AI will not fix what transformation has failed to build

South Africa's financial services sector faces a strategic contradiction it can no longer afford to ignore. Most firms are running a transformation agenda and an AI innovation agenda simultaneously, yet the two rarely converge. The consequences of that separation are becoming measurable, and they cut both ways.

A recent EY survey of 12 South African financial institutions found that most have moved beyond AI pilots but are still struggling to scale AI into a genuine enterprise capability. Investment remains concentrated on efficiency and productivity, with limited impact on how firms compete or differentiate themselves. The sector is investing heavily in technology, but is not yet transforming. 

AI does not operate in a neutral environment 

The most frequently mentioned risk in AI discussions - that it will displace workers - overlooks a more immediate threat. AI will expose organisations' true nature: the quality of their data, their governance maturity, and whether transformation is truly integrated into their operations or just a tick-box exercise. 

AI systems do not operate in neutral settings. They are introduced into organisations that already have established histories, hierarchies, assumptions, and structural blind spots. Without intentional efforts, these technologies won't enhance the inclusiveness or intelligence of financial services. Instead, they risk amplifying existing inequalities and making them more difficult to identify. 

A faster system isn't always a fairer one. Similarly, a smarter model doesn't necessarily mean it's wiser. An institution that is technologically advanced isn't necessarily transformed. 

Homogeneous teams produce compromised models 

Research consistently shows that diverse teams lead to better AI outcomes. Teams with varied experiences tend to ask more insightful questions, explore a broader array of innovative initiatives, and uncover blind spots that homogeneous teams might overlook. This is especially critical in AI development because the models produced reflect the perspectives of the teams that create them. The choices regarding data, feature selection, and the significant cases tested are all influenced by the diverse viewpoints in the room. 

When rooms lack diversity, models silently inherit these gaps, remaining invisible until they are not. 

The EY survey also highlighted data quality as the main obstacle to expanding AI in South African banks. This issue is more than just technical; data that only mirrors those with past access to financial services is not a true data quality problem. Instead, it signals a failure in transformation embedded within AI infrastructure. 

Treating transformation as a compliance requirement while viewing AI as a performance goal not only hampers transformation but also yields inferior AI. 

A talent deployment problem, not a talent shortage 

South Africa is often seen as lacking AI-ready talent, but this diagnosis is inaccurate. The country actually has technically skilled, quantitatively trained, and contextually aware individuals, many of whom understand the South African market deeply. The real issue is their absence from the decision-making spaces for AI strategy, model assumption testing, and questioning whether systems truly serve their intended populations. 

This isn't a pipeline issue. It reflects a structural decision about who is trusted with the important decisions. The consequences go beyond fairness: a financial model that works solely for those already included isn't innovative in the South African context. It remains incomplete. 

A product assuming stable income, formal employment, and English proficiency isn't a universal solution; it's designed for a specific segment. Critical questions include who is excluded from this data, who might be disadvantaged by this model, and which barriers are being treated as risk factors. These questions are central to determining whether AI can provide real commercial value in this market. 

In the era of AI, the quality of questions posed can be as important as the quality of the code produced. 

Governance in an AI world is a cross-disciplinary function 

At Prescient Investment Management, we treat AI governance as a shared responsibility across multiple disciplines, not merely a technical task. Ownership is assigned to teams across investment, operations, compliance, and HR, ensuring diverse viewpoints are integrated into AI decisions from the outset, rather than after the strategy is finalised. The standing question is simple: who is involved in these discussions, and is there anyone who should be involved but isn’t? 

That structure reflects a broader principle. Diversity of thought has to be built into governance by design. It cannot rely on individuals having the courage to raise dissenting views in the wrong room. 

South Africa’s structural advantage in AI development 

South Africa's unique situation enables it to offer a distinct contribution to global AI progress. Instead of merely consuming solutions developed elsewhere, it has the potential to pioneer governance models and institutional strategies tailored for complex environments. 

Operating within a market shaped by the divide between formal systems and informal economic practices, the fragility of institutional trust, and the rich diversity of its people, South African financial institutions have gained contextual insights that directly inform the toughest challenges in AI deployment. Financial access extends beyond mere product availability; it encompasses language, culture, affordability, and dignity. Designing with this reality in mind leads to more resilient systems, rather than more limited ones. 

Trust, not data, platforms or sophisticated algorithms, will become the most important currency of the next era of financial services. Clients might not always appreciate the sophistication of the underlying model that drives an investment suggestion or an onboarding process. However, they will recognise whether they are treated with respect, whether there is real recourse when issues arise, and whether technology has made them feel more included or more invisible. 

The question the sector must answer 

The sector's key challenge is not just to adopt AI, as market forces will drive that outcome on their own. The more challenging mandate is to implement AI in a way that shows the sector has learned from its past. This involves creating models that avoid repeating historical exclusions, assembling teams that represent the country's diversity, establishing governance capable of slowing deployment when needed, and developing products that can distinguish between risk and exclusion. 

Transformation is not just a cost of doing business in South Africa; it is the essential intelligence needed to succeed. It involves market, social, and contextual understanding that ensures products and services work effectively within the actual economy, not just the assumed one. 

AI can recognise patterns, but transformation assesses whether those patterns reflect the people served. While AI can improve systems, transformation considers if those systems are valuable enough to optimise. AI can speed up processes, but transformation questions who benefits from that acceleration and who might be left behind. 

The future leaders of South African financial services won't be those with the most advanced algorithms. Instead, they'll be institutions with a clear purpose, strong ethics, and a single guiding question: who benefits most from this progress?