The Hidden Tax of AI in Contact Centres – Sally Hodgin, Principal AI Consultant at Connect discusses the true cost of AI in the contact centre
AI is increasingly being positioned as a way to reduce cost-to-serve in the contact centre, but in reality, many businesses are adding an extra layer of cost long before they see any return. As organisations race to implement AI-first voice automation, a new cost layer is quietly emerging alongside that human spend: LLM compute, orchestration infrastructure, and the operational inefficiencies that can accompany poorly designed Voice AI architectures. Many organisations are unknowingly layering AI costs on top of existing operating costs, rather than replacing them.
The UK contact centre industry already spends billions of pounds each year on servicing inbound calls, the majority of which are still managed by human agents. That cost reflects more than just answering queries. It covers complex regulated interactions, customer authentication, issue resolution and the need for consistency across millions of conversations every day. In response, many organisations have accelerated AI adoption in an attempt to reduce operational cost and improve efficiency.
Sally Hodgin, Principal AI Consultant at Connect, believes that the issue isn’t AI itself, it’s how it’s being applied, and discusses how businesses can avoid being stung by the ‘hidden tax’ of AI.
The reason AI increases cost before it reduces it
Billions of interactions are processed by AI systems each year, regardless of whether they are ultimately resolved by automation or a human agent. In many cases, indiscriminate use of large language models introduces unnecessary compute, latency and risk, particularly when applied without clear justification.
This is because most interactions now pass through an AI layer before a decision is made. Every inbound call typically moves through a Voice AI triage layer to determine whether it can be automated or routed to an agent. At production scale, even small inefficiencies in AI architecture compound rapidly across millions of interactions. For businesses globally, this creates a significant new baseline cost.
These costs are often further compounded by design inefficiencies. For example, the overuse of large models, prompt bloat, and the processing of irrelevant information all increase the number of tokens per interaction, directly driving up costs. Even a few seconds of unnecessary delay on a call can increase average handling time. When large complex models are misapplied and irrelevant information is processed it can add anything between £200 million and £500 million per year purely through increased interaction time alone. This is a hidden tax no business wants to bear.
When the AI business case starts to erode
The economics of AI start to fail when the cost per interaction exceeds the human effort it replaces. This often happens when large, general-purpose models are applied to high-volume, low-complexity tasks. These models carry significantly higher compute and energy requirements, yet many contact centre interactions are highly structured and deterministic, such as identity verification or balance enquiries, where that level of reasoning is unnecessary. When automation scales, even minor inefficiencies in precision, latency or model orchestration compound rapidly. The result is a model where cost increases faster than value, undermining the original business case for AI adoption.
A more disciplined approach to AI architecture
A more effective approach involves aligning model complexity to the operational requirements of the task. Instead of relying on a single large model, organisations should distribute workloads across different model types. Micro-models can handle precision-based tasks such as intent detection, entity capture and ID&V. For example, ID&V remains one of the largest barriers to scaled automation in the contact centre, despite being highly deterministic in nature. This creates a bottleneck that limits automation and slows the efficiency gains that could be realised further down the line. By enabling specialised models to handle these tasks, that constraint is removed. These models can operate faster, deliver greater accuracy for specific use cases, and do so far more cost-effectively.
Additionally, small language models can manage orchestration, routing and workflow logic, while large models should be reserved for interactions that genuinely require deeper reasoning or contextual understanding. This layered approach ensures that organisations only pay for high-intensity compute where it adds measurable value. It improves governance, predictability and scalability across enterprise AI operations, as well as key operational metrics, including average handling time, containment, and cost to serve. Ultimately, this is about control. Model architecture becomes a commercial decision, not just a technical one, directly influencing cost, performance and return on investment.
Conclusion.
AI has a clear role to play in the future of the contact centre, but cost reduction is not automatic. Without careful design, it introduces a parallel cost structure that sits alongside the existing human workforce. The organisations seeing the strongest results are not those deploying the largest models indiscriminately, but those applying intelligence with operational discipline. Avoiding the hidden tax of AI comes down to one principle: only pay for complexity where complexity is required.
Sally Hodgin is Principal AI Consultant at Connect
Connect is a global customer experience specialist, systems integrator and digital transformation partner with industry-leading, technology-enabled capabilities. We orchestrate modernised and operationally sustainable CX that is ready to scale and mature. Founded in 1990, we’ve evolved alongside every major industry shift; from on-premise to cloud, voice to omni-channel, and now AI-enabled experience. Built on this extensive market experience, our approach is focused on delivering outcomes-based solutions that accelerate value, informed by what it takes to operate, scale, and continuously improve CX in live environments.
Fort additional information on Connect visit their Website
