Is AI making call centres better, or replacing them?

The South African picture: hiring and automating at the same time

If AI were simply replacing call centres, the employment numbers would be falling. In South Africa they are doing the opposite.

The business process outsourcing sector employs more than 270,000 people in South Africa, and industry body BPESA counted 8,180 new international CX jobs created between April and June 2025 alone, with the Western Cape and KwaZulu-Natal leading. Global operators keep expanding here: Teleperformance has publicly targeted growing its South African workforce to around 10,000 by the end of 2026.

That growth is happening while the same operators roll out AI tooling aggressively. The resolution: international clients offshore work to South Africa for the combination of language quality and cost, and AI raises what each agent-hour produces rather than eliminating the need for agents. The work changing hands is the repetitive tier-1 layer; the work growing is everything above it.

Where AI is measurably making centres better

Strip out the vendor theatre and four use cases are delivering in South African centres today.

After-hours and overflow answering

An AI voice agent answers at 22:00, on public holidays and during call spikes, takes a structured message or resolves the routine query, and hands anything complex to the morning queue. Before AI the realistic alternative for an SME was voicemail or nothing.

Routine tier-1 resolution

Balance queries, order status, bookings, password resets: high-volume, low-judgment interactions where callers mostly want speed. Modern voice bots resolve a meaningful share of these end to end, and every one resolved is a queue minute returned to human agents.

Agent assistance during the call

Live transcription, suggested answers, next-best-action prompts and automatic after-call summaries. The summary alone routinely saves minutes of wrap-up per call, which compounds across a floor into real capacity.

Quality assurance at 100% coverage

Traditional QA samples a handful of calls per agent per month. AI-driven QA scores every call for compliance phrases, sentiment and script adherence, and flags the exceptions for human review. Supervisors coach from evidence rather than anecdote.

Where AI still falls short of the sales deck

The gap between demo and production is where budgets go to die. Three limits matter for buying decisions.

Complexity and exceptions. AI handles the process it was trained on; the customer with three interlinked problems, a billing dispute and a contract nuance still needs a person with authority. Escalation design, not bot capability, decides whether those callers leave angrier.

Emotion and trust. Distressed, elderly or high-value callers measurably prefer humans, and a business that forces them through a bot to prove a point pays in churn. The honest deployment gives callers an early, obvious route to an agent.

Language reality. South African English accents are handled well by current speech models, but seamless multilingual service across all official languages remains a work in progress. Test with your actual caller base, in your actual languages, before believing any accuracy claim.

The jobs question, answered without spin

“Will AI take call centre jobs?” deserves a straight answer: it is changing them faster than it is cutting them, but the change is real.

The entry-level, script-following role is shrinking as tier-1 work automates. What is growing is the layer above it: agents handling complex and emotional interactions, QA analysts working with AI-flagged calls, bot trainers and conversation designers, and team leads coaching from full-coverage analytics. For a sector that is one of South Africa’s most important entry-level employers, the pipeline question is serious: if the bottom rung automates, operators and government programmes have to build new on-ramps into the sector.

For buyers the implication is simpler: vendors selling “cut half your headcount” are selling a spreadsheet, not an outcome. The operators getting results are redeploying saved hours into service quality, retention calls and revenue work.

The buyer playbook: adopt without breaking your service

A sequence that captures the augmentation wins first and defers the risky bets.

Keep reading