South African businesses are adopting artificial intelligence at pace, but faster individual work is not necessarily translating into faster businesses.
Dane Walsh | image supplied
PwC's May 2026 AI performance research found that 82% of African organisations are running AI pilots, while KPMG's latest Global AI Pulse, which included South Africa, found that only 7% of leaders had established a return on their AI investment.
For Dane Walsh, head of research and development at sales operations platform Skynamo, part of the problem is that businesses can improve individual tasks without addressing the processes surrounding them.
A developer may use AI to write code faster and a designer may move through concepts more quickly, Walsh says, but the work still needs to move between people and systems.
AI can increase the amount of information moving through those handovers. If the next person still has to interpret the output, find missing context or work through a fragmented process, the bottleneck remains.
Research by OfferZen among South African technology leaders in 2025 points to a similar pattern. Ninety-seven percent of respondents said their teams were using AI, with most reporting faster code writing, while leaders also reported that some blockers were shifting further downstream.
Measure what reaches the customer
Walsh argues that businesses should focus less on AI usage and more on whether it is improving the work that ultimately reaches customers.
AI adoption, he says, is not itself a business outcome. Neither is the number of pilots being run.
Instead, organisations should track how frequently useful improvements reach customers, how long delivery takes and how quickly teams can recover when something goes wrong.
Research by METR in 2025 illustrates why those measurements matter. In a study involving experienced open-source developers, participants expected AI tools to make them faster and subsequently believed they had done so. However, the study found that tasks actually took 19% longer when AI was allowed.
METR has since said newer AI tools may be producing better results, so the study should not be treated as a verdict on current AI systems. Its broader lesson is that perceived productivity and measured productivity are not necessarily the same.
Human interaction still matters
Walsh also cautions against allowing AI to replace too much interaction between people.
AI can be easy to work with because it readily responds to a user's assumptions, whereas colleagues can challenge ideas and force people to explain their thinking.
For software development in particular, Walsh argues that teams still need opportunities to share context and develop ideas together rather than relying entirely on individual interactions with AI tools.
He says AI adoption should therefore not become a competition to maximise individual usage. Different people may use AI extensively or selectively, with the more important question being what the team achieves collectively.
Start with the business problem
Walsh's recommendation is to begin with the problem a business is trying to solve rather than with the technology itself.
AI may form part of the solution, but organisations should first establish the customer problem and desired outcome.
That approach also matters as AI tools continue to evolve. Businesses need systems that can be modernised as better technologies become available, rather than becoming locked into decisions based on what worked at a particular point in time.
For businesses evaluating AI investments, Walsh suggests asking three questions: are handovers becoming shorter, have the bottlenecks holding up work actually been removed, and are customers seeing better results?
If those outcomes are not improving, greater AI activity does not necessarily translate into greater business value.
