Adrian Chiles needed help with a broken EV charger. What he got instead was “Rachel” — an AI voice assistant dressed up in fake office sounds, pretending to be somewhere it wasn’t. The call failed three times. A real person, Stuart, eventually solved the problem whilst chatting about nothing in particular. This is not just a funny column about technology frustration. It is a warning sign that many UK businesses are choosing to ignore.
The customer service gap nobody measures
Chiles’s experience, published in The Guardian, follows a pattern that customer experience teams should find uncomfortable. The AI assistant “Rachel” offered help in a polished American accent with carefully curated background noise — keyboards clacking, ambient office chatter. It was designed to deceive.
And it failed. Not once, but three times. Each time, the line went dead after the handoff.
Strategic Reality: Most organisations track AI deflection rates and cost-per-contact. Almost none measure the reputational damage when automated systems fail customers who genuinely need help.
The numbers tell one story. A 2025 Zendesk report found that 72% of UK consumers have abandoned a purchase after a poor automated service experience. Gartner projects that by 2027, 80% of customer service organisations will use generative AI in some form. The gap between those two data points is where businesses are bleeding value without realising it.
| Metric | What gets measured | What gets missed |
|---|---|---|
| Resolution rate | Tickets closed by AI | Customers who gave up before resolution |
| Cost per contact | Savings vs human agents | Revenue lost from frustrated customers |
| CSAT scores | Post-interaction surveys | Customers who never complete the survey |
| Deflection rate | Calls avoided | Trust eroded per deflected call |
What Chiles’s story actually reveals
The column reads as a personal anecdote. Underneath, it exposes three systemic failures in how businesses deploy AI for customer-facing work.
The deception problem. The fake background noise is not a minor design choice. It represents a deliberate attempt to make an AI system pass as human. When customers discover the deception — and they always do — trust damage compounds. Chiles described it as “chilling” and “an attempted act of deception.” His reaction is not unusual.
Critical Context: The EU AI Act classifies AI systems that “manipulate persons through subliminal techniques” as high-risk. Fake environmental audio designed to simulate a human call centre sits uncomfortably close to that boundary. UK businesses operating in European markets should take note.
The handoff failure. Rachel could acknowledge the problem and promise to connect Chiles with someone who could help. Then the line went dead. This is the most common failure mode in AI customer service: the system handles the easy part (greeting, routing) but collapses at the point where the customer actually needs something done. Three consecutive failures suggest a systemic issue, not a one-off glitch.
The hidden phone number. Before reaching Rachel, Chiles had to dig through a glossy website full of sustainability messaging to find a support number “hidden in the base rock” of a page. This is design by discouragement — making it deliberately difficult to reach any form of support, human or otherwise.
Hidden Cost: Every customer who cannot find your support contact is a customer whose next interaction with your brand may be a negative review or a cancelled subscription. The cost of a buried phone number is invisible but cumulative.
Who loses when AI customer service gets it wrong
The impact extends well beyond one frustrated columnist.
Customers in genuine need. Chiles needed to charge his car to drive to work. This was not a casual enquiry. When AI systems fail people with urgent, practical problems, the emotional impact is disproportionate. A 2025 UK Institute of Customer Service study found that 68% of consumers rate “being able to speak to a person when needed” as their top service priority.
Frontline staff who pick up the pieces. Stuart, the installer, eventually called back and spent a long time walking Chiles through complex IT configuration. The human agent inherits the customer’s accumulated frustration from every failed automated interaction that came before.
Implementation Note: Organisations that route complex technical issues through AI first, then to humans, are creating a worse experience than direct human contact. The AI interaction sets expectations that go unmet, and the human agent starts from a deficit.
Brand perception over time. Chiles writes for The Guardian. His column reaches millions. Most dissatisfied customers do not have a national newspaper column, but they do have social media, review platforms, and word of mouth. The reputational arithmetic works against companies that get this wrong.
| Stakeholder | Immediate impact | Long-term consequence |
|---|---|---|
| Customer | Frustration, wasted time | Brand abandonment, negative advocacy |
| Support staff | Higher complaint volume | Burnout, attrition |
| Brand | Single poor experience | Cumulative trust erosion |
| Competitors | None | Differentiation opportunity |
What good AI-assisted customer service looks like
The answer is not to remove AI from customer service entirely. It is to deploy it honestly and with clear escalation paths that actually work.
Success Factor: The best AI customer service implementations share one trait: they never pretend to be human. They set clear expectations, solve simple problems well, and hand complex issues to people without dropping the connection.
Tier your support honestly. Use AI for genuinely routine queries — password resets, order tracking, FAQ lookups. For technical problems, complex complaints, or emotionally charged situations, route directly to human agents. Do not force customers through an AI interaction they do not want.
Fix the handoff. The most dangerous moment in AI customer service is the transition from automated to human. If your system cannot guarantee a warm handoff — where the human agent receives full context and the customer stays on the line — you have a system that is not ready for production.
Drop the theatre. Fake background noise, synthetic vocal warmth, deliberately ambiguous identities — these are trust liabilities, not engagement features. The UK’s Online Safety Act and emerging AI transparency requirements both point toward mandatory disclosure of AI interactions. Getting ahead of this is straightforward and costs nothing.
SME Advantage: Smaller businesses can turn the human element into a genuine competitive advantage. When a large competitor routes every call through an AI system, being the company where a real person answers the phone becomes a differentiator worth marketing.
Four risks hiding in your AI customer service deployment
1. Compliance exposure from synthetic identity. Using AI voices that mimic human characteristics without disclosure creates regulatory risk. The UK government’s AI white paper emphasises transparency as a core principle. Businesses that blur the line between human and AI interaction may face enforcement action as regulation matures.
2. Data loss from silent churn. Customers who abandon AI interactions rarely file complaints. They simply leave. Without tracking abandonment patterns at each stage of the AI interaction, organisations cannot see the damage. Most analytics platforms are not configured to capture this.
3. Agent morale collapse. Human agents who only receive escalated, already-frustrated customers experience higher stress and faster burnout. If AI handles all the “easy” interactions, the human role becomes exclusively adversarial. Retention of experienced support staff becomes harder.
Warning: ⚠️ If your customer service AI has a higher handoff failure rate than 5%, you are actively harming customer relationships at scale. Audit your handoff completion rates before expanding AI deployment.
4. The uncanny valley of trust. Chiles compared Rachel’s voice to deepfake videos — technically impressive but fundamentally unsettling. There is a growing body of research suggesting that AI systems designed to closely mimic humans provoke stronger negative reactions when they fail than systems that are transparently artificial. The more realistic the imitation, the deeper the disappointment.
Getting this right: three things that matter
The organisations succeeding with AI in customer service share a practical philosophy. They use AI where it genuinely improves speed and accuracy, and they protect human interaction where it genuinely matters.
Transparency builds more trust than sophistication. A message that says “You’re speaking with an AI assistant. I can help with [X, Y, Z]. For anything else, I’ll connect you with a person” outperforms any amount of synthetic warmth. Customers who know they are dealing with AI adjust their expectations accordingly and report higher satisfaction when the AI performs within its stated capabilities.
Measure what customers experience, not what the system reports. Deflection rates and resolution percentages are operational metrics. They tell you what the system did. They do not tell you how the customer felt. Add post-interaction surveys that specifically ask: “Were you able to get the help you needed?” and “At any point, did you want to speak with a person but couldn’t?”
Take Action: Audit your AI customer service touchpoints this quarter. Map every interaction where a customer might need human help and verify that the escalation path works — not in testing, but in production with real customers under real conditions.
Invest in the handoff, not the chatbot. The technology for AI greeting, routing, and simple query resolution is mature and works well. The technology for seamless human handoff is where most implementations fall apart. Allocate budget and engineering time to the transition, not just the front end.
Where this leaves UK businesses
Chiles ended his column grateful not for the technical fix, but for “the interaction with a living being.” He called human customer service agents “a dying breed.” That phrase should concern every business leader reading this.
The companies that will win customer loyalty over the next five years are not necessarily those with the most advanced AI. They are the ones that understand when AI adds value and when it subtracts it. They deploy automation for efficiency and protect human interaction for trust.
A phone number that is easy to find. An AI that says what it is. A handoff that does not drop the call. These are not expensive innovations. They are basic operational standards that an alarming number of organisations fail to meet.
The question is not whether to use AI in customer service. It is whether you are honest about what it can and cannot do — with your customers and with yourself.
This analysis is based on Adrian Chiles’s column “The AI assistant was offering me any help I needed. All I wanted was a living, breathing human” published in The Guardian on 11 March 2026.