Chatbot

Short answer: Customers or staff usually reach a chatbot through a website, a messaging app or an internal help desk. There are two broad types. Rule-based chatbots follow scripted paths, like a fixed menu of questions and answers, or match keywords to prewritten replies. AI-based chatbots usually use a large language model to understand questions in everyday wording and write their own replies. Many business chatbots combine both.

Typical uses include answering frequent customer questions, giving order or delivery status and booking appointments. Chatbots also collect details before passing a case to a person, and answer staff questions about internal policies. Rule-based chatbots are predictable and easy to check. But they handle only the paths someone scripted, and they frustrate users who ask anything else. AI-based chatbots handle varied wording and can answer from company documents, but they can also give wrong answers. In both cases, much of the value depends on access to company systems, such as the order system for status questions. It also depends on a clear route to a person.

When a user sends a message, the chatbot works out what the person is asking. It either matches a scripted rule or passes the message to a language model. It then looks up the information it needs. That can be an approved answer, a passage from the company's help articles or a record from a business system through a direct connection. Next, it replies, asks a clarifying question or hands the conversation to a person. The team sets handover rules for when the chatbot is unsure, when the user asks for a person or when the topic is sensitive. Sensitive topics include complaints, refunds and legal matters. The team also decides which topics the chatbot may cover, what data it can see and how it logs conversations for review.

An online retailer handles 6,000 support chats a month, and 40% of them, or 2,400, are order status questions. It connects a chatbot to the order system to answer those questions. In a test, the chatbot resolves 75% of order status chats without a person, which is 1,800 chats a month. If each of those chats took a support staff member 6 minutes, the team saves 10,800 minutes, or 180 hours, a month. The team also tracks repeat contacts to confirm the chatbot actually solved the issues.

A chatbot speaks for the company, so leaders need to decide its scope. That means which questions it answers, what it may promise, when it hands over and which systems it can read. Measure success by issues resolved, customer satisfaction and repeat contacts. Counting the conversations the chatbot handled can overstate its value, because customers may then call or email about the same issue.

AI-based chatbots can state wrong prices, policies or refund terms. A company may have to stand by what its chatbot tells customers. Users can sometimes steer a chatbot into off-topic or inappropriate replies. Conversation logs often contain personal data, which brings privacy duties and rules on how long to keep it. In some countries, businesses must tell users they are talking to an automated system. Customers should always have a way to reach a person.