One of the principal areas where AI was expected to deliver value was in customer service, but recent research by Gartner found that 95% of organisations intend to retain their human agents, leading it to predict that half of businesses will abandon their plans to replace these workers with the technology by 2027.
There’s a general sense that AI hasn’t lived up to the hype. But the real issue isn’t AI itself, it’s that the industry has failed to treat AI adoption as a digital transformation journey. Many contact centres are attempting to implement AI without first building the data, systems, and operational foundation required to make it succeed.
That said, AI will undoubtedly be a core component of the contact centre, not because it replaces humans, but because it enables smarter, faster, more contextual interactions for humans, with humans, and sometimes instead of humans. The real question is not whether AI will continue to be used, but how we adopt it in a way that works effectively, which will require a structured form of implementation.
Understand Before You Automate
Firstly, before automating anything in the Contact Centre, AI should be used to augment our understanding. Most organisations still lack a clear, joined-up view of their customer journeys across digital and assisted channels. What enquiries are being made? What drives repeat contact? Where is customer effort being exerted? To answer these questions, contact centres must develop a robust data strategy. This should ensure real-time access to both structured and unstructured data, as a mechanism to create a single interrelated data model that layers the contextual understanding of why people are contacting the organisation into each individual customer journey, including all associated agent activity.
Only with this data strategy in place can AI be used to provide a granular understanding of the customer experience, identify friction, and highlight areas where transformation (digital or AI-led) could be most effective.
The second phase involves using AI to contextually analyse every customer interaction across every channel. This allows organisations to determine the volumetrics and costs of different enquiry types and to identify the root causes of high-effort or repeat contacts. But it also goes further, pinpointing where unproductive demand is being generated by upstream processes, and the analysis makes it possible to understand agent behaviour and performance in a real context, not just by handle time or Average Speed of Answer (ASA).
AI therefore isn’t just summarising the conversation; it’s interpreting the full journey, including queue times, sentiment, resolution, outcome, agent behaviour, and can even involve desktop application usage. This gives a truer picture of how performance (and customer experience) is being delivered.
Agents as Human-in-the-Loop
One of the most valuable yet often overlooked opportunities is to leverage contact centre agents as an ongoing human-in-the-loop (HITL) mechanism. Rather than viewing agents as legacy resources to be automated away, organisations should position them as active contributors to the continuous learning and refinement of AI systems. For example, agents can validate auto-summaries before they’re committed to Customer Relationship Management (CRM) platforms or refine or correct the categorisation of contact reasons or outcomes. They can help flag gaps or inaccuracies in chatbot and knowledge management content and provide feedback on AI-generated recommendations in real time.
This turns frontline staff into a dynamic quality assurance (QA) layer that strengthens the relevance, reliability, and usability of AI applications, from Agent Assist to FAQ bots to Retrieval Augmented Generation (RAG)-based knowledge retrieval. It also closes the loop between customer context, knowledge quality, and employee enablement, improving outcomes on all fronts.
Once these opportunities are clearly understood and your AI is fuelled by real-time, validated data, the next phase is to apply automation and augmentation where it delivers the most value. This could be to enhance digital self-service with better design and proactive guidance or to resolve low-complexity, repetitive contacts using Agentic AI. Or perhaps the contact centre can use AI Assistants to improve agent productivity and accuracy by surfacing contextual knowledge, suggesting next best actions, and auto-updating systems. Or the use might be more subtle, with AI used to continuously categorise, route, and summarise interactions using AI trained by and with human input.
This is where AI becomes additive rather than disruptive. It frees up agent time, reduces cognitive load, and improves consistency, while allowing the human workforce to focus on high-value, emotionally nuanced, or complex interactions.
Breaking Out of the Walled Gardens
Another reason many AI projects stumble is not because the technology isn’t sound but because organisations lack access to the validated data needed to power it. RAG, for example, promises smarter, context-rich responses but it fails without high-quality reference data and real-time context about the customer journey.
If your AI tools don’t know who the agent is, what system they’re using, or what the customer already tried online, then they’ll always be ineffective. To fix this, the industry needs to abandon the “walled garden” architectures of legacy vendors and embrace open, interoperable, plug-and-play platforms that enable this interrelated real-time data modelling. Only then will AI move beyond the prototype stage.
From Cost Centre to Value Engine
The digital transformation brought about by AI is about more than meeting technical goals; it’s an opportunity to rethink the entire CX cost model. By deflecting low-effort interactions through improved self-service and automation, companies can reduce their overall cost to serve while reinvesting in richer, more meaningful human interactions that build relationships, loyalty, and lifetime value.
In doing so, we move beyond cost-per-contact and ASA as dominant metrics, toward more customer-centric measures such as Customer Effort Score (CES), resolution quality, and customer sentiment. It’s here that AI not only reduces friction but empowers both agents and customers to engage in more valuable ways.
In summary, AI is not failing in CX because it lacks capability. It’s failing because it’s being deployed on shaky foundations that miss context, utilise fragmented data, and demonstrate a lack of integration. But with a phased, pragmatic approach, rooted in understanding, analysis, and then intelligent application it’s possible to reverse that trend.
Crucially, human contact centre agents are not being replaced by AI but remain integral to its success. They are the real-time validators, curators, and enablers of AI quality and customer relevance. With their assistance, and if we get this right, AI will not just reduce cost but will help us reinvest that value into human-led engagements that create lasting relationships and drive real business outcomes.
