Bruce von Maltitz, CEO of 1Stream.
While traditional automated systems have long supported the foundations of digital customer service in SA, the integration of AI voice agents is a turning point that will revolutionise the way consumers engage with brands.
So claims Bruce von Maltitz, CEO at 1Stream, who explains that for many years, interactive voice response systems provided the necessary framework for routing calls. However, substantial investment in local infrastructure has moved the contact centre industry forward and it now facilitates the combination of advanced large language models with high-speed speech-to-text engines.
AI can handle high volumes of routine queries and serve as a sophisticated copilot for the workforce, claims 1Stream.
While there is significant interest in deploying AI voice agents, Von Maltitz says 1Stream is limiting its engagement to proofs of concept with five enterprises.
“Making the AI voice agent available is relatively quick and easy, but integrating into customers’ environments takes time and slows deployment,” he says.
Von Maltitz advises that companies ensure their data infrastructure is ready to support memory-rich AI voice agent interactions. This is because an AI voice agent is at its most powerful when it has context – for example, knowing that a caller has already engaged with the brand via WhatsApp or e-mail earlier in the day.
“AI voice agents have moved beyond keyword recognition and understand context and intent through a layered process that combines linguistics, mathematics and massive amounts of data. So it is a multi-layered process,” says Von Maltitz.
“It starts with high-speed automatic speech recognition engines hosted locally in South Africa to eliminate the latency gap. This is followed by natural language understanding and large language models that identify the intent behind a customer’s words. Crucially, we use retrieval-augmented generation, which grounds the AI’s responses in an organisation’s proprietary data (manuals, FAQs and CRMs) rather than general internet knowledge.”
These agents can make mistakes or misinterpret a call, he adds.
“This is why the system uses confidence scoring and escalation rules. If the AI is unsure about a caller’s intent, it will ask for clarification or transfer the call to a human agent,” Von Maltitz continues.
Adherence to local POPIA requirements and global ISO 42001 standards for AI management will help to ensure AI voice agents are secure and ethical, adds Von Maltitz.
Another consideration is whether this technology will replace humans within the contact centre environment.
Von Maltitz believes while AI voice agents are currently not replacing people, there is a chance it could happen in the future. He also suggests the technology is creating jobs for people delivering the solution.
“Overall, we view this as a shift in role complexity. AI handles high-volume, routine tasks – like identity verification or balance enquiries – which are the primary drivers of employee burnout. This allows human employees to focus on high-value, empathy-driven interactions. For the organisation, it provides the ability to scale during peak periods (like Black Friday) without the disruptive cycle of temporary hiring.”
According to Von Maltitz, human teams still monitor performance, update information and control when calls are escalated, so the AI acts as a first-line assistant rather than a replacement for human agents.
Alignment with business objectives
Technology professionals continue to emphasise the need to align AI with business objectives, operational goals and requirements.
“By first identifying the unique, quantifiable business challenges faced by an organisation and then using rapid, execution-led AI approaches, companies solve real-world problems more efficiently,” says Mark Walker, director at technology consultancy T4i.
AI expert Johan Steyn, founder of AIforBusiness.net, adds to the discussion. In a blog post titled: ‘The digital workforce moment: Why AI agents are no longer just software’, he writes: “When an AI agent can open a case in a CRM, update a customer record, create a purchase request, schedule an interview, reconcile a payment exception or escalate a compliance issue, it starts behaving less like software and more like a new class of employee.”
He suggests that business leaders need a practical operating model to manage this new class of employee. This means defining the agent’s role, its purpose, level of access and, initially, let agents operate in constrained environments before expanding autonomy.
“The biggest mistake organisations can make is to treat AI agents as a shiny feature, rather than a new category of operational actor. Once software can do work, not just describe work, the organisation must respond with the same seriousness it applies to hiring, supervision and risk management. In South Africa, where skills constraints and service pressure are real, digital operators could unlock meaningful capacity,” says Steyn.
While the local level of AI voice agent technology adoption remains low, says Von Maltitz, he predicts that 2026 will see a surge in the number of these agents entering the local market.
