Dr Evans Sagomba
Everything AI
THE news that OpenAI’s latest model, GPT 5.5, failed half of a junior financial analyst exam should give us pause.
For months, the global conversation has been filled with bold claims about artificial intelligence revolutionising industries, replacing jobs, and reshaping economies. Yet here we are, confronted with evidence that even the most advanced systems struggle to perform tasks that many entry-level analysts are expected to handle competently.
The benchmark in question, Vals AI Finance Agent v2, is not a trivial quiz. It is designed to replicate the real work of financial analysts: extracting figures, running calculations, and connecting conclusions across multiple steps. GPT 5.5 managed just over 52 percent accuracy.
That is marginally better than its rivals, but still far below the threshold of reliability.
In a real firm, such performance will raise red flags. No board will entrust its shareholder value to an analyst who gets half the answers wrong.
This is where the hype around artificial intelligence collides with reality.
For all the talk of machines replacing humans, the truth is that AI remains brittle when confronted with complex, multi layered tasks.
It can mimic language, generate reports, and even draft financial summaries. But when asked to execute the full chain of reasoning that underpins professional work, it falters.
The Zimbabwean corporate sector should take note: it is not yet Uhuru with AI.
Consider the role of a junior analyst in Harare’s financial institutions. They are expected to interpret balance sheets, assess credit risks, and prepare reports that feed into board decisions.
Accuracy is not optional.
A miscalculation can distort valuations, mislead investors, and expose companies to regulatory penalties. If an AI system delivers results with barely half the accuracy of a human analyst, it is not a tool to replace staff. At best, it is an assistant that requires constant supervision.
The danger lies in over reliance.
Many companies, dazzled by marketing claims, are tempted to cut costs by substituting analysts with AI agents. Yet this experiment risks undermining governance.
Shareholders expect boards to exercise prudence.
Regulators demand compliance. Clients want trust.
None of these can be guaranteed if decisions are based on flawed outputs. Zimbabwe’s financial sector, already navigating economic volatility, cannot afford the luxury of misplaced faith in machines.
It is important to acknowledge progress.
Two years ago, older systems scored around 30–40 percent on similar tasks. Today’s models have improved to the high 40s and low 50s. That is a step forward.
But progress is not the same as readiness. A bridge that is half built cannot carry traffic.
Likewise, an AI system that is half accurate cannot carry the weight of corporate finance.
The lesson is clear: AI should be treated as a complement, not a substitute.
In Zimbabwe, where companies are exploring digital transformation, the prudent path is to integrate AI as a support tool.
Let it automate repetitive tasks, flag anomalies, and generate drafts. But let human analysts retain the final say. The human mind, with its ability to contextualise, question, and exercise judgement, remains indispensable.
There is also a cultural dimension.
In our context, trust is not built on algorithms. It is built on relationships, accountability, and transparency.
A board member in Mutare does not simply want numbers; they want assurance that those numbers reflect reality.
AI cannot yet provide that assurance. It can process data, but it cannot carry responsibility. Responsibility rests with people. This is why governance frameworks matter.
Boards must establish clear policies on how AI is used. They must demand transparency from vendors, scrutinise outputs, and ensure that human oversight is never removed.
The failure of GPT 5.5 is not just a technical issue; it is a governance warning. It tells us that reliance without oversight is reckless.
Zimbabwe’s National AI Strategy speaks of harnessing technology for inclusive growth. That vision is noble. But inclusivity requires caution.
If companies rush to replace analysts with machines that fail half their tasks, the result will not be growth. It will be erosion of trust, misallocation of capital, and potential regulatory breaches.
The global narrative often paints AI as inevitable. Yet inevitability is not the same as readiness.
Electricity was inevitable, but it took decades to build grids that were safe and reliable.
The motor car was inevitable, but it required traffic laws, road networks, and driver training. AI is no different. It must be integrated carefully, with safeguards and human oversight.
For Zimbabwean companies, the message is simple: do not be seduced by headlines. Look at the evidence. If the most advanced models cannot pass a junior analyst exam, they are not ready to replace your staff.
Use them wisely, but do not entrust them with responsibilities they cannot fulfil.
This is not to dismiss AI entirely. It has the potential to enhance efficiency, reduce costs, and open new opportunities. But potential is not performance. Boards must distinguish between the two.
The prudent director will ask: what tasks can AI handle reliably, and what tasks require human judgement? That distinction is the line between innovation and recklessness.
In the end, the story of GPT 5.5 is a reminder that technology is a tool, not a saviour. It can assist, but it cannot replace the human capacity for reasoning, accountability, and trust.
Zimbabwe’s companies must embrace AI with eyes wide open, recognising both its promise and its limits. It is not yet Uhuru.
The journey towards reliable AI is ongoing, but we are not there yet.
Until then, let us keep our analysts at their desks, our boards vigilant, and our governance rooted in human judgement. That is the path to resilience in a digital age.
Dr Evans Sagomba is a Doctor of Philosophy and Chartered Marketer (CMktr, FCIM) with an MPhil and PhD in Philosophy. He specialises in AI, Ethics, and Policy Research, and is an AI Governance and Policy Consultant. Master’s and PhD supervisor. AI Ethics and Governance Lecturer. [email protected] ;Social media handles; LinkedIn; @ Dr. Evans Sagomba (MSc Marketing) (FCIM)(MPhil) (PhD) ; X: @esagomba



