Chatbots

Chatbots — questions and answers

What's the difference between intent and entity in chatbot development?

Intent represents what the user wants to accomplish—booking an appointment, checking order status, reporting a problem. Entity refers to specific data within that request: a date, product name, account number. A single user message like 'book a table for two on Friday' contains one intent (reservation) and two entities (party size, date). Training involves providing example phrases for each intent and marking which words represent entities.

How do chatbots handle conversations that go off-script?

Most chatbots define fallback handlers that trigger when confidence scores drop below a threshold or no intent matches. These might prompt the user to rephrase, offer menu options, or route to human support. Well-designed bots also track context across turns, so if someone asks a follow-up question without repeating details, the system remembers what was discussed previously. Handling tangents and recovering gracefully separates functional bots from frustrating ones.

Can a chatbot built for Facebook Messenger work on a website or WhatsApp?

The core logic—intents, entities, conversation flow—transfers across platforms, but each channel has different technical requirements and interface capabilities. Messenger supports quick reply buttons and carousels; WhatsApp has stricter message templates for business-initiated contact; web widgets control their own UI completely. Most frameworks provide channel adapters that translate between the bot's internal format and each platform's API, though you'll often adjust responses to suit each medium's constraints.

Why do some chatbots require training data while others work immediately?

Rule-based chatbots use pattern matching or decision trees defined by the developer, so they function as soon as the rules are written. Machine learning-based bots need example conversations to learn how people express intents, requiring dozens or hundreds of sample phrases per intent before accuracy becomes acceptable. Hybrid approaches use rules for structured tasks and ML for understanding varied natural language, balancing immediate functionality with adaptability.

What happens when a chatbot needs information from an external system during a conversation?

The bot makes an API call to the external system—a CRM, inventory database, payment processor—using data extracted from the conversation as parameters. This typically happens via webhooks: when a specific intent triggers, the chatbot platform sends a request to your server, which queries the external system and returns formatted data. The bot then incorporates that information into its response. Handling timeouts, errors, and authentication adds complexity beyond the conversation logic itself.

How is conversation state maintained when users return hours or days later?

Chatbot platforms store session data—collected entities, conversation history, current position in a flow—in databases keyed to user identifiers. Session timeout policies determine whether returning users resume where they left off or start fresh. For long-running processes like multi-step forms, you'll explicitly save progress and implement logic to detect abandoned sessions, remind users, or clear stale data. Stateful conversations require careful design around what persists and when context resets.