Chatbots and Customer Service: What AI Is Actually Changing

British consumers sent over 2.3 billion messages to AI customer service systems in 2024, according to Salesforce research. Most of those interactions resolved without a human agent. Some left customers more frustrated than if they had never contacted the company. The quality gap between the best and worst AI customer service deployments is enormous, and understanding what drives it matters whether you are a business deploying these systems or a consumer trying to get something resolved.

What Modern AI Chatbots Actually Are

The customer service chatbots of 2026 are fundamentally different from the decision-tree bots of 2018. Modern systems are built on large language models — the same technology underlying ChatGPT and Claude — trained on a company’s specific knowledge base, product documentation, and customer interaction history. They understand natural language, handle ambiguous phrasing, and can maintain conversational context across a multi-turn interaction.

The result is dramatically better than their predecessors for routine queries — account balance, order tracking, basic troubleshooting, FAQ responses. The result is still unreliable for complex, emotionally charged, or edge-case scenarios that require genuine judgment or authority to resolve.

Where AI Customer Service Works

High-volume, high-repetition interactions are where AI chatbots deliver clear value. HSBC deploys an AI virtual assistant called Amy that handles over 60% of UK customer queries without human escalation — primarily account queries, payment confirmation, and basic fraud reporting. The deflection rate has reduced the cost per customer interaction by an estimated 40%, according to HSBC’s 2024 annual report.

Retail and e-commerce see some of the strongest deployments. Marks & Spencer’s AI assistant handles order tracking, returns initiation, and delivery issue resolution with resolution rates above 80% for those query types. Customers who contacted the assistant at 2am about a delivery issue got an immediate resolution — something a human agent team could not have provided at that hour.

Insurance claim first notification — capturing the basic facts of a claim — is another area where AI has proven reliable. Aviva UK and Direct Line both use AI for initial claim intake, reducing average first notification time by over 60% compared to phone-only channels.

Where AI Customer Service Fails

Complaint escalation is where many deployments fall apart. When a customer is already frustrated and contacts a business to complain, an AI that responds with scripted empathy phrases and cannot make exceptions to policy makes things worse. The research is clear: customers who experienced AI-handled complaints that were not resolved rate their satisfaction lower than customers who waited longer for a human agent.

Complex, multi-step problems — a billing dispute that requires accessing three different systems, cross-referencing historical transactions, and making a judgment call about whether to credit an account — exceed the practical capability of current AI systems deployed in most contact centres. Recognising these handoff points and escalating gracefully to human agents determines whether a chatbot deployment improves or damages customer satisfaction.

The UK Regulatory Picture

The FCA’s Consumer Duty requirements, which came into full force in 2023, created new obligations for financial services firms using AI in customer interactions. Firms must ensure AI-driven customer service does not lead to poor outcomes — meaning the AI cannot make the customer journey so difficult that they give up on making a claim, disputing a charge, or switching provider.

The ICO published guidance on AI in customer service in 2024, specifying that customers must be informed when they are interacting with an AI system rather than a human, must be able to request a human agent, and must not be subject to automated decision-making that significantly affects them without human review available. Several UK firms are not fully compliant with these requirements yet.

What Good Deployment Looks Like

The best AI customer service deployments share four features. They are transparent — customers know they are talking to AI. They have clear, accessible escalation paths to human agents. They are trained on current, accurate product and policy information. And they are monitored for customer satisfaction and continuously improved based on interaction data.

Companies that deployed chatbots primarily to cut costs, without investing in training quality, escalation design, or monitoring, consistently see worse customer satisfaction scores than companies that approached the deployment as a genuine service improvement.

What This Means for You

As a consumer, AI customer service has improved significantly and will continue to. For routine queries — especially outside business hours — it is often faster and more effective than phone queues. For complex complaints or situations requiring human judgment, push for human escalation early rather than cycling through AI responses that cannot resolve your issue. Saying “I need to speak to a human agent” should work on any FCA-authorised firm’s service and is your right under current ICO guidance.

This article is for educational purposes only and does not constitute financial advice. Cryptocurrency investments involve significant risk. Always do your own research.