In the global AI conversation, size often steals the spotlight. Every month, it seems another model is released with more parameters, more training data, and more complexity than the last.
OpenAI’s GPT-2, which made waves earlier this year for being “too dangerous” to release in full, drew massive attention. Just last month, Google introduced BERT into its search engine, calling it the biggest leap in search relevance in five years. These breakthroughs are impressive. But in Africa, they raise a different question entirely.
Do we need bigger models, or better questions?
That question has shaped how I view intelligence systems from a distinctly African marketing perspective. Working across fintech, retail tech, and platform strategy in Nigeria, I’ve found that local success with AI rarely depends on model size. Instead, it depends on context, accessibility, and the clarity of the problems we’re trying to solve.
Africa is not a monolith. It’s multilingual, economically diverse, and deeply informal in structure. The consumer experience in Lagos differs wildly from that in Accra, Nairobi, or Cape Town. Data is limited, mobile costs are high, and systems are often offline by default. An enterprise chatbot with multi-language NLP or a recommendation engine trained on millions of global users might sound exciting in theory, but may not hold up in the markets that need AI the most.
That’s why some of the most effective AI applications I’ve seen on the continent are the simplest ones.
Take Kudi.ai. Founded in 2017, Kudi didn’t build a massive model. Instead, it created a conversational financial assistant that worked on Facebook Messenger and USSD.
It helped users send money, pay bills, and check balances,all through text. No broadband, no downloads, no friction. Just immediate, intelligent service.
Or Lara.ng, which launched in 2018 to solve an everyday challenge: navigating public transport in cities like Lagos and Ibadan. Instead of a flashy app or map interface, Lara used a chatbot to give users fare estimates, directions, and route options by simply typing questions. Again, the intelligence wasn’t artificial. It was contextual.
These tools asked smarter questions of their environments, not bigger ones. They respected local limitations, optimised for human behaviour, and delivered real value.
In my own experience, especially while working at Fast Pay and LONTOR Hi-Tech, this became increasingly clear. At Fast Pay, we were scaling a fintech product across road and marine transport vendors. The bottleneck wasn’t technology. It was user confidence.
We needed to know when and why vendors hesitated, then design onboarding journeys that were informative and intuitive. That wasn’t machine learning, it was human learning applied to systems.
Later at LONTOR, a high-growth consumer electronics brand, we relied on a mix of brand intelligence, market analytics, and customer behaviour data to improve our go-to-market execution.
It wasn’t about chasing AI trends. It was about asking: What insights are we missing? How do we test faster? Where do field activations fall flat?
That mindset, of inquiry and observation, was often more powerful than any algorithm.
Too often in African tech circles, AI is framed as a luxury add-on. Something to be layered in after success, not as a foundation. But what if we flipped that?
What if AI began with better segmentation, smarter targeting, and empathy-led system design? Not just automation, but augmentation, tools that make people more capable, not replace them.
And what if the best way to build AI in Africa was to start from marketing logic? Where you test, learn, observe, iterate. Where questions come before answers, and humans before systems.
OpenAI’s GPT-2 is undoubtedly remarkable. So is Google’s BERT, now transforming search in ways unimaginable a few years ago. But even as these models rise in prominence, we can’t afford to forget what matters here: relevance, usability, and fit.
We don’t need to out-scale the world. We need to out-context it.
The future of AI on the continent won’t be led by the companies with the largest datasets. It will be led by those who know what questions to ask and who they’re asking them for.
That’s not just artificial intelligence. That’s human insight at scale.
