Artificial intelligence rapidly advances across global health tasks, from forecasting outbreaks and optimising supply chains to supporting diagnostics, clinical decision making and drug discovery. Its appeal has intensified amid tightening donor budgets, workforce shortages and rising demand. Efficiency promises to help do more with less, such as flagging patients at risk of treatment failure earlier or predicting treatment interruptions to target follow up and retention. Yet a historical insight known as the Jevons Paradox cautions that improvements in efficiency can increase rather than reduce overall consumption. Applied to health systems, cheaper and more capable computational tools can lower unit costs while expanding use, revealing unmet need and creating new expectations. The result is a shift in how benefits are realised: not necessarily through absolute cost reductions, but through improved value and impact per unit of investment.
Efficiency Gains Can Increase Overall Demand
The Jevons Paradox emerged from nineteenth-century observations that more efficient steam engines led to higher coal consumption by making energy cheaper and more productive. A similar dynamic is visible in AI, where advances in computational performance and reduced inference costs can drive wider adoption. In health, machine-learning algorithms have outperformed traditional risk screening tools in identifying individuals at highest risk of HIV acquisition, enabling more efficient targeting of pre-exposure prophylaxis and testing strategies. Falling costs for cloud-based tools make once cost-prohibitive interventions scalable, facilitating better optimisation across the prevention, diagnosis, treatment and management continuum.
Must Read: Closing the Governance Gap in Healthcare AI
This broadening reach does not guarantee lower total spend. As efficiencies are realised, demand can expand in step, particularly when tools expose the full extent of unmet need. Efficiency can also raise expectations among funders and implementers for more precise and continuous service. Over time, increasingly sophisticated decision support and predictive analytics can redefine what is considered acceptable care, reinforcing the cycle in which greater capability encourages broader use. Generative AI adds another layer by enabling novel capabilities that stimulate new categories of demand beyond incremental gains in existing workflows.
Cost Pressures Shift from Detection to Intervention
There are areas where AI has reduced costs, including automated claims processing, fraud detection and virtual call centres that have generated savings for national health insurance schemes. AI can also lower specific service delivery costs. Even so, efficiencies at the component level do not necessarily translate into lower system-wide expenditure. When models predict who is likely to drop out of HIV care, earlier detection of attrition risk can lead to additional follow up visits, the use of clinical co-pilots, provision of psychosocial support through empathetic chatbots and verification of benefits such as transportation subsidies using digital identifications. The burden shifts from identifying risk to delivering interventions that address it, introducing workload and costs downstream.
Short-term dynamics further constrain the prospect of headline savings. AI-enabled efficiency gains are likely to be absorbed by latent demand rather than converted into cashable reductions. Rather than requiring fewer community health workers, for example, improved efficiency may simply allow coverage of all households within the available time. As coverage expands, so too can demand for services, which increases resource requirements. At system scale, the effect resembles a redistribution of cost pressures rather than their elimination. Tasks become faster or more targeted, yet total activity can grow, sustaining or raising overall spend as programmes meet needs that were previously unaddressed or only partially addressed.
From Cost Savings to Value Creation in Health Systems
These dynamics have implications for how AI is positioned and financed. If the principal expectation is absolute cost reduction, disappointment is likely, particularly in resource-constrained contexts where unmet needs are substantial. A more resilient framing is value creation: using AI to increase health impact per unit of spend. In HIV programmes, this can include enabling earlier diagnosis, increasing the number of individuals initiated and retained on antiretroviral therapy, improving viral suppression rates and preventing new infections through more efficient targeting of prevention services. The emphasis shifts from reducing spend to improving outcomes with the resources available.
Under this framing, AI functions as a force multiplier rather than a substitute for people or infrastructure. It can support better targeting of scarce donor resources, help reduce waste and augment decision making, with particular relevance in under-resourced settings. However, efficiency tools do not negate the need for skilled providers, robust data systems, ethical governance and sustained funding commitments. Aligning expectations with these realities helps maintain focus on the most meaningful returns: reach, quality and equity. By recalibrating success metrics around improvements in effectiveness rather than headline savings, stakeholders can integrate AI in a way that enhances programme performance without assuming it will resolve structural underfunding or workforce gaps.
Efficiency gains from AI can expand access, sharpen targeting and lift programme performance, but they can also stimulate demand and reveal unmet need, limiting the scope for absolute cost reductions. A pragmatic approach recognises that the greatest benefits lie in value creation: earlier detection, better retention, improved suppression and more effective prevention achieved with available resources. Positioning AI as a force multiplier within health systems, rather than a shortcut to lower budgets, supports clearer expectations and more sustainable investment. Success rests on reframing objectives, reinforcing core capabilities and directing AI toward measurable gains in reach, quality and equity.
Source: The Lancet Digital Health
Image Credit: iStock
