In a contact centre, quality-assurance staff used to listen to about five per cent of calls because checking every conversation by hand was too expensive. Today, speech-AI can transcribe, index and audit each call for a fraction of that cost, so managers are moving from spot checks to analysing every interaction, from spotting billing errors to measuring customer sentiment.
That shift illustrates the point made by TechCentral: artificial intelligence is not shrinking the amount of intelligence enterprises need, it is making each unit of intelligence cheaper, and that cheapness unlocks new uses.
What is the Jevons paradox?
The term comes from the 1865 work of English economist William Stanley Jevons. He observed that when James Watt’s improved steam engine cut the coal needed for a given amount of work to about a quarter of earlier designs, Britain did not burn less coal. Instead, cheaper steam power opened thousands of new applications, textile mills, deep-shaft mines and railways, and total coal consumption rose sharply. Economists call this the Jevons paradox: a drop in the unit cost of a resource can increase overall demand for that resource.
Applied to AI, the paradox means that as the cost of processing information falls, organisations do not simply replace human workers with machines and stop. They apply intelligence to problems that were previously too costly to tackle.
Take radiology. In 2016, AI pioneer Geoffrey Hinton warned that deep-learning models would soon outperform radiologists. The prediction proved premature. The US Food and Drug Administration (FDA) now lists more than a thousand AI-enabled medical devices, three-quarters of them for imaging, yet the Mayo Clinic’s radiology staff has grown by 55 % since 2016 to around 400 doctors. The rise is driven by an ageing population and higher imaging volumes, but AI’s ability to read scans faster and cheaper also makes it feasible to handle more cases.
In South Africa, the effect is most visible where global AI models stumble. Speech models trained in the Global North assume monolingual, high-resource environments and falter when conversations switch between English, isiZulu, isiXhosa and Afrikaans amid background noise. Local firm Saigen builds models that work with as little as 100 hours of domain-specific recordings, turning a previously unaffordable capability into a routine tool for call-centres, banks and insurers.
For executives, the practical question is not whether AI will replace jobs, but where the cost barrier has been suppressing demand for operational intelligence. When that barrier falls, new workflows emerge, from real-time compliance monitoring to predictive maintenance, and businesses that recognise the shift can capture value that was previously out of reach.
However, the paradox has limits. In areas where demand is fixed, for example, entry-level roles that perform a set amount of routine work, cheaper AI can indeed reduce headcount. The challenge for South African companies is to identify elastic demand, where AI can expand the scope of work, and to reskill staff for higher-value tasks that the technology uncovers.
Understanding the Jevons paradox helps firms avoid the false binary of AI as either a job-killer or a miracle efficiency tool. Instead, it frames AI as a cost-reduction engine that can unlock new business opportunities, provided leaders map where intelligence was previously too expensive to apply.
Read more about AI trends in the Tech & Telco section.
TechCentral reported that the debate about artificial intelligence has settled into a tedious equilibrium, with executives hearing either apocalyptic job-displacement warnings or promises of operational efficiency. The source adds that both camps share an unexamined premise that enterprise demand for intelligence is fixed, a view history disproves. It points out that when the unit cost of a vital resource collapses, demand rarely stays flat and often rises dramatically. The article also notes that AI is collapsing the unit cost of processing information, turning previously unaffordable intelligence into a routine expense for organisations across sectors.
This kind of cost-driven shift matters to South African business owners because it changes the economics of data-intensive processes. When transcription, indexing or image analysis become cheap enough to run on every transaction, managers can move from sampling to full-scale monitoring, uncovering issues that were hidden by cost constraints. Companies should watch for the point at which the expense of a new AI tool falls below the threshold of marginal benefit, because that is when adoption accelerates and new use cases emerge. Keeping an eye on internal cost-benefit analyses and the speed at which AI licences are priced will signal when the technology moves from a pilot to a core capability.
TechCentral reported that global speech models are trained on hundreds of thousands of hours of audio, while efficient local models can operate with as little as 100 hours of domain-specific recordings. It highlights that Saigen builds models to handle code-switching, turning a previously unaffordable capability into a routine tool for call-centres, banks and insurers. The source also mentions that South African executives have spent the past few years asking which workforce capabilities AI will replace, underscoring the shift from questioning replacement to identifying suppressed demand.
Understanding how a lower cost input unlocks new applications helps executives decide where to invest. The key is to differentiate elastic demand, where cheaper AI expands the scope of work, from fixed demand, where it may simply reduce headcount. Business owners should map processes that were previously limited by review costs, such as compliance checks or predictive maintenance, and evaluate whether AI can make those processes continuous. Monitoring adoption rates and reskilling plans will indicate whether the organisation is capitalising on the expanded intelligence capacity or merely trimming staff.
