutual restraint would raise all profits, but defection is individually rational regardless of what others do. Because automation is strictly dominant — not merely a best response to rivals automating — communication and voluntary coordination cannot resolve it. This is a true externality, not a coordination failure. The over-automation that results is a deadweight loss that harms both workers AND firm owners.

The Over-Automation Wedge

The gap between competitive equilibrium automation (α^NE) and the socially optimal rate (α^CO) is:

α^NE − α^CO = ℓ(1 − 1/N) / k

ℓ = demand loss per automated task · N = number of competing firms · k = integration friction cost
The wedge grows with N (more competition) and shrinks with k (higher friction). It is independent of AI productivity.

Policy Responses: What Works and What Doesn't

Instrument	Changes Automation Incentive?	Fixes Externality?	WBN Assessment
Universal Basic Income	No	No	Raises living standards floor but leaves automation race unchanged. May worsen it by attracting new market entrants.
Capital Income Tax	No	No	Operates on profit levels only — cancels out of the firm's per-task optimisation entirely.
Worker Equity / Profit-Sharing	Partially	No	Narrows but cannot close the wedge. Requires profit-sharing above 100% to fully correct. Will not arise voluntarily.
Upskilling / Retraining	Partially	No	Reduces demand loss per displaced worker but cannot eliminate the competitive externality.
Coasian Bargaining	No	No	Automation is a dominant strategy — no non-binding voluntary agreement is self-enforcing.
Pigouvian Automation Tax ✔	Yes	Yes	Charges each firm for the demand it destroys for rivals. The only instrument operating on the correct margin.

Tinbergen's Principle applies: a distinct market failure requires a distinct instrument.
No combination of income support, retraining, or bargaining will slow the arms race.
Only a correctly calibrated automation tax changes the calculus that drives it.

The Pigouvian Solution

τ* = λ(1−η)w × (1 − 1/N)

λ = worker marginal propensity to consume · η = income replacement rate
w = wage · N = number of competing firms

For large N: τ* ≈ λ(1−η)w — requires only observable sector-level data

Self-calibrating. For large competitive markets the rate simplifies to approximately λ(1−η)w — requiring only observable sector-level data.
Potentially self-limiting. Revenue can fund retraining that raises η (income replacement rate), which lowers the demand loss per task, reducing the required tax rate in future periods.
Imprecision is acceptable. Because welfare loss is quadratic in the over-automation wedge, even an approximately calibrated tax yields a first-order welfare gain.
Multilateral coordination advisable. Unilateral implementation risks pushing automation offshore — border-adjustment mechanisms analogous to carbon border taxes are a natural complement.

Strategic Implications

For Policymakers

Reframe the policy debate. The problem is not only what happens after displacement but the competitive incentives driving it. Reactive policies are necessary but insufficient without a corrective on the automation margin itself.
Act before peak displacement. Most forecasters identify 2027–2029 as the window of maximum disruption. Multilateral automation tax discussions should begin now, not after the cascade is underway.
Revenue recycling matters. Directing tax revenue into retraining rather than general revenue creates a self-reinforcing correction: higher reabsorption rates lower the required tax rate over time.

For Business Leaders

Watch for profit erosion alongside mass layoffs. Standard competitive models predict cost-reducing technology raises profits. Profit erosion coinciding with sector-wide layoffs is the empirical signature of the externality activating.
Fragmented industries are most exposed. The firms most at risk are not dominant technology companies but competitive industries deploying capable AI: financial services, logistics, healthcare administration, professional services.
Customer base income is a strategic asset. Firms that treat workforce income as purely a cost miss that it is also the consumer demand their revenue depends on. The two are not separable at scale.

For Investors

Sector-level returns will disappoint firm-level pilots. Pilot results do not capture the demand destruction that occurs when an entire industry automates simultaneously. Private returns to AI systematically overstate economy-wide returns.
Monitor η — the income replacement rate. If reabsorption keeps pace with displacement, the externality is contained. If displacement outpaces reabsorption — the more likely near-term scenario — the cascade accelerates.
The Red Queen effect. Each improvement in AI capability widens the distortion rather than resolving it. "Better AI" is not a mitigant — it is an amplifier of the structural problem identified here.

Conclusion

⬤ WBN Research Desk — Final Assessment

The multiplier effect transforms AI's job displacement from a labour market problem into a macroeconomic one.

The direct job counts in standard forecasts are the floor, not the ceiling, of the economic impact. Each wave of displacement removes spending power that was sustaining other jobs, in a cascade that compounds faster than reabsorption can offset it — particularly when displacement is simultaneous across sectors and income is shifted from high-velocity wage earners to low-velocity capital holders.

The research is unambiguous on one point: rational firms cannot stop this on their own. Only a correctly calibrated Pigouvian automation tax changes the calculus that drives it. Until that instrument is in place, every other response — however well-intentioned — is treating symptoms while the structural cause continues unaddressed.

Source & Methodology

Primary source: Hemenway Falk, B. & Tsoukalas, G. (2026). The AI Layoff Trap. arXiv:2603.20617v1 [econ.TH], March 21, 2026.

Supporting data: World Economic Forum Future of Jobs Report 2025; Goldman Sachs Global Investment Research; Eloundou et al. (2024) in Science; Brynjolfsson et al. (2025); CNBC labour market reporting (2025–2026); BEA consumer spending multiplier estimates.

WBN synthesis: Multiplier-adjusted estimates apply the standard Keynesian employment multiplier range of 1.5–2.0 to published direct displacement forecasts. These are indicative central estimates; actual outcomes depend on reabsorption speed, policy response, and sectoral composition.

"The AI Layoff Trap" by Brett Hemenway Falk and Gerry Tsoukalas (March 2026) is an economics research paper that formally proves why competitive markets will always over-automate with AI, even when every firm can see the damage it causes. Using a game-theory model, the authors show that each firm captures 100% of the cost savings from replacing workers with AI but bears only a fraction of the demand destruction it creates — because displaced workers stop spending, thereby shrinking the customer base all firms share. This trap is a Prisoner's Dilemma: every firm knows collective restraint would be better for everyone, but no individual firm can afford to hold back. The paper evaluates every major proposed policy fix — UBI, capital taxes, retraining, worker equity, Coasian bargaining — and finds that only a Pigouvian automation tax actually corrects the problem at its source. Below i the downloadable copy of the entire paper.

2603.20617v1

2603.20617v1.pdf

2 MB