AI Agents in Call Centres

AI Agents in Call Centres: Is it Working?

In founding a research business, you are required to interact with a lot of AI chatbots for support, building a website, and other activities, a rather strange symbiotic occurrence. My research leads me to review chatbot AI assistance, and to do so I need to interface with the very systems I am studying. What did I find?

AI in call centres is not failing. From what I can tell, it is doing exactly what it was designed to do.

I will expand on this. Systems are optimised for cost and throughput, not resolution.

To clarify, you get: – high automation – lower cost per interaction – but inconsistent outcomes

In my articles I point out how this works at small scale, where it is manageable. My intent is for organisations to further consider how systems behave when volume increases, because the problems compound.

The gap is not adoption. It is what happens when these systems are under pressure.

AI is now embedded across contact centres, but performance is uneven and often overstated. Industry reporting converges on the same pattern: high adoption, mixed outcomes, and a widening gap between efficiency gains and customer experience.

Adoption is not the issue. After reading surveys from organisations such as Gartner and McKinsey, I consistently see reports of contact centres that have deployed some form of AI (typically virtual agents, chatbots, or agent-assist tooling) with penetration rates approaching 80–90%, depending on how "AI" is defined.

What matters is not presence, but effectiveness. Resolution performance remains constrained.

Independent studies and vendor benchmarks indicate that fully automated self service resolution rates for nontrivial queries commonly sit well below what adoption figures suggest, with higher success rates only in tightly scoped use cases such as password resets or order status checks. Escalation to human agents remains the dominant path for anything involving ambiguity, exceptions, or emotional context. In practice, this means AI is handling volume but not necessarily solving problems end to end.

Understandably, cost pressure is the primary driver behind continued rollout. Firms report material reductions in cost per interaction, driven by deflection, shorter handling times, and lower labour intensity. ROI expectations are framed accordingly, with many programmes justified on multifold returns rather than service quality improvements. This incentive structure shapes system behaviour. Optimisation targets throughput and containment, not necessarily resolution quality.

NOTE: ROI = Return on Investment: the financial return generated from deploying AI in the contact centre relative to the cost of implementing and operating it. Formally expressed as: ROI = (Net Benefit ÷ Total Investment Cost)

Automation is scaling regardless of these limitations. In high volume environments, AI is already handling the majority of routine inbound traffic. "AI first" routing (where automation is the default entry point) is becoming standard operating practice. At this scale, edge cases accumulate, system behaviour shifts, variability increases, consistency reduces, and drift emerges as the system optimises under load. [the core topic around my “AI Workforce” articles]

The structural gap in the market is measurement. Organisations track adoption metrics: deployment rates, containment, deflection, average handling time. Fewer measure outcome quality in a way that is operationally enforced. Resolution accuracy, recontact rates, and customer effort are often secondary or lagging indicators.

This creates a false signal. Systems appear successful because they are active and cost efficient, even when they are not reliably solving the problem.

This is increasingly acknowledged in industry reporting. Research from Deloitte and Forrester highlights that AI adoption without integration into workflows, data systems, and governance models limits realised value. In practice, the issue is not model capability. It is system design under constraint.

What this means operationally is straightforward. AI in contact centres is behaving like any scaled system under pressure. It optimises for the metrics it is given, within the constraints it operates under. If those constraints prioritise cost and throughput, quality will degrade at the edges.

Human-in-the-loop does not resolve this at scale unless it is structured as a control mechanism rather than a fallback.

The relevant question is not whether AI is deployed. It is whether the system is designed to maintain outcome quality once volume, cost pressure, and optimisation dynamics take hold. That is where most implementations currently fail.

References

·         Gartner — Customer Service and Support Technology Trends https://www.gartner.com/en/customer-service-support/trends/customer-service-predictions Adoption of AI in contact centres is widespread, but performance varies significantly by use case and implementation maturity.

·         McKinsey & Company — The Economic Potential of Generative AI https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier AI delivers measurable cost and productivity gains, with the highest value realised when embedded into workflows rather than deployed as standalone tools.

·         Deloitte — State of Generative AI in the Enterprise https://www.deloitte.com/us/en/pages/consulting/articles/state-of-generative-ai-in-enterprise.html Adoption without integration into systems, processes, and governance models limits realised outcomes.

·         Forrester — Predictions 2024: Automation https://www.forrester.com/report/predictions-2024-automation/RES179955 Automation initiatives underperform when not integrated into end-to-end workflows and operational processes.

·         IBM — Global AI Adoption Index https://www.ibm.com/reports/ai-adoption High enterprise adoption of AI, with ongoing challenges in scaling value beyond initial deployment.

·         Zendesk — CX Trends Report https://www.zendesk.com/resources/customer-experience-trends/ AI is widely deployed in customer support, but resolution quality and escalation remain key challenges.

·         Intercom — State of AI in Customer Service https://www.intercom.com/blog/customer-service-trends/ Automation handles increasing ticket volume, but complex queries continue to require human intervention.

·         Microsoft — Work Trend Index (AI) https://www.microsoft.com/en-us/worklab/work-trend-index AI adoption is accelerating, with organisations prioritising productivity gains, often ahead of governance maturity. Automation initiatives underperform when not integrated into end-to-end workflows and operational processes.

Previous Article: Human-in-the-Loop Doesn’t Scale the Way You Think

Stay Connected

If you're interested in discussions on AI or how AI is actually delivered, across infrastructure, energy, networks, materials, and supply chains, please subscribe:

https://digitalbackbone-be8806.beehiiv.com/

Footnote

This article is part of a series exploring topics: AI is constrained by physical infrastructure, and increasingly shaped by economic behaviour at scale

Disclaimer

The views expressed in this article are my own and are intended for general information and discussion purposes only. They do not represent the views of any employer, organisation, or client.
© 2026 Rodney Terry – Digital Backbone. All rights reserved.