Customer service is one of the business functions most affected by generative AI. Contact centers handle huge volumes of repetitive questions in text and voice — exactly the type of work language models can assist with or automate. The evidence so far shows real productivity gains, along with new risks for customers and companies.

The evolution of chatbots

Early chatbots followed scripts and decision trees, often frustrating customers who couldn’t find the right option. Generative AI chatbots understand free-form questions, answer from a company’s knowledge base and handle follow-up questions naturally. The latest systems are AI agents that can take actions — checking an order, processing a refund, changing a booking — through connections to business systems.

Agent assist

Much of the value comes from helping human agents rather than replacing them:

  • suggesting answers in real time during chats and calls;
  • surfacing relevant knowledge base articles;
  • summarizing conversations and writing case notes automatically;
  • translating between languages;
  • coaching agents on tone and compliance.

What the evidence shows

A prominent study, Generative AI at Work by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, examined more than 5,000 customer support agents at a company that introduced an AI assistant. Productivity, measured by issues resolved per hour, rose about 14% on average — and about 34% for novice and less-skilled workers, with little effect for the most experienced. Customers’ sentiment toward agents improved, and employee attrition fell. The AI appeared to spread the knowledge of top performers to newer staff.

Companies have also reported large automation figures. Payment company Klarna said in 2024 that its AI assistant handled two-thirds of customer service chats in its first month. It later emphasized that customers should always be able to reach a human — a reminder that automation rates alone don’t capture customer experience.

The risks

Wrong answers. Chatbots can give incorrect information about policies, prices or rights. In Moffatt v. Air Canada (2024), a Canadian tribunal held the airline responsible for incorrect bereavement-fare information given by its chatbot.

Frustration. Customers dislike being trapped with a bot that can’t solve their problem. Easy escalation to humans is essential.

Security. Agents connected to account systems must verify identity and resist manipulation, including prompt injection and social engineering.

Privacy. Conversations contain personal data and must be handled under data protection rules.

Job impact. Automation may reduce the number of entry-level roles, which have traditionally been a gateway into many careers.

Measuring success

Good metrics go beyond cost per contact:

  • resolution rate — was the problem actually solved?
  • customer satisfaction and effort scores;
  • repeat contacts about the same issue;
  • escalation rates and reasons;
  • accuracy audits of AI answers;
  • employee satisfaction and attrition.

A typical modern support flow

  1. A customer writes in the chat: “My order hasn’t arrived.”
  2. The AI assistant verifies the customer’s identity through their logged-in account, looks up the order and sees it is delayed at a depot.
  3. It explains the situation, gives a new estimated delivery date and offers options allowed by policy, such as waiting or cancelling for a refund.
  4. If the customer is unhappy or the case falls outside the rules, the assistant hands over to a human agent with a summary.
  5. The human agent sees suggested replies and relevant policies, resolves the issue and the AI writes the case notes.
  6. Conversations are sampled for quality review, and recurring problems — like a depot with frequent delays — are flagged to operations.

The best implementations treat AI as part of a system that includes people, policies and continuous improvement, not as a standalone bot.

The future of the contact-center job

As routine questions are automated, human agents handle a higher share of complex, emotional or unusual cases. That makes the job more demanding and potentially more skilled, with greater emphasis on empathy, problem-solving and judgment. Training, career paths and workload management will matter as much as the technology.

What to watch next

  • AI agents handling complex multi-step requests end to end.
  • Voice AI for phone support becoming indistinguishable from humans.
  • Rules requiring access to a human agent.
  • Effects on entry-level contact-center employment.

Key takeaways

  • Generative AI chatbots answer free-form questions and increasingly take actions.
  • Agent assist tools suggest answers, summarize calls and translate.
  • A large study found 14% average productivity gains and 34% for novices.
  • Companies are responsible for what their chatbots tell customers.
  • Easy escalation to humans and accuracy audits are essential.

Frequently asked questions

Can I always ask for a human?

Not always by law, but many companies offer it, and some jurisdictions are considering requirements. Phrases like “speak to an agent” often trigger escalation.

Are AI customer service agents accurate?

They can be accurate when grounded in up-to-date company knowledge and limited to clear tasks, but they still make mistakes that need monitoring.

Does AI customer service save money?

Often yes through automation and faster handling, but poor implementations can increase repeat contacts and harm loyalty.