AI has moved into businesses faster than many organisations have worked out how to use it well. PwC’s May 2026 AI performance research found that 82% of African organisations are running AI pilots, yet few have scaled them across the enterprise. KPMG’s latest Global AI Pulse, which included South Africa, found that only 7% of leaders had established a return on their AI investment.

Dane Walsh, Head of Research and Development at unified sales operations platform Skynamo, sees part of the problem in what happens around the technology. People can become faster at individual tasks without the organisation becoming faster as a whole.

“It is visible in software development. A developer can use AI to write code faster, while a designer can move through concepts more quickly. The gain at that point in the process can be real, but the work still has to pass between people and systems,” says Walsh.

AI puts more information into those handovers. Someone still has to understand what has been produced and carry the right context into the next part of the job. If that process remains slow or fragmented, making one person faster does not remove the delay further down the line.

OfferZen’s 2025 research among South African technology leaders suggests this is already happening. Ninety-seven percent of those surveyed said their teams were using AI. Most reported faster code writing, while leaders also saw blockers shifting downstream. The work had accelerated in one part of the process without necessarily speeding up the rest.

Measure what reaches the customer

“In R&D, leaders should be far less interested in how many prompts somebody writes as opposed to whether the team solves the customer’s problem. AI usage is not a useful business outcome on its own. Neither is the number of pilots running. If a team produces more work but releases less reliable software, a business has gained activity rather than value,” says Walsh.

“The useful measure is delivery. We should know how frequently useful improvements reach customers, how long that takes and, when something goes wrong, whether the team can recover quickly. AI can improve those outcomes. It can also create extra review and rework when it is inserted without enough thought.”

A 2025 METR study warns against confusing the feeling of productivity with the result. Experienced open-source developers expected AI tools to make them faster and afterwards believed this to be the case. Tasks actually took 19% longer when AI was allowed. METR has since said newer tools may be producing better results, so this is not a verdict on current AI. It is a reminder to measure what happened rather than how the tool felt.

Keep human friction in the system

“AI is easy to talk to and can be remarkably agreeable,” says Walsh. “A colleague may disagree with your premise or make you explain an idea properly. That can be uncomfortable, but that is where good thinking happens.

“It is tempting for many developers to interact with an AI model rather than another person. Over time, that can leave people working inside increasingly comfortable individual confirmation-bias loops. Software development is creative work. We still need people around whiteboards, sharing context and arriving at ideas they would not have reached on their own.”

Walsh suggests that leaders should avoid turning AI adoption into a race to maximise individual usage. Teams need room to experiment and share what works, he says. One person may use AI constantly while another may use it selectively. “A leader would do well to judge what they achieve together, as a team,” he says.

Start with the problem

Walsh is wary of technology discussions that begin with AI itself because they can limit the conversation before the business has had the opportunity to define what needs to change.

“If you start with the technology, you are already looking for somewhere to use it. Instead, you should start with the customer problem and the outcome you need. AI may form part of the solution, but the tool should follow the problem,” he says.

That thinking also shapes how Walsh looks at longer-term technology decisions. AI is changing too quickly for businesses to assume that whichever tool works best today will remain the right choice several years from now. Systems need enough flexibility to adapt as the technology changes.

“Technology ages. You need to be able to modernise as better options become available rather than finding yourself stuck with a decision that might have made sense years ago,” he says.

At Skynamo, Walsh applies the same principle to technology used by sales teams in the field. Its role is to carry information and reduce administrative work without getting between the salesperson and the customer.

“Useful technology often disappears into the background. People have better information and less admin to deal with, which leaves them more attention for the relationship in front of them. I think AI should work in much the same way.”

He said businesses would be wise to watch three things: Are handovers shrinking? Are the bottlenecks that hold up work actually gone? And, is the customer seeing better results?

“If the answer is yes, the return follows. If it is no, you have bought activity, not value,” he says.