A2B Strategy

Originally published on LinkedIn, June 15, 2026

Digital v Human

On Friday I had two unbelievably good “banking” experiences at opposite ends of the human v digital divide. First , I accompanied my gardener to FNB South Africa at Cresta to understand why half a salary top-up had been “stolen by the bank”. The lady that helped him was kind, empathetic, reassuring, walked him slowly through what had happened, took steps to ensure that it would not happen again, set in motion steps to reverse penalty fees and shut down any talk of additional credit which the system was recommending. Her name was Nomonde. Never ever have I witnessed such brilliant branch service. She embodied “how can we help you”, and she should be used as Exhibit A for in-house training. A more general point though is that human connection matters when things go wrong, and she did not just save an account, she gained an advocate. Digital cannot match that…

And so saying, walking out of the Branch I came across a GoTyme Bank kiosk. It was the first time I had seen one so I stopped to take a few photeys, then thought what the heck, let’s trial their account opening. The barbarians are at the gate! I had entered my details, been KYC’d, issued a personalised debit card, installed the app, transferred money in, and activated the account – all in 5 minutes. A new benchmark has been set!

To be clear, my very strong view is that the notion of “digital-led” banking is bollocks. In a world in which code is free, digital offers no moat. The moment-of-truth for banking experiences and customer retention is when the pawpaw hits the fan. Call centres, Chatbots, AI personae exacerbate distrust and frustration. This is not to say that digital is not important. On the contrary digital is critical because it represents “table stakes”. When legacy banks talk of being digital-led what they mean is they are scrambling desperately to catch up and maybe aren’t that far behind. Gotyme has raised the table stakes! Very impressive!

AI and Corporate Sovereignty

Saturday morning witnessed household panic. Fable…wtf… where are you? Ah, America has gone skiing off-piste again!

Dr. Mark Nasila is one of South Africa’s leading AI gurus and someone I have tremendous respect for. But when he starts talking about the importance of sovereign AI (I mean he wrote a whole flipping book about it) I confess to switching off. Go ahead, build a SA data centre… you going to power it with what, Eskom’s finest?! But this little Fable episode should have caught every C-suite exec’s attention. If you rely on closed source American frontier models to power your agents, workflows, processes et al, how do you feel about them being turned off on a whim. In the new world AI economy, notice has been served that any US president can turn off any country’s and company’s supply of “electricity” at a moment’s notice. A second point, when you leverage the respective hyperscalers’ secure “private sandboxes” do you really believe that your data is not visible to the service providers. The best predictor of future behaviour is past behaviour, and we have all seen the “bending of the knee” of the American CEO class. The SA CEOs of these American companies will all deny compromising data privacy of course, and probably in good faith. But they themselves “only need to know what they need to know”, especially when it comes to the all-encompassing national security interests of the US. Finally, there is the straight-up economic risk of being bound to a toll road. A year or two ago I paid R1000 for my Microsoft subscription. At renewal time it went up 70% because I now had access to that vapid Copilot. But what could I do? My ecosystem is tied to that toll road. A better general example: when I listened to Standard Bank’s results they talked of significant savings in moving off legacy hardware and onto the cloud. Elsewhere though I had observed that annual unit subscription costs were up 9%. That is 3x inflation. The promised land in the Cloud is not all milk and honey. Cloud dependence is probably inescapable. AI, not so much. Corporates probably need a few smart peeps who design on frontier models, replicate on retuned pretrained open-source models, and shrink depending on use case to run locally.

We live in interesting times!