The changing conversation

AI and the human touch.

How automation can support better conversations—and where people need to stay involved.

A new role for AI in chat

AI can answer common questions, help visitors find information, summarize a conversation, or suggest responses to a human agent. These tools can be useful for straightforward requests and for giving agents context, especially when they are grounded in accurate, current information.

AI-assisted support is not automatically the same as live agent chat. A bot may respond instantly, but it can misunderstand a question, provide an outdated answer, or sound confident when it is wrong. A human should be able to take over when the issue is sensitive, complicated, or unresolved.

What the research suggests

One study of 5,172 customer-support agents found that access to a generative AI assistant increased issues resolved per hour by 15% on average, with different effects across workers. That result describes a particular workplace and tool; it should not be treated as a promise for every business. [1]

Design a useful handoff

  • Be clear: let visitors know when they are interacting with an automated assistant.
  • Offer a path to a person: explain when an agent is available and how to request one.
  • Carry context forward: with appropriate privacy safeguards, let the agent see the relevant conversation so the visitor does not need to repeat everything.
  • Check the answers: maintain the information the assistant uses and review errors and unresolved questions.
  • Protect information: understand what the provider stores, who can access it, and whether it is used to improve models.
A useful question to ask: does automation make it easier for this visitor to get the right answer, or does it add another step?

Measure the whole experience

Look beyond the number of conversations handled automatically. Review whether issues were resolved, whether people needed to contact you again, how easily they reached an agent, and what feedback they gave. A fast answer is only valuable if it is useful and trustworthy.

References

  1. Generative AI at Work, The Quarterly Journal of Economics.
  2. Protecting Personal Information: A Guide for Business, Federal Trade Commission.