AI and Automation in CX: What Actually works in African Markets

Most of the AI conversations I join in customer experience start in the same place. The case studies come from Silicon Valley, London or Stockholm, the customer is assumed to have a smartphone and a generous data plan, and the question on the table is how quickly a chatbot can replace a queue. African markets begin somewhere different. I have come to believe that, in several important ways, it is a better place to begin.

What works here is rarely the most sophisticated technology. It is the technology that has been designed around how African customers actually live, connect, speak and decide whom to trust. The organisations getting this right are not simply importing an AI strategy. They are building one that fits.

Start where your customers already are

The GSMA Mobile Economy Africa 2025 report makes a point every CX leader on the continent should keep close. Despite widespread network coverage, almost one billion people in Africa are still not using mobile internet, around 63% of the population. The barriers it names are device affordability, limited digital skills and a lack of relevant content. Any automation strategy that quietly assumes a modern smartphone, a data bundle and a stable connection has been designed for a minority of the market, however polished it looks in a demo.

Safaricom in Kenya offers one of the clearest examples of building around that reality. When it introduced its AI assistant, Zuri, in December 2018, it did not position the chatbot as a replacement for anything. Zuri launched on Telegram and Facebook Messenger as an addition to self-service options customers already relied on, including USSD and the mySafaricom app. At the time, Safaricom reported 24 million daily uses of its USSD service and more than 700,000 customers using the app every day. In early 2020, Zuri arrived on WhatsApp, the messaging platform Kenyans were already using most, handling everyday tasks such as M-PESA reversals, PUK retrieval, buying bundles and checking balances.

The lesson is not that WhatsApp is the answer for everyone. The lesson is that the channel should be chosen by the customer's habits, not by the vendor's roadmap. Automation earns adoption when it shows up in a place customer already trust, alongside the options they already understand.

Language is the experience, not a feature

South Africa has twelve official languages, with South African Sign Language added in 2023. Across the continent, the number of spoken languages runs into the thousands. Most customers do not move neatly between them either. They switch languages within a single sentence, mixing English with isiZulu, Sesotho, Afrikaans or Pidgin depending on the moment and the subject. An AI model trained mostly on standard English will hear that as noise. A customer hears it as normal conversation.

Having spent more than eleven years at Vodacom earlier in my career, I was encouraged to read about the work behind its TOBi assistant. In a case study published by Microsoft, the local team explained that it was starting with isiZulu, and Avi Moodley, a principal machine learning specialist at Vodacom, noted that some customers in rural areas speak only one language other than English. In Ghana, Vodafone Ghana brought TOBi to WhatsApp and added the ability to respond to queries in Pidgin English, with complex cases handed over to live agents.

The local ecosystem is growing too. Lelapa AI, a South African research and product lab co-founded by Pelonomi Moiloa, launched its Vulavula platform with support for English, Afrikaans, isiZulu and Sesotho. It is built to handle code-switching and is designed to run where computing power and connectivity are limited. That last point matters more than it might seem. A model that needs a fast connection to work well is a model that works best for the customers who were already well served.

When a customer can explain a problem in the language they think in, they explain it faster and more accurately. Fewer misunderstandings mean fewer repeat contacts. More importantly, the customer feels respected before the problem is even solved, and that feeling is the foundation of trust.

Keep the door to a person open

Many of the tasks being automated in African markets involve money. A mobile money reversal, a SIM swap or a banking query is not a casual interaction. When something goes wrong, the customer needs to know that a capable human being is within reach. The strongest examples I have seen treat that handover as a designed part of the journey, not an escape hatch. Vodafone Ghana was explicit that complex scenarios would pass to live agents. Safaricom and Vodacom both built one-time password security into their automated journeys, recognising that confidence in the channel matters as much as convenience.

The practical test is simple. Can a customer reach a person without repeating everything they have already told the bot? If the answer is no, the automation has moved the effort onto the customer rather than removing it.

Use AI behind the frontline as much as in front of it

Some of the most valuable opportunities sit where customers never see them. In many contact centres across the continent, a large share of calls happens in languages that quality teams cannot easily review or analyse. Speech analytics and transcription built for African languages can change that, giving leaders visibility of conversations they have never been able to learn from. Real-time support for agents, such as live transcription and summaries during code-switched calls, helps the person on the line understand the customer and document the conversation accurately.

This is AI that makes people better at their jobs rather than removing them from the conversation. It also tends to deliver value faster, because it improves the interactions that already carry the most complexity and emotion.

What works, in practice

When I look across the examples that are genuinely working, a few principles keep repeating. They meet customers in the channels they already use, including the ones that do not need a smartphone. They take language seriously from the very first design decision, not as a later upgrade. They keep a person reachable and make the handover feel seamless. They protect trust around money and identity. They also measure success by whether the customer's problem was resolved, not by how many conversations the bot contained.

None of this is about holding back on AI. It is about building it with the confidence that African customers deserve technology designed for them, rather than technology adapted for them as an afterthought. The continent has every opportunity to show the rest of the world what inclusive automation looks like.

“The best AI in African customer experience will not be the most advanced. It will be the one that speaks the customer's language, meets them where they are, and always knows when to hand them to a person.”