When Every Company Calculates Correctly and In the End Everyone Loses
5 min read
In Q1 2026, tech firms shed around 80,000 jobs – nearly half due to automation. Two recent studies reveal why this calculation works for individual businesses yet becomes a trap for the economy as a whole. For mid-sized companies weighing AI investments, a sober look at both sides is well worth the effort.
Key Takeaways
- The math checks out. Replace a role with AI and you keep the full savings – its cost is borne by the market as lost income curbs demand.
- The collective loses. When everyone automates at once, demand evaporates faster than the economy can replace it. The study calls this the automation trap.
- The payoff never materializes. According to an MIT review, 95 % of AI pilots deliver no measurable return. Maturity of the solution decides the outcome.
Related:What truly hidden costs lurk behind interchangeable AI text / Asia sourcing: what it really costs mid-sized firms
The calculation that works for each firm in isolation
Economist Gerry Tsoukalas of Boston University and Brett Hemenway Falk of the University of Pennsylvania modeled the mechanics behind the curtain. Their headline finding sounds almost trivial: swap a worker for AI and the firm saves the full wage bill. The hidden costs, however, land elsewhere.
That displaced worker was also a customer. The lost income shrinks demand, and the ripple effect hits every seller far beyond the former employer. The single firm barely feels the gap; the market absorbs it.
Why, in the end, everyone still loses
From many rational individual choices emerges a collective dilemma. Each firm automates because it pays off for them. The result: purchasing power falls faster than new jobs can be created. The researchers frame it as a prisoner’s dilemma – everyone optimizes in isolation, yet all end up worse off.
Their uncomfortable prescription: levy a tax on automation, akin to a carbon fee. Firms that replace labor would pay for the demand they erase, and the proceeds would fund upskilling. Standard remedies like universal basic income or retraining programs, they argue, are not enough on their own. Whether such a tax ever reaches the statute books is another question entirely.
The second finding: big money, little impact
While one study examines the macro level, an MIT analysis provides a view inside companies. Around 300 corporate projects were examined. The results are sobering: 95 percent of AI pilots delivered no measurable return, with only five percent significantly accelerating revenue.
What’s behind this is interesting. In the evaluation, purchasing from specialized providers succeeded about twice as often as in-house development. The greatest returns came from silent automation in back-office operations, while visible marketing projects lagged behind. Wherever you start with processes that are already expensive and repetitive, you’ll get more out of it than with the next chatbot on the homepage.
What SMEs should take away from this
For mid-market decision-makers, there are two clear benchmarks against which to measure AI investments.
Expensive, recurring back-office processes with clear rules. Purchased, proven tools instead of in-house development. Measurable before-and-after figures before roll-out.
Visible prestige projects without a clear process behind them. Job cuts as the sole objective. Pilots that never produce a single figure yet keep running anyway.
The automation trap is an argument for closer scrutiny where efficiency ends and harm begins. SMEs decide project by project, rarely in one sweeping move.
Frequently Asked Questions
What is the automation trap?
The automation trap describes a situation in which every business automates rationally to cut costs, but the collective loses out because the resulting loss of purchasing power reduces demand faster than new jobs are created.
Who authored the study?
Gerry Tsoukalas of Boston University and Brett Hemenway Falk of the University of Pennsylvania. They model the effect as a competitive demand externality.
Why don’t basic income or retraining suffice?
In the authors’ model, these instruments fail to address the root cause: each firm’s incentive to automate more than is socially optimal. Only a levy on automation tackles the issue at its source.
What does the MIT review say about AI ROI?
Of roughly 300 projects examined, 95 percent delivered no measurable return. Success was concentrated among companies that bought proven tools and applied them in back-office processes rather than splashy marketing initiatives.
Should SMEs therefore postpone AI projects?
Postponement is rarely the answer. A better approach is selective automation: target expensive, repetitive processes with clear metrics before you start. Where the sole goal is cutting headcount, take a second look.
More from the MBF Media Network
Image source: AI-generated (May 2026), C2PA certificate embedded in image
