Best Practices for Getting Started - Amazon Lex
Conversation Design PrinciplesReal-World Use Cases and Examples

Best Practices for Getting Started

Conversation Design Principles

Following these conversation design principles from the start will help you build more effective, maintainable, and user-friendly Amazon Lex V2 chatbots that provide natural interactions.

Core Design Principles

  • Start with User Goals - Design your bot around what users want to accomplish, not what your system can do. Focus on the user's journey and desired outcomes.

  • Use Natural Language - Write prompts and responses conversationally. Avoid technical jargon and speak as a helpful human would.

  • Provide Clear Options - When users get stuck, offer specific examples of what they can say rather than generic help text.

  • Keep It Simple - Start with basic functionality and gradually add complexity. Users should be able to complete common tasks quickly.

  • Handle Errors Gracefully - When the bot doesn't understand, provide helpful guidance rather than just saying "I don't understand."

  • Confirm Important Actions - Always confirm before taking actions that can't be easily undone, like placing orders or deleting information.

  • Provide Escape Routes - Always give users a way to start over, get help, or connect with a human when needed.

Conversation Flow Best Practices

  • Set Clear Expectations - Let users know what the bot can and cannot do early in the conversation.

  • Use Progressive Disclosure - Ask for information one piece at a time rather than overwhelming users with multiple questions.

  • Provide Context - Remind users what information you've already collected and what you still need.

  • Make Corrections Easy - Allow users to correct information without starting over completely.

Real-World Use Cases and Examples

These practical examples show how to apply conversation design principles to common scenarios that new Amazon Lex V2 users encounter.

Use Case 1: Appointment Booking

Scenario: A medical office wants to automate appointment scheduling.

Challenge: Users need to provide multiple pieces of information (service type, date, time, contact info) and may want to change details.

Solution Approach:

  • Start Broad: "What type of appointment would you like to schedule?" (dental, eye exam, consultation)

  • Narrow Down: "When would you prefer your dental appointment?" (Accept flexible input like "next week" or "Friday afternoon")

  • Confirm and Offer Changes: "I have you scheduled for a dental cleaning on Friday, March 15th at 2 PM. Does this work for you?"

  • Handle Changes: If user says "Can we do 3 PM instead?", update the time without restarting the entire process.

Key Techniques:

  • Use AMAZON.Date and AMAZON.Time for flexible date/time input

  • Create custom slot types for appointment types

  • Use confirmation prompts before finalizing bookings

Use Case 2: Order Status Inquiry

Scenario: An e-commerce company wants customers to check order status without calling support.

Challenge: Users might not have their order number handy, or might ask in different ways.

Solution Approach:

  • Multiple Entry Points: Accept "Where's my order?", "Track my package", or "Order status"

  • Flexible Identification: "I can help you track your order. Do you have your order number, or would you prefer to use your email address?"

  • Helpful Guidance: "Your order number is usually in your confirmation email and starts with 'ORD-'"

  • Clear Results: "Your order #ORD-12345 shipped yesterday and will arrive tomorrow by 8 PM. Would you like tracking details?"

Key Techniques:

  • Use built-in slot types like AMAZON.AlphaNumeric for order numbers

  • Provide multiple ways to identify orders (email, phone, order number)

  • Give clear, actionable information in responses

Use Case 3: FAQ and Support

Scenario: A software company wants to handle common support questions automatically.

Challenge: Users ask the same questions in many different ways, and some issues require human help.

Solution Approach:

  • Broad Intent Recognition: Train intents to recognize variations like "I can't log in", "Login problems", "Password not working"

  • Guided Troubleshooting: "Let's try a few quick steps. First, have you tried resetting your password?"

  • Escalation Path: "If these steps don't work, I can connect you with our support team. Would you like me to do that?"

  • Follow-up: "Did this solve your problem? You can always say 'help' if you need assistance with something else."

Key Techniques:

  • Use many sample utterances to capture different ways users express problems

  • Provide step-by-step guidance rather than long explanations

  • Always offer a path to human support when automated help isn't sufficient

Handling Complex Scenarios

Even simple bots encounter complex situations. Here's how to handle them gracefully:

  • When Users Change Their Mind:

    • User: "Actually, I want roses instead of tulips"

    • Bot: "No problem! I've updated your order to roses. Your roses will be ready for pickup on Friday at 2 PM. Is that correct?"

  • When Information is Unclear:

    • User: "I need flowers for next week"

    • Bot: "I'd be happy to help! Which day next week works best for you? You can say something like 'Monday' or 'next Tuesday'."

  • When the Bot Doesn't Understand:

    • User: "I want the purple ones with the long stems"

    • Bot: "I want to make sure I get this right. Are you looking for purple flowers? We have purple roses, purple tulips, and purple irises available."

  • When Users Need to Start Over:

    • User: "Cancel this, I want to start over"

    • Bot: "Of course! I've cleared your current order. What type of flowers would you like to order today?"