As AI transitions from answering questions to executing tasks, African enterprises face a unique frontier. Beyond simple chatbots, voice is emerging as the primary digital interface. In this interview with ADEYEMI ADEPETUN, the Founder of BimpeAI, Samuel Adekunle explains how AI explores local language capabilities, navigating shadow voice AI and building native infrastructure designed specifically for African institutions.

Before now, the saying was the next phase of AI was just chat, but execution. What does that mean in practice for banks, public sector organisations and large enterprises?
Chat is when AI answers a question. Execution is when AI completes a task. For a bank, that could mean following up with customers, supporting loan workflows, helping with onboarding, routing complaints or checking the status of a request.

For a public sector organisation, it could mean helping citizens get answers, complete applications, or follow up on service requests. The value is not that the AI can talk. The value is that it can connect to the right system, follow the right process, and help the organisation move faster.

Why is voice becoming such an important interface for AI in African markets?
Voice is natural for African markets because many people already prefer to call. Not everyone wants to use an app, fill a form, or type a long message in formal English. In Nigeria, especially, people move between English, Pidgin and local languages very easily.

Voice makes AI more accessible. It also helps enterprises serve more people without putting all the pressure on call centre teams. For me, voice AI is not just a feature. In markets like Nigeria, it can become the main way people interact with digital services.

Lately, we have been hearing about shadow voice AI. What is it and why should enterprises pay attention to it?
Shadow voice AI is AI voices speaking to your customers without your approval, control or knowledge.

Picture this: someone clones the voice of a bank’s customer-service line and calls account holders asking them to confirm a transfer. Or a third-party vendor spins up an automated calling agent to chase loan repayments, using the bank’s name, with no script approval and no audit trail. The customer hears an official-sounding voice and trusts it; that is exactly what makes it dangerous.

For a bank or a public institution, this is a fraud and liability problem waiting to happen. The technology to clone a voice convincingly now costs almost nothing. The answer is not to ban voice AI; it is to govern it: verified agents, approved scripts, consent, full audit logs and a clear handover to a human. Enterprises that don’t put this in place now will be reacting to an incident later.

Nigeria is a multilingual country with complex accents, code-switching and local language patterns. What does it take to build voice AI that works in Yoruba, Igbo, Hausa, Pidgin, and Nigerian English?
It takes more than translation. Nigerians do not speak in one clean language pattern. We mix English, Pidgin, Yoruba, Igbo, Hausa, slang, and local expressions. The same person can switch tone and language depending on who they are speaking to.

To build useful voice AI, you need local speech data, real accents, domain-specific context, and a strong understanding of how people actually speak.

That is why we are investing in African voice capability. The goal is not just to make AI speak local languages. The goal is to make it useful in real conversations.

A lot of AI infrastructure is built on Western language, data and regulatory contexts. What are the risks of simply importing those systems into African markets?
The first risk is the model simply doesn’t understand the user. A speech system trained on American or British English can hit 95 per cent accuracy in a demo and then fall apart on a Lagos phone line.

Nigerian English, Pidgin, a code-switch mid-sentence, background market noise, a 2G connection dropping syllables. We’ve watched off-the-shelf models mis-transcribe Nigerian names into something else.

The second risk is trust. A misunderstanding in a casual app is annoying. A misunderstanding when someone is checking a loan balance or a benefit payment breaks the relationship, and the institution carries the blame. That’s what we try to eliminate.

The third risk is dependency. If the whole stack is built elsewhere, Africa is always renting infrastructure that wasn’t designed for its languages, its regulations, or its institutions. We think that layer has to be built here.

How do you think about trust, safety and accountability when an AI agent is speaking to a bank customer or citizen on behalf of an institution?
For enterprise AI, trust is the product. An AI agent speaking for a bank or public institution cannot behave like a general chatbot. It needs limits. It needs approved workflows. It needs to know what it can answer, what it can access, and when to hand it over to a human. There should also be logs, permissions, and clear accountability.

The institution remains responsible for what the AI does. So, the agent has to be designed with safety from the beginning, not added later.

What makes deploying AI agents inside enterprise environments different from building consumer chatbots?
Consumer chatbots can be broad and experimental. Enterprise agents need to be reliable. They have to connect to real systems like CRMs, databases, support tools, payment systems, and internal workflows.

They also have to follow compliance rules and protect customer data.

The enterprise does not care that the AI sounds impressive. They care whether it reduces workload, improves response time, supports customers, and creates a clear business outcome. That is the difference.

Where do you see the biggest opportunities for AI voice agents in Nigeria: banking, public services, healthcare, telecoms, education, or commerce?
The strongest near-term opportunity is financial services. Banks, lenders, and fintechs deal with huge volumes of repetitive customer interactions: onboarding, KYC, loan follow-up, account support, complaints, and customer education.

Public services and telecoms are also major opportunities because they serve millions of people and already depend heavily on call centres. Healthcare and education are important too, but they need more careful deployment because the stakes are sensitive.

What does Africa need to build for itself in the AI era, rather than only consuming tools built elsewhere?
Africa needs to build its own language and voice infrastructure. We need AI that understands African accents, African English, Pidgin, and local languages. We also need AI that connects into our payment systems, banks, public institutions, and business tools.

The opportunity is not just to use AI. It is to shape the AI stack around our own realities.

That is where I think African founders can make a real contribution globally.

What does success look like for BimpeAI over the next five years?
Near-term success is measurable, so let me be specific. In five years, I want BimpeAI’s agents handling 200 million customer conversations a year across voice and messaging, deployed in 50 banks and 80 public institutions, in at least 10 African markets beyond Nigeria.
 
I want our agents to support banks, public institutions, telecoms, and large service businesses across voice and messaging.
 
I also want us to prove that African founders can build serious AI infrastructure, not just local apps on top of global platforms. If we do that, BimpeAI will not just be a company using AI. It will be part of the infrastructure layer for how AI works in African markets.

BimpeAI started with WhatsApp and hospitality use cases, but the company has now moved into enterprise AI deployments. What changed in your understanding of the market?
We started with hospitality because the problem was in our face. A restaurant would get 40 WhatsApp messages an hour asking, “Are you open,” “can I book for six,” “did my payment go through,” and a human had to answer every one while none of it touched the booking system, the till, or the customer record.
 
The lesson wasn’t “businesses need chatbots.” It was that a chatbot that only talks is useless.

Businesses need AI that can do the thing: take the booking, confirm the payment, update the record.
 
Once we saw that, the move to enterprise was obvious. A bank has the same gap, just with higher stakes and bigger volumes. The conversation is disconnected from the systems that matter. That’s the problem we now solve for banks, lenders, and public institutions.