AI is making call centres more expensive – not cheaper

The call centre industry, once a beacon of cost-efficiency through AI integration, is now grappling with unexpected expenses. Inside South African call centres, AI tools have become a double-edged sword, with the true cost not merely the software licence, but the robust infrastructure required to power it. As AI tools, from voice neutralisation software to live agent coaching, become deeply embedded in customer engagement environments, the processing demands have surged over the past 18 months.

Operators underestimated this backend impact, leading to a financial dilemma: run AI workloads directly on endpoint devices, shift processing to the backend environment, or move infrastructure entirely off-premises through cloud providers.

Running AI workloads on endpoint devices pushes BPOs towards higher specification machines, moving away from Intel i5 to demand for i7 powered devices on the call centre floor. This shift prompts earlier hardware upgrades than originally planned. Alternatively, BPOs can keep devices standard, but this escalates backend server infrastructure needs, requiring higher compute density, advanced networking, and greater scalability.

Renting AI infrastructure from hyperscale providers like Amazon Web Services or colocation environments presents another option. While this avoids large upfront investments, it introduces ongoing rental costs, creating a complex long-term financial equation, particularly for BPOs with strict latency and compliance requirements. Traditional procurement models are under pressure as operators consider leasing and rental models for AI infrastructure, spreading costs over time while maintaining flexibility.

The shift underscores that AI is increasing operational costs within BPOs, shifting from labor efficiency to infrastructure efficiency. The competitive landscape may now hinge on the ability to afford powering AI at scale, rather than the cheapest labour model.