AI in Africa. The future of AI for B2B is exciting but it may not be LLMs and Deep Learning

We are still in the middle of the AI hype bubble, but the rate of change is increasing so we will exit the hype markets faster. When this happens we will go directly onto a business roller coaster, it’s going to be a white-knuckle ride. The sales and marketing AI markets look likely to exit the hype cycle first, before the end of this year, but our focus is on go-to-market for B2B businesses so our comments have to be taken in the context of how we see the market.

The global AI markets are driven by LLM’s and Deep Learning but are they relevant to emerging markets? Will they work and add the same value in our B2B markets? They will lift the base case but offer average or lowest common denominator outputs which may not be relevant to our business markets. They will lack the most vital ingredient of all, context. I believe the future for emerging markets B2B AI is Small Language Models and Simple Statistical Models. Here’s why.?

Emerging Market AI Drivers??

1. African B2B Tech Market Drivers. The West sees AI innovation as digital-to-digital, and African markets are driven by digital to bricks-and-mortar innovations. We use tech differently and are solving different problems.
2. USD Exposure. High currency volatility in Africa is driving business models and buying behaviours. Centralised compute off 100% USD-based backbone is becoming an increasingly fragile business model. Exchange rate fluctuations often push marginal business models over the line. Two-tier models are emerging where the primary load is carried through the USD backbone and the larger secondary load is carried on cheaper, locally priced servers or bare tin. Managing USD exposure is easier for smaller models. India and China are present in these markets as a result.??
3. Data Connectivity remains the universal African currency. Mobile data remains expensive and we are a mobile-first market. Large centralised compute AI models mean lots of data has to move and someone has to pay. Digital and financial inclusion and creating sustainable business models means collapsing the cost to serve, not increasing it.?
4. Our Markets are Complex, Distributed and Fragmented, culturally, geographically and skills-wise and we remain people-first economies. The TAM is enormous but we are a market of micro-markets. Smaller models offer higher value in each market and make micro-markets financially viable.?
5. The Western Model of Centralised Compute AI to cater for LLMs and Deep Learning with high GPU costs makes many business models unviable in our markets. Access to market and the cost of distribution remain high, adding high back office costs will not be sustainable. Decentralised compute models are quicker to build, cheaper to run and collapse the cost to enter.
6. We are Data-poor. Our data is fragmented, distributed, unstructured and often not easily available. Centralising, structuring and cleaning it is time-consuming, slow and expensive. Our markets are fragmented so centralising the data may not yield good results. Smaller models are much easier to build with poor data. In data-poor markets improving simple decision making can yield exponential results.?We don't need to be right, we need to be better than we were. It's better to be roughly right than precisely wrong.
7. Skills. The gold rush for data scientists makes launching AI-first businesses expensive. High GPU costs and high AI skills costs increase and front-load the cost to launch. Our markets are largely unfunded, we are not able to compete with the West for skills. A decentralised skills pool building small, simple, locally relevant models off poor data is something we can excel at and are markets we can lead and defend.???

We are working hard to gather the data to understand the market better, we work closely with founder-led AI businesses, so the comments above, while being informed are not supported by data yet. We don’t include the “We’re building a bot” market in the AI market. This market will remain a blood bath as big guys continuously collapse it in the race to the bottom.

Centralised compute models requiring large clean data sets using USD GPUs with high connectivity costs will make many AI business models unsustainable.

We remain complex fragmented distribution-first markets. We are a market of micro-markets where every model has a small TAM and ROI will come from more efficient, cost-effective models.

Will SLMs and Simple Statistical Models outperform LLMs and Deep Learning Models? If the context is different and the models specific, they have to.

Elevage in winemaking is improving every step of the process by 1% to produce a great wine. We need to apply these principles to AI in our markets but in a bottom-up way. We don’t have to compete with the West, we need to help our local markets be better than they were. As always, customer-centricity is the best defence. This is a market we can develop, monetise, lead and defend. The West is building aircraft carriers, we need to build an armada of speed boats.

After the mature digital economies have been served by their centralised compute models customer demand will fragment into micro-markets. If you want to see the future, come to Africa.