By Bruce von Maltitz, CEO of 1Stream

What are the many AI voice bots that are about to start helping companies with customer engagement actually going to say? For all the hype and buzzwords around AI, what really counts is everything behind the voice.

There’s a tendency to think the magic is in the conversation itself; the tone, the speed and the way the bot sounds. We can safely assume that no one wants to talk to something that sounds like a generic robot when the alternative is an AI voice bot that sounds local and feels natural, all while offering genuine help.

But a good voice bot experience isn’t created by the voice alone but depends on the quality of the data, knowledge and intent behind every response. If the underlying information is inconsistent, outdated or disconnected from how the business actually works, the interaction is bound to reflect that.

A good AI voice bot starts with purpose

Before accents, personality or tone, businesses need to be clear on what the bot is there to do. It should not try to be everything to everyone – especially during the first iteration. A sensible place to start is the top 5 or 10 reasons customers are getting in touch and ask: can we resolve these well, quickly and in a way that feels useful? And that is where the data comes in.

Historical emails, chats and voice recordings are the knowledge base of what customers are really asking, and offer insight into the intent behind the questions, and which contact centre human responses have consistently led to the best outcomes.

This is also where modern AI starts to separate itself from traditional IVR. Instead of only listening for keywords, AI brings in context. With the right mix of speech recognition, language understanding and retrieval from a company’s own knowledge base, the bot can start responding based on what the customer means, not just the exact words they used.

Local really is lekker

Human speech is dynamic by nature. People interrupt, they switch tone, and there are many different accents in which the same issue can be phrased in completely different ways. If AI voice bots can’t handle that natural messiness, customers will feel frustrated very quickly. 

Then there’s the unique South African context to consider. What is the accent? What is the language? What is the tone? What is the personality? These factors shape whether the experience comes across as cold and generic, or as something that feels natural and relevant to a South African customer.

In some use cases, that can even influence how willing a customer is to engage. In areas like debt collection, for example, some people may feel more at ease speaking to a bot because the exchange is purely transactional and less judgmental.

The good news is none of this removes the need for human agents. If anything, it changes where they add the most value. AI is well suited to handling high-volume, routine tasks such as identity verification or balance enquiries, which are often among the biggest drivers of agent burnout. That gives agents more room to focus on higher-value, empathy-driven interactions where human discretion is needed most.