Zoom in on CX: Why Most AI in CX Fails Before It Even Launches

Every year, more organizations pour budget into AI tools promising to transform customer experience — and every year, many of those same organizations quietly struggle to show results. The technology works. The rollout happens. But the transformation never quite materializes the way leadership hoped. Why? According to Andy Kingston, Head of Consumer Channels Transformation at Lloyds Banking Group, the answer usually has nothing to do with the AI itself. In a recent episode of Zoom in on CX, Kingston unpacked what separates real CX transformation from expensive tech theater — and it starts long before anyone touches a platform.

The Problem With "Hope-Driven" Transformation

Kingston is direct about the pattern he sees across enterprise organizations: teams get seduced by the technology before they've done the harder work of defining the problem. "There's probably three core elements," he explains. Architects and technologists come in early to map what's technically possible — which is useful, but incomplete. Too often, teams skip a crucial step before that: getting brutally clear on what they're actually trying to solve, whether it's long handling times, low automation containment, or simply customers who aren't satisfied. Skipping that step is what Kingston calls "the magpie effect" — getting "obsessed by the shiny technology side" before knowing what problem it's meant to fix.

The third element, he argues, is the one leaders resist most: bringing frontline colleagues in early, not as a rubber stamp at the end. "We don't always know best," Kingston admits. Colleagues serving customers every day will tell you the truth about whether a proposed solution actually helps — if you ask them before the investment is already locked in.

Personalization Without Crossing the Line Into Creepy

A recurring theme in modern customer experience strategy is personalization — but Kingston is cautious about chasing hyper-personalization for its own sake. "I'm not sure you need to go to that depth," he says, referring to the push toward "segment of one" targeting. Instead, he sees more near-term value in giving human colleagues strong, real-time data and letting them lead the conversation. "It's the conversation where the value add really comes," Kingston notes — the technology should support that conversation, not replace it prematurely.

The deciding factor for any AI-driven personalization, he argues, is transparency: customers need to understand why an organization is engaging with them. Skip that step, and even well-intentioned personalization starts to feel invasive rather than helpful.

Listen to all episodes

Click here

Where AI Belongs — and Where Human Judgment Still Wins

Fraud detection is one of the clearest examples of AI's real value in a contact centre setting, according to Kingston. AI models can process the dozens of data points a colleague would otherwise need to manually cross-reference, building a propensity score for whether an interaction looks fraudulent. But even in highly automatable scenarios, Kingston points to a simple truth: "We still go and see a GP" even though a search engine could answer most basic medical questions. The reassurance of human judgment matters — especially in AI in customer service scenarios involving financial risk, vulnerability, or high emotion.

Rethinking the Metrics: Experience Over Cost-to-Serve

Perhaps the sharpest point in the conversation is Kingston's rejection of cost-to-serve as the primary metric for CX success. "I don't think cost to serve is the primary currency anymore," he states plainly. Organizations focused purely on cost reduction are, in his view, missing where the real competitive advantage now lives: experience. Getting the human-AI blend right doesn't just cut costs — it drives customers to choose you, and buy more from you, because the experience itself becomes the differentiator.

Key Takeaways

• Define the problem before evaluating the technology — don't let the tools dictate your CX strategy
• Involve frontline colleagues early, not as a final validation step
• Personalization should add clear, transparent value — not just maximize data usage
• AI works best supporting human judgment in fraud, risk, and vulnerability cases, not replacing it outright
• Cost-to-serve is no longer the right lens for measuring AI success in CX
• The colleague's role is evolving toward complex, high-emotion interactions as AI absorbs routine tasks

Want the full conversation? Listen to this episode of Zoom in on CX on Spotify or Apple Podcasts. To listen to all our Zoom in on CX episodes early, visit our dedicated Zoom in on CX landing page.

Disclaimer

Zoom in on CX is produced by Zoom. Guests appear in an independent capacity and the views expressed are their own, not necessarily those of Zoom. Any research, statistics, or third-party findings referenced in this post are attributed to their respective sources and have not been independently verified by Zoom. Where research has been conducted with the support of Zoom, this is noted in the relevant episode. This content is intended for informational purposes only and does not constitute legal, regulatory, or professional advice.