When Every African Bank Has the Same AI, What Exactly Is the Competitive Advantage?
When Every African Bank Has the Same AI, What Exactly Is the Competitive Advantage?
The more African banks spend on the same AI, the less AI itself will be a competitive advantage.
I want to state that plainly at the start, because everything else here follows from it. The real advantage will not sit in the technology. It will sit in what the institution puts around the technology: proprietary context, judgement, permission to act, and the ability to learn faster from outcomes.
Hold the chain in mind as you read. Intelligence, context, judgement, permission, learning. The farther right a bank moves along it, the harder its advantage becomes for a competitor simply to buy.
AI adoption is not the same as AI advantage
The numbers tell one story. African banks are increasing AI investment. Staff adoption is rising. Use cases are multiplying. Generative AI is moving into customer service, fraud detection, credit, coding, marketing, compliance and operations.
But most of those capabilities arrive through the same vendors, the same cloud platforms, the same foundation models.
So the question quietly changes. It is no longer who has AI. It is this: if our largest competitor can buy essentially the same intelligence tomorrow, what exactly are we building today that they cannot easily reproduce?
The signals are worth reading honestly. In recent African banking research, 83.2% of professionals expect to increase AI investment. Only 67.1% formally measure what that investment returns. And 50.2% name legacy integration as their biggest internal obstacle.
Read those three together. We are spending confidently, measuring loosely, and building on foundations that resist the very technology we are buying.
The strategic mistake would be to confuse AI activity with AI advantage.
That is the thesis. Let me prove it.
Intelligence is becoming easier to buy
A few years ago, sophisticated AI capability could distinguish a bank. That window is closing.
Today a bank can buy AI credit analysis. Fraud detection. Coding copilots. Customer-service agents. Compliance tooling. Investment research. Marketing intelligence. The menu is the same menu your competitor is reading.
Adoption figures make the point rather than settle it. When Standard Bank reports that 72% of its employees are using generative AI, across 87 approved use cases, that is impressive this year. But if every major bank on the continent reaches a similar level within a few years, what does the number actually tell us about advantage?
Adoption is becoming a measure of participation in the AI economy. It is not, by itself, a measure of differentiation.
If the intelligence itself is becoming widely available, the first serious source of advantage has to lie somewhere else.
Context is where the institution starts to matter
It lies, first, in context. What does your bank know that the generic model does not?
Transaction history. Merchant behaviour. Repayment and collections patterns. Salary cycles. Remittance flows. Local sector knowledge. Fraud signatures particular to your markets. The texture of how an SME in your city actually trades.
This is where the African position can be unusually strong. Our banks hold knowledge about markets where a great deal of economic life does not sit neatly inside formal financial records. That knowledge is hard for an outside model to reproduce, because it was never fully written down anywhere a model could read it.
But I want to challenge the lazy version of this argument. Owning data is not the same as having usable institutional context.
Consider what it takes just to see your own customer clearly. Absa consolidated customer information that had been spread across 32 systems. The information already existed. The strategic work was making it usable.
That distinction matters more than most dashboards admit. A bank may hold twenty years of customer history and still have less usable context than a competitor with ten.
In one bank, the constraint is the data itself. In another, the data is fine and the problem starts much later. Knowing which is which is most of the work.
Better intelligence does not guarantee better judgement
Suppose two banks receive the same AI assessment of the same SME. Strong recurring cash flows. Low customer churn. Stable transaction history. High probability of repayment.
Bank A reads it as an attractive working-capital opportunity.
Bank B reads it as insufficient collateral.
Same intelligence. Opposite decision. The difference is not the model. It is risk appetite, credit philosophy, sector knowledge, pricing discipline, strategic priority, institutional experience.
AI can improve what a bank knows without improving what a bank decides.
This is the point at which advantage stops being technological. The same analysis passes through very different institutional lenses, and the lens is the thing that is hard to copy. It is also, not incidentally, the thing boards exist to shape.
The biggest gap may be between knowing and acting
Now the harder question. When AI knows something useful, can the institution act on it?
The model says this customer is likely to leave. This transaction is probably fraudulent. This SME qualifies for credit. This borrower should be priced differently. This customer needs an intervention now.
And then what? Does the insight move? Or does it wait for manual verification, several approvals, an old credit policy, a committee sign-off, a legacy workflow, next week’s review?
Without institutional permission to influence real decisions, AI becomes very expensive advice.
I am not arguing that institutions should let AI act unchecked. I am arguing the opposite of drift. Governed decision rights are themselves a competitive capability. The difference between two banks may have less to do with what their AI knows and more to do with what their institutions are actually willing and able to do about it.
Learning is where the advantage compounds
The last layer is the one that compounds. Learning.
In Bank A, the AI makes a recommendation, someone makes a decision, and the outcome disappears into another system. In Bank B, the recommendation leads to a decision, the decision produces an outcome, and that outcome is fed back so the next decision is better.
Credit approval, then repayment result. Fraud alert, then confirmed or not. Retention offer, then accepted or rejected. Pricing move, then customer response.
The real moat may not be the model. It may be the speed at which the institution converts its own decisions into proprietary learning.
That learning history is the one thing on this list a competitor cannot buy from a vendor tomorrow morning. It exists only because the institution lived through the decisions that produced it.
Which completes the chain. Intelligence, context, judgement, permission, learning. The farther right the institution moves, the harder the advantage becomes to copy.
The board may be measuring the wrong race
I sit in these rooms, so let me be direct about what boards are about to be shown.
Bank boards will increasingly receive impressive AI dashboards. Models deployed. Staff trained. Use cases launched. Productivity gained. Savings booked.
None of those measures, on its own, tells the board whether the institution is building a competitive advantage.
The board’s job here is not to choose the model, design the architecture or run the implementation. Its job is to understand what strategic advantage management believes the investment will create, what assumptions that advantage rests on, and how the institution will know whether it is actually materialising.
Four questions get you most of the way there.
Which parts of our AI capability could our largest competitor buy tomorrow?
What proprietary context materially improves our decisions, and are we organising it or merely storing it?
Which important decisions are genuinely changing because of AI, rather than simply becoming faster?
What are we accumulating today that becomes more valuable with every decision we make?
And behind all four, the capital-allocation question a board cannot avoid. If AI spend is rising, what exactly is this institution accumulating that will still matter after everyone else has bought similar technology?
The technology may become common; the institution does not have to
African banks are entering an AI investment race. The winner may not be the bank with the largest AI budget, the most use cases, the highest adoption figures, or even the smartest model. Those advantages are becoming replicable, and quickly.
What is harder to reproduce is an institution that has organised its proprietary context, developed sharper judgement, created the permission to act on new intelligence, and built a learning loop in which every outcome improves the next decision.
The irony of this era is worth sitting with. As artificial intelligence becomes more powerful, the quality of the institution around it matters more, not less.
The technology may become common. The institution does not have to.
So here is the question I would put to any bank leader reading this. If your biggest competitor had exactly the same AI tomorrow morning, what advantage would your bank still possess?