AI implementation in contact centres is introducing a parallel cost structure that organisations are failing to account for, rather than delivering the promised efficiency gains. The industry's rush to deploy large language models as a cost-reduction mechanism has created a hidden tax: every inbound interaction now passes through compute-intensive AI layers for triage and routing decisions, yet most contact centre work remains highly structured and deterministic. When billions of interactions are processed annually through unnecessarily complex models, even marginal inefficiencies—prompt bloat, irrelevant data processing, latent delays—compound rapidly across the operation. The economics deteriorate further when token costs per interaction exceed the human labour they theoretically replace, a threshold many organisations cross without realising it. For teams already embedded in platforms like Zendesk or Salesforce, this raises a critical question: are your voice automation implementations genuinely reducing cost-to-serve, or are they simply adding a new operational expense layer that masks deteriorating unit economics?
The root cause is architectural misalignment. Organisations are applying general-purpose large models to high-volume, low-complexity tasks—identity verification, balance enquiries, intent detection—where specialised micro-models would deliver superior accuracy, lower latency and dramatically reduced compute costs. This design inefficiency can add £200m to £500m annually through increased handling time alone. The solution requires operational discipline: distributing workloads across model types, reserving large models only for interactions requiring genuine reasoning, and using smaller language models for orchestration and routing. This approach transforms model architecture from a purely technical decision into a commercial one, directly influencing ROI and operational metrics including containment and average handling time. For support leaders and CX consultants, the implication is stark—the organisations seeing strongest results are not those deploying the largest models indiscriminately, but those applying intelligence with measurable cost justification at every layer of the stack.
The hidden AI costs impacting contact centre efficiency allthingsbusiness.co.uk