Why Competitive Automation May Weaken Consumer Demand
The core problem: private savings, shared costs
The researchers describe this as a demand externality. Each company receives the full private benefit of automation while bearing only a fraction of the wider demand loss associated with reduced employment income.
In a single-firm setting, an organisation may have a strong reason to weigh labour savings against the decline in its own customers’ purchasing power. Competition changes that calculation.
If one company delays AI adoption while competitors automate quickly, it may face higher costs and lose market share. From the perspective of each individual decision-maker, rapid automation can therefore remain the rational choice, even if widespread automation leaves every firm worse off in the longer term.
This is the central tension identified in the research. Better AI capabilities and more intense competition can increase the incentive to replace workers, while also increasing the risk of aggregate demand being eroded too quickly.
The outcome resembles a strategic trap: no individual firm can easily justify holding back when its competitors are moving forward.
Automation and augmentation are not the same strategy
The research does not argue against the use of AI. Its policy implications distinguish between AI that augments human work and AI that fully replaces it.
Augmentation can improve productivity while retaining human employment, income and organisational knowledge. It may allow employees to focus on higher-value judgement, relationship management, creative work, complex problem-solving and oversight.
Full replacement has a different economic profile. It can reduce direct labour costs, but its wider consequences depend on how quickly displaced people can move into productive new work and maintain their spending power. Where reabsorption into new roles is slow or incomplete, the model suggests that firms may collectively undermine the demand base that supports their own revenues.
For leaders, this makes workforce design as important as technology deployment. The relevant question is not only whether AI can perform a task, but whether it should replace, redesign or complement the human role attached to that task.
It may seem reasonable to expect businesses to recognise the collective risk and moderate their use of labour-replacing AI. The study challenges that assumption.
Even if executives understand the long-term danger, voluntary agreements are vulnerable to defection. A firm that believes competitors will delay automation has an immediate incentive to automate faster and gain a cost advantage. This makes informal coordination difficult to sustain.
The authors’ model also finds that several widely discussed responses—including universal basic income, worker equity participation, upskilling, capital-income taxation and bargaining among firms—do not directly correct the automation incentive under the conditions assumed in the model.
This does not mean that such measures lack social or economic value. It means that, in this specific framework, they do not eliminate the underlying gap between the private return to automation and its shared effect on consumer demand.
A policy approach focused on incentives