What is a chatbot?
A chatbot is a program that talks with people in text. In customer service, it usually appears as a chat window on a website, a messaging channel such as WhatsApp or Facebook Messenger, or inside an app. The customer types a question and the chatbot replies, either answering directly, collecting details, completing a simple task, or passing the conversation to a human agent.
Chatbots are one form of conversational AI, though not every chatbot is especially intelligent. The phone equivalent is an AI voice assistant.
Types of chatbot
Rule-based chatbots
These follow a fixed decision tree. The customer clicks buttons or types keywords, and the chatbot follows a scripted path. They are predictable and easy to control, but they break as soon as a customer asks something outside the script.
AI chatbots
These use natural language understanding and intent recognition to interpret free-text questions. Many now use generative AI to compose replies, often drawing on a knowledge base. They cope with far more variety, but because they generate answers rather than pick from a list, they need guardrails to keep them accurate and on topic.
Hybrid chatbots
Most production chatbots combine both: AI to understand the question, and controlled flows for anything that must be done exactly right, such as identity checks or payments.
Where chatbots help
- Simple, frequent questions like opening hours, how to update details or where to find a form.
- Gathering information before a human picks up, so the agent doesn’t need to ask again.
- Out-of-hours cover in webchat and messaging channels.
- Status checks when connected to a back-end system, such as an order, application or outage.
Text has some natural advantages over voice. Customers can multitask, scroll back, and paste in reference numbers, and the conversation leaves a written record. That makes chatbots a common tool for channel shift away from the phone.
Where chatbots go wrong
Chatbots have a mixed reputation, mostly because of poor deployments rather than the technology:
- Looping. The bot repeats “Sorry, I didn’t understand” without offering a way out.
- Hiding the human. Customers who can’t reach a person usually switch to the phone, angrier, adding to the workload rather than reducing it.
- Overreach. Bots that try to handle complaints, hardship or sensitive situations tend to make them worse.
- Wrong answers. A generative chatbot without grounding can state incorrect policy with confidence.
- Counting the wrong thing. A high containment rate means little if customers gave up rather than got an answer.
Good practice
- Tell customers they are chatting with an automated assistant.
- Keep the scope narrow at first and widen it as you learn from real transcripts.
- Make the handover to a person obvious, quick and complete, passing the full chat history so the customer doesn’t repeat themselves.
- Staff the human side of the channel properly. A chatbot is only as good as what happens when it reaches its limits.
- Measure resolution and customer effort, not just how many chats the bot closed.