Why deploying AI in the customer journey is not a substitute for a competent, confident workforce.
There’s a seductive logic taking hold across financial services and beyond: why train people to a high standard when AI can simply supply the answer at the point of need? It sounds efficient. It sounds modern. It is also, I’d argue, a category error that firms will come to regret—commercially, legally and morally.
1. The state of UK customer service
Let’s start with the evidence, because the “AI will fix CX” narrative depends on customer service being broken enough to need fixing, and fixed by the thing being sold. The picture is more complicated than either story.
The Institute of Customer Service’s UK Customer Satisfaction Index has been tracking the nation’s mood since 2008. After a sustained slide—the index dropped to 75.8 in July 2024, its lowest level since 2010—there has been a modest recovery. By July 2026 the UKCSI stood at 78.3, up a point on the year but only 0.1 points higher than January 2026. That’s not a turnaround. That’s a plateau after a decade of underperformance, with satisfaction still trailing the highs of the early 2020s.
An employee who can retrieve an AI-generated answer but cannot explain, contextualise or exercise judgment over it is not meeting that bar—whatever the screen in front of them says.
What makes this sobering rather than merely disappointing is the backdrop against which it’s happening: an unprecedented, multi-year wave of investment in customer-facing technology. Chatbots, IVR, self-service portals and now generative AI have been rolled into contact centres at scale, all promising to “transform” the customer journey. And yet the dial has barely moved. If technology alone were the answer, we would expect to see it in the numbers. We don’t.
2. When it matters, people want people
The other inconvenient truth for the “AI will handle it” school of thought is that customers themselves don’t buy it—not when the stakes are real. Survey after survey tells the same story with remarkable consistency.
A SurveyMonkey study found 79% of consumers strongly prefer speaking with a human rather than an AI agent, and 84% believe human representatives give more accurate support. Separate research puts the figure even higher for anything consequential: 85% of people would rather speak to a real person than an AI when contacting a business, rising to 89% for law firms—professions, tellingly, where competence and trust are inseparable. Crucially, this isn’t blanket AI scepticism; it’s risk-weighted. Over two-thirds of consumers say they’d be uncomfortable using AI for medical advice (69%) or investment advice (68%), even while being perfectly happy to let a bot sort out a food order or a returns request. The pattern holds when researchers move from stated preference to scenario testing too: when consumers are asked about calling their bank for support, 81% want some level of human involvement in the call.
This is the point that gets lost in the “just deploy the system” logic: customers aren’t indifferent to how a problem gets solved. They are actively, and increasingly, distrustful of firms that lean on AI for anything that matters—and financial services, almost by definition, is where things matter. Trust in a business falls for 57% of consumers when it relies mainly on AI for customer service, and 70% believe service quality would get worse if humans were removed in favour of AI. You cannot underwrite that risk away with a better prompt.
3. The legal reality of competence in UK financial services
Here the argument stops being about customer sentiment and becomes about law. The FCA’s Training and Competence sourcebook, the Senior Managers and Certification Regime, and—since 2023—the Consumer Duty all rest on a shared premise: that the individual delivering advice or service to a customer must themselves be competent, not merely equipped with a tool that is.
The Consumer Duty in particular requires firms to ensure customers are given the support and understanding they need to make good decisions—and regulators have been explicit that outsourcing the substance of that support to a system does not discharge the firm’s obligation to have competent staff overseeing and standing behind it. Certification under SM&CR attaches personal accountability to individuals in customer-facing and advice roles. An employee who can retrieve an AI-generated answer but cannot explain, contextualise or exercise judgment over it is not meeting that bar—whatever the screen in front of them says. If something goes wrong, “the AI told them to say that” is not a regulatory defence, and it will not read well in a Final Notice. Firms that treat competence as something the technology now owns are building a compliance liability, not removing one.
4. Cognitive outsourcing isn’t new—it’s just better branded
None of this should feel unfamiliar, because contact centres have been here before. Call scripting arrived decades ago on precisely this promise: strip judgment out of the interaction, hand the agent a decision tree, and consistency—and cost control—would follow. Voice prompting and IVR systems did the same thing one layer up, routing customers through menus so that, in theory, fewer decisions needed to be made by anyone at all, human or otherwise.
What’s notable is not that these things happened, but that they didn’t work. Two decades of scripting and prompting have coexisted with the flatlining satisfaction scores above. If stripping cognition out of the front line were the route to better CX, we’d have got there already—repeatedly, and at enormous expense. Generative AI is a more sophisticated version of the same bet: that the answer to inconsistent human performance is to need less from the human. The technology has changed; the wager hasn’t, and neither has its track record.
5. What this does to the people doing the job
There’s a human dimension to this that rarely makes it into the business case. Ask anyone who has worked a scripted contact centre floor what it does to their sense of purpose, and you’ll hear some version of the same thing: it is deskilling, demoralising and it drives attrition. If the future role of an agent is to relay whatever the AI surfaces, without needing to understand why it’s right, you have not elevated the job—you have hollowed it out. You’ve turned a career into a mouthpiece.
Employees want more than a monitor to read from. They want mastery, some discretion and the dignity of knowing their judgment matters to the outcome. Strip that away and you get exactly what contact centres have struggled with for years: high churn, low engagement and a workforce that has quietly concluded the company doesn’t actually need them to be good at the job—only present for it. That’s a retention problem and a culture problem, long before it’s a regulatory one.
6. The law is starting to push back
If UK regulators haven’t yet legislated a “right to a human”, other jurisdictions are getting there. Spain’s Ley 10/2025—now in force—gives customers the right to request a human agent at any time, requiring bots and AI systems to provide a clear route to a person with full context transferred. The law goes further than sentiment: it requires 95% of calls to be answered within three minutes, on average, and applies not just to essential-service providers but to any company meeting certain size and turnover thresholds, including financial services.
This matters beyond Spain’s borders. It is a signal of direction. Regulators and legislators are increasingly treating “access to a competent human” not as a customer-service nicety but as a consumer right—which means firms betting their entire CX strategy on automation-first, human-optional models are building on ground that may not hold for long. It would be a brave UK strategy that assumes we will be the exception.
The allure of the system, and the discipline we’ve abandoned
Put all of this together and the appeal of “just deploy the technology” becomes obvious—and so does its hollowness. A system is measurable. You can point to a dashboard, a deflection rate, a cost-per-contact figure, and tell your board a story of progress. Improving people, by contrast, is slow, unglamorous and resistant to a single quarterly metric. Too many firms have quietly concluded it isn’t worth attempting—that competence is somebody else’s problem now that the AI is in the loop.
But the evidence above suggests this is precisely backwards. Customer satisfaction hasn’t moved despite decades of exactly this kind of substitution. Customers themselves reject it for anything that matters. The regulator expects genuine, individual competence, not a proxy for it. Employees experience it as an erosion of purpose. And the law, in at least one major economy, is starting to enshrine the opposite principle.
The better bet—commercially and ethically—is not choosing between people and AI, but using AI to make people demonstrably, continuously better at their jobs: monitoring competence, closing gaps and supporting wellbeing, rather than replacing judgment with retrieval. That’s the harder path. It’s also, I’d suggest, the only one that actually moves the dial. Firms like Elephants Don’t Forget exist precisely because that gap between “deploying a system” and “improving your people” is where the real, durable advantage sits—and it’s increasingly possible to close it without the trade-off firms have assumed they had to make.
