McKinsey’s newest global survey found something that should genuinely worry most executives. Eighty-nine percent of companies now use AI regularly in at least one part of the business. Only thirty-nine percent can point to any real earnings impact from it, and that number hasn’t moved in a year despite all the extra spend. Just six percent qualify as genuine high performers.
So the question isn’t whether AI works anymore. Clearly it does, somewhere. The question is why it isn’t showing up in your numbers.
Here’s the first thing I check, before anything else, when a client tells me they’ve rolled out AI everywhere and it’s not moving the needle. Was this technology brought in to solve a specific problem, or because it looked exciting? A few years ago there was a running joke about having an AppSumo addiction, constantly buying shiny new software and trying to squeeze it into your strategy after the fact. I see the same pattern with AI now, except worse, because most people implementing it don’t even understand what they’re actually buying. Wrong model for the task. No real grasp of what the usage actually costs. Using something built for heavy lifting to do something simple. It’s a sledgehammer to crack a nut. The question is never “have you got AI.” It’s, did you identify the actual problem first, and then find or build the tool that solves it? Or did you just buy the shiny object and hope it would find you a use?
McKinsey’s own data backs this up. The six percent who are genuinely seeing returns share one trait above all else. They redesigned the actual workflow the AI sits inside, rather than bolting AI onto a process that was already broken.
Let me give you a real example of what that actually looks like, because it’s easy to say and harder to do. We built a system for sales teams in contact centres that replaced traditional call listening and periodic training entirely. There’s a well-known pattern in sales training, a spike of enthusiasm right after a workshop, then a steady drop-off before the next one. So instead of another workshop, we built continuous professional development. The AI compares every single call, from every salesperson, in real time, against transcripts from top performers, and gives specific, actionable feedback after every call, not once a quarter.
The result I’m proudest of isn’t what it did for the top performers. It’s what it did for everyone else. Over eighteen months, one engagement, under NDA so I won’t share the client or the exact figures, conversion rates rose dramatically, and the real driver wasn’t the stars getting even better. It was the middle eighty percent of the floor improving incrementally, on every single call. That’s always where the real money is sitting, and it’s exactly why bolting a chatbot onto an existing process almost never moves the number that matters.
Now here’s the part specific to South Africa, because this isn’t a story that only applies to global enterprises with unlimited compute budgets. McKinsey also found that AI operating costs are already constraining about one in five organisations, even while most plan to keep spending more. For a South African mid-market business, that risk is sharper, not softer.
A failure to prepare is preparing to fail, and with AI that means one thing specifically. Know exactly what you’re trying to do before you touch a pricing page. Chasing the shiny object instead of the specific problem is like putting a Ferrari in your driveway to nip to the shops for milk. It’s not just unaffordable, it’s impractical, there’s nowhere to put the shopping. AI is genuinely accessible to a South African mid-market business now. The barrier isn’t cost anymore, it’s clarity. The more focused your strategy, the cheaper it is to run, because you’re only ever paying for exactly what you need.
So before you sign up for anything, ask one question. What is the single task I need this to do, and does the tool I’m about to pay for actually match that task, or am I about to pay Ferrari prices for a grocery run?
That’s the reckoning. Most companies haven’t asked it yet. The six percent already have.
