The AI models are already built. OpenAI built them. Anthropic built them. Google built them. Meta built them. If your competitive strategy in AI is to build a better foundation model than these organisations, you have not thought carefully about what you are up against. The infrastructure layer is effectively settled for now, and the winners are not going to be African companies working at that level. Not in this generation.
The application layer is crowded. Every week there is a new announcement of an AI tool built for African markets — customer service bots, content generation tools, translation apps, productivity software. Many of them are good. Some of them will survive. But the application layer is commoditising fast. The marginal cost of building an AI-powered application is falling every six months. What is genuinely scarce, genuinely difficult to build, and genuinely defensible is not the model and not the app.
It is the data that makes the model output actually relevant to how Africans think, decide, behave, and buy.
The application layer is commoditising fast. What is genuinely scarce and defensible is the data that makes AI output actually relevant to how Africans think, decide, and buy.
1. The calibration problem nobody is talking about
Every large language model was trained primarily on text produced by and for Western markets. This is not a conspiracy. It is just what the internet looked like when the training data was collected. The majority of written content available in English — articles, forum posts, consumer reviews, survey responses, academic papers — reflects the assumptions, reference points, cultural frameworks, and decision-making patterns of people in the US, UK, Western Europe, and East Asia.
When you use that model to generate insight about African consumers, it is not working from knowledge of African consumers. It is working from what it knows — which is mostly non-African — and projecting. The output sounds confident because large language models always sound confident. But underneath the confidence is a calibration problem. The model is making inferences that assume a set of cultural defaults that do not match the reality of the market you are asking about.
This shows up in practical ways. Consumer surveys designed by international consultancies that use Western psychological frameworks to segment African markets and produce findings that do not match what happens when you actually talk to the consumers. Marketing strategies built on behavioural assumptions that were derived from studies done in the US and applied to Lagos or Nairobi because nobody bothered to check whether the assumptions held. AI tools generating “African market insight” that is really just standard global market insight with the word “Africa” inserted.
The data problem is not dramatic. It is quiet. It produces outputs that are almost right, that sound authoritative, and that lead you slightly off course in ways that compound over time.
2. What building Hadisi revealed
I built Hadisi because I kept running into this problem from multiple directions. As a founder operating in Nigerian markets, I needed research that was specific to how actual Nigerians — across different income levels, different regions, different ages, different education backgrounds — actually think about money, about brands, about decisions, about risk. What was available was either too expensive, too slow, or not calibrated for African contexts.
We built over 1,400 AI personas representing real demographic groups across African markets. Not generic personas. Specific: the 28-year-old petty trader in Kano whose banking behaviour is shaped by a specific combination of trust patterns around fintech, religious attitudes toward interest-bearing products, and the specific economics of daily cash flow. The Lagos professional in her late thirties whose consumption is shaped by specific pressures from a family network that has expectations of her, by the specific status dynamics of her social class, and by the memory of specific economic shocks that created specific risk aversion in specific categories.
Building those personas was not quick. It required genuine cultural research — not Google searches, not survey data collected by someone who had never visited the communities in question, but serious rigorous work on the actual texture of how different groups of Africans live, think, and decide. That work is expensive. It requires expertise that is not abundant. It cannot be faked with a prompt and a hope.
Building those personas required genuine cultural research — not Google searches, not survey data from someone who had never visited the communities in question. That work is expensive and cannot be faked with a prompt and a hope.
3. Why this is defensible in a way that applications are not
An application can be copied in weeks. If you build a customer service bot for Nigerian banks and it works well, someone else can build a similar one in a month. The technology is not the barrier. The software architecture is not the barrier. The underlying AI capability is not the barrier because you are both renting it from OpenAI or Anthropic or whoever.
Data is different. Cultural intelligence of the kind that makes AI output actually relevant to African markets takes years to build correctly. You need researchers who understand the communities. You need methodologies that have been tested and refined against reality. You need the institutional knowledge of what questions to ask and how to ask them in ways that produce honest responses rather than socially desirable ones. You need the longitudinal data that shows how attitudes are shifting, not just a static snapshot.
All of that accumulates over time. It does not respond to a sudden influx of capital the way a software sprint does. You cannot buy your way to cultural intelligence in six months. You build it or you do not have it.
The companies that will be genuinely valuable in African AI markets, in five to ten years, are not going to be the ones that built the slickest application on top of GPT-4. They are going to be the ones that know something true about African consumers that nobody else knows, that have the research infrastructure to generate that knowledge continuously, and that have embedded that knowledge deeply enough in their products that it produces real differentiation.
4. The investment community has not priced this correctly
African AI funding has gone predominantly to applications. Understandably — applications are easier to demo, easier to explain, faster to reach revenue. A bot that answers customer service queries in Yoruba is immediately legible to an investor. A research platform that produces culturally grounded synthetic African personas is harder to explain in a pitch meeting.
But the investment logic is backwards. The applications are in a race to commoditisation. The research infrastructure is not. The value in the AI stack, in any geography, migrates to whoever controls the most irreplaceable layer. In Western markets, that is increasingly the proprietary data companies have about their specific users. In African markets, it is the foundational cultural and behavioural data that makes any AI output about African consumers actually accurate.
The value in any AI stack migrates to whoever controls the most irreplaceable layer. In African markets, that is the cultural and behavioural data that makes AI output actually accurate.
This is beginning to be understood. Not fast enough. The window for building the research infrastructure properly, before commoditisation pressure forces everyone toward shortcuts, is not infinitely wide. The companies that invest in the slow, expensive, unglamorous work of genuinely understanding African consumers — at the level of specificity that actual decisions require — are building something that will not be easy to dislodge.
The models are available to everyone. The apps are being built everywhere. The data that makes the models useful for African markets is being built by almost nobody. That gap is where the real opportunity is. And it will be filled by whoever is willing to do the hard work that the attention economy is ignoring.