The argument for artificial intelligence in business is an economic one: the traditional inputs of capital and labour have been producing diminishing growth, and automation that both replaces routine work and augments skilled work behaves like a new input rather than an efficiency gain. The argument against careless adoption is also economic, because the cost of getting it wrong lands on the business that deployed it.
Work on responsible adoption in South African conditions has consistently identified the same tension: the productivity case is real, and the social and ethical consequences are sharper in a market with high unemployment and significant inequality.
Automation and augmentation are different decisions
Replacing a routine task removes a cost and often a job. Augmenting skilled work makes existing people more productive without displacing them. These are separate choices with separate consequences, and businesses that describe both as adopting technology tend not to have thought about which one they are doing.
The employment consequence is a live commercial risk here
In a market with high unemployment, a business that automates visibly and without a plan for the people affected carries reputational and labour relations exposure that the technology case does not price in. Handling it well, through redeployment and retraining, is cheaper than handling it badly.
An automated decision still has to be explainable
Where a system decides who gets credit, who gets hired or what someone is charged, the business remains accountable for the outcome and for the personal information used to reach it. Those obligations are enforced by the Information Regulator, and pointing at a supplier’s model is not a defence.
Bias enters through the data, not the intention
A system trained on historical decisions reproduces the patterns in those decisions. In a country with a particular economic history, that is a specific and foreseeable risk rather than an abstract one. Any business deploying an automated decision should test outcomes across groups before trusting it, and keep testing afterwards.
Start where the task is repetitive and the error is cheap
For a small business the sensible entry points are drafting, categorisation, scheduling and first-line customer queries, where mistakes are visible and recoverable. Applying it to pricing, credit or hiring decisions requires the testing and accountability described above. Local research into where technology adoption actually raises productivity is published by the Council for Scientific and Industrial Research.
Frequently asked questions
Why is artificial intelligence described as a new factor of production?
Because capital and labour have been delivering diminishing growth, while automation that both replaces routine work and augments skilled work adds capacity rather than only efficiency.
What is the difference between automating and augmenting?
Automating removes a task and often a job. Augmenting makes existing people more productive. They carry different consequences and should be decided separately.
Who is accountable for an automated decision?
The business that deployed it, including for the personal information used, regardless of who supplied the system.
How does bias get into a system?
Through the historical data it learns from, which reproduces the patterns in past decisions, making outcome testing across groups necessary before and after deployment.
Where should a small business start?
With repetitive tasks where errors are visible and recoverable, such as drafting, categorisation, scheduling and first-line queries, rather than pricing, credit or hiring.
Further reading
Originally published in July 2017. Updated September 2026 to set out the practical decisions a business faces when adopting automation, using the lead story from the original roundup.
