The broader promise of Artificial Intelligence (AI) requires deliberate fine-tuning for African markets, not the assumption that what works in New York will work in Nairobi, Writes Ken Oidamae Tobiko, an investment consulting analyst based in Nairobi.
There is a quiet arms race underway in global finance. Investment banks, hedge funds and asset managers are deploying large language models to read earnings calls, scan news sentiment and generate research at a speed no human analyst can match.
In developed markets, sentiment analysis has become a credible input into investment decisions. The inputs are abundant: earnings transcripts, analyst notes, structured disclosures going back decades.
The signal is not perfect, but it is consistent and historically deep. Over time, these systems learn how executives hedge, how risk is disclosed and how optimism is signalled.
Africa offers little of that consistency.
Without localised training data, models cannot distinguish between stylistic convention and genuine signal.
The data is not just thinner — it is different.
Outside a handful of names, corporate disclosure across African exchanges is thin, irregular and often unstructured. On the Nairobi bourse, most listed companies receive fewer than two analyst reports a year while some receive none. Financial histories are fragmented across exchanges, regulators and company websites and sometimes are rarely structured and rarely current.
This is not just a smaller dataset. It is a fundamentally different one.
Most off-the-shelf models are trained on Western corporate data and English-language financial commentary shaped by developed-market norms. Apply that model to Sameer, Uchumi or Tanga Cement and the problem becomes clear. The model is not just extrapolating. It is potentially misreading.
Language, in finance as elsewhere, is local. The management tone in Nairobi, Lagos or Johannesburg reflects different incentives, regulatory environments and cultural norms.
A statement that reads as cautious in a US context may be standard practice elsewhere. What appears neutral may mask meaningful risk. Without localised training data, models cannot distinguish between stylistic convention and genuine signal.
The result is a subtle but material risk: false precision. An investor presented with a clean sentiment score may assume analytical rigour where there is none. The model may simply be projecting Western language patterns onto markets that do not share them.
The Allocation Consequence
There is a second-order effect worth naming. Institutional investors increasingly use AI-assisted tools to screen markets. If African equities produce weak or noisy outputs, they risk being deprioritised. The problem is the data, not the fundamentals.
Over time, that feeds directly into capital allocation. Markets legible to machines attract more informed capital, tighter spreads and better price discovery. Those that are opaque risk remaining chronically under-invested.
The risk is not just that Africa goes undercovered. It is that African markets get systematically filtered out of the next generation of portfolio construction entirely.
One could argue the gap is temporary. As African markets deepen, disclosure will improve, datasets will grow and models will adapt. That is directionally true but incomplete.
Data does not just need to expand, it also needs to be relevant. Effective models require localised context and an understanding of how information flows within African markets. The constraint is not just quantity of data, it is also the design.
The Opportunity is Significant
For now, current off-the-shelf AI tools should be used with caution in African equity research. They are useful for summarisation and coverage expansion. As signal generators, particularly for sentiment, I would argue they are structurally limited.
But the opportunity is real. Firms that invest early in building proprietary African datasets and training region-specific models could develop a meaningful informational edge. In markets where market inefficiencies persist, better interpretation of limited data is itself a source of alpha.
The broader promise of AI in investing is convergence where AI tools become universally applicable across different investment geographies. That promise is worth holding onto. However, it requires deliberate fine-tuning for African markets, not the assumption that what works in New York will work in Nairobi.
