South African businesses that measure the success of artificial intelligence primarily through a smaller payroll risk losing the people they need to make the technology work, according to Francois van der Merwe, CEO and founder of Otinga.io.
He argues that the immediate savings from reducing headcount can obscure longer-term costs, including lost institutional knowledge, recruitment expenses and the time needed to bring new employees up to speed. Companies could find themselves hiring for capabilities they previously had the opportunity to develop internally.
The concern centres on how boards assess the business case for AI. A proposal that links automation directly to fewer employees presents a clear financial return. Assessing what those employees could contribute after training requires a more detailed understanding of the business and its future needs.
“The saving on the salary bill is easy to calculate. The value of someone who understands your customers, your systems and the exceptions that keep the business running is much harder to put on a spreadsheet. If you cut that person before exploring what they could do with AI, you may be removing your next source of growth,” says Van der Merwe.
His argument challenges the assumption that automating a function makes everyone working within it interchangeable with technology. Even where AI can complete individual tasks, people still need to assess results, handle exceptions and adapt processes as business requirements change.
Employees’ readiness to take on that work also differs. Some are already experimenting with AI and improving their own workflows. Others are willing to learn but need training, access to suitable tools and clear guidance. Some remain reluctant to change.
Van der Merwe believes a restructuring decision based only on current job descriptions can miss these differences. It may remove employees with the strongest potential to help the organisation develop its AI capabilities.
He does not suggest that every role will remain necessary or that businesses should retain their existing structures indefinitely. His concern is whether leaders have assessed the capabilities they will need before deciding which people to lose.
The cost of rebuilding knowledge
One risk emerges after the initial implementation, when a business needs to improve or extend its AI-enabled processes.
A workflow that performs well under standard conditions may still struggle with unusual customer requests, incomplete information or exceptions to company policy. Employees who know why a process operates in a particular way can help identify where automation needs adjustment and where human judgement remains essential.
If that knowledge has left the organisation, the company may need to recruit people who can work with AI while also learning its customers, systems and internal practices.
The resulting delay matters, Van der Merwe argues, because competitors may be developing similar capabilities with employees who already understand their businesses. Buying access to comparable technology does not guarantee comparable execution.
“Before approving an AI-led restructuring, boards should model the competitor that keeps and retrains its capable people. That business could be improving its service and building new offerings while you are recruiting people to recover the knowledge you have just lost,” he says.
That assessment should extend beyond the current cost base. Leaders need to consider how AI could change customer expectations, pricing and demand for their services.Where automation reduces delivery costs, businesses may be able to serve customers or undertake work that was previously uneconomical. At the same time, competitive pressure could push prices down.
Van der Merwe sees a strategic risk in reducing capacity just as those opportunities emerge. A company may achieve its savings target but lack the people needed to respond to a larger market or develop more relevant services.
Reinvesting the return
His proposed response is to direct a substantial share of AI-generated savings towards building internal capability.
As a starting principle, he suggests reinvesting between 50% and 75% of realised savings, with the appropriate allocation depending on the organisation’s circumstances. This is a strategic recommendation rather than a universal formula.
The investment should support sustained skills development and give employees opportunities to apply what they learn to actual business processes. A once-off training session is unlikely to produce lasting changes if staff return to the same responsibilities without time, support or authority to improve them.
Employees closest to a process can often identify practical opportunities for automation. Their involvement still requires oversight, particularly where workflows use sensitive information, affect customers or influence financial decisions.
Businesses therefore need clear responsibilities for reviewing outputs, approving changes and monitoring performance. They also need evidence that the investment is delivering value.
That means looking beyond the number of licences purchased or employees trained. Relevant measures could include fewer errors, shorter turnaround times, improved service quality and additional work completed with existing resources.
“The return from AI should fund the organisation you need to become. Put a meaningful share back into skills, practical workflow development and the oversight needed to make those systems reliable. A smaller payroll tells you very little about whether you have built a stronger business,” says Van der Merwe.
For boards, his recommendation is to require a fuller business case before approving major workforce changes. Alongside projected savings, it should explain which capabilities must be retained, how employees will be assessed for retraining and how the company will respond to changing demand.
The test of an AI investment, he argues, is whether it leaves the business better equipped to compete once the initial savings have been realised.