Pre-configured Responses, Sequences, and Cached Responses Guide

Overview

ChatIQ uses three systems to provide fast, cost-effective responses:

• Pre-configured Responses
• Response Sequences
• Response Caching

Understanding how these work will help you optimize your chatbot's performance and reduce costs.

What Are Pre-configured Responses?

Pre-configured Responses are instant, zero-cost replies that you manually configure for specific patterns or messages. They are checked before any AI processing happens, making them perfect for common queries like greetings, FAQs, and simple interactions.

Key Features

• ✅ Instant responses (<5ms, no API calls)
• ✅ Zero cost (no OpenAI API usage)
• ✅ Always available (work even when evaluation ends)
• ✅ Per-bot libraries (organize with response sets)
• ✅ Flexible editing (basic intent view or advanced pattern view)
• ✅ Action triggers (booking, privacy, takeover, sequences)

How They Work

1. User sends a message
2. System checks if message matches any pre-configured response pattern
3. If match found → Returns the pre-configured response immediately and may start an optional action
4. If no match → Continues to cache check, then LLM

When to Use Pre-configured Responses

Perfect for:

• Greetings: "Hi", "Hello", "Hey"
• Common questions: "What are your hours?", "How do I contact support?"
• Simple acknowledgments: "Thanks", "Thank you", "OK"
• Frequently asked questions with fixed answers
• Error messages or system status updates

Not ideal for:

• Complex questions requiring AI reasoning
• Questions that need document retrieval
• Dynamic responses that change based on context

Setting Up Pre-configured Responses

1. Go to Dashboard → Pre-configured responses
2. Choose a bot to manage its response library
3. Use Response sets:
   • New response set from source to generate responses from a FAQ, menu, catalog, or gallery
   • New response set (empty) to build a manual set
4. Add responses manually with Add pre-configured response
5. Use the Pattern helper (AI) for quick pattern suggestions when needed
6. Use Basic or Advanced mode per response:
   • Basic: Write a plain-language Query/Intent + Response and we generate the pattern
   • Advanced: Edit the exact Pattern and Pattern Type directly
7. If needed, choose an Action such as Start sequence

Basic vs Advanced Mode

• Basic mode is best for non-technical users: enter a natural question and a response, and we generate the matching pattern automatically.
• Advanced mode is for precision: edit pattern + type directly for edge cases or highly specific matches.

Response Sets from Sources

Use response sources (FAQ, menu, catalog, gallery, services, custom) to generate pre-configured responses automatically. Each source creates a response set you can filter, edit, and extend just like manual responses.

How it works:

• Create a response set from a source document
• The system generates patterns + responses per item
• You can edit any response without breaking the source

Manual Suggestions (Quick Replies)

You can also add manual suggestions to any pre-configured response.

How it works:

• Each suggestion has a label (button text) and payload (sent as user text).
• When clicked, the payload is treated like a new user message.
• If the payload matches another pre-configured response, it fires immediately.

Inline Button Shortcuts for LINE

For LINE-focused responses, you can also define buttons directly inside the response text using markdown-style shortcuts on their own lines.

Supported formats:

What would you like to do next?

[Demo](https://example.com/demo)
[Survey](reply:survey)

How it works on LINE:

• https://... or http://... creates a URL button
• reply:survey creates a reply button that sends survey back into chat
• Reply payloads can trigger another pre-configured response immediately
• These shortcut lines are removed from the visible LINE message body and shown as buttons instead

Per-response LINE rendering:

• Quick replies: render the buttons as standard LINE quick replies
• Flex buttons: render the same buttons inside a simple LINE Flex bubble

Use the LINE button style setting in the pre-configured response editor to choose between quick replies and Flex for that response.

Important note:

• The reply: shortcut is currently interpreted by the LINE delivery runtime only.
• For web/custom clients, use structured suggestions[] if you need button rendering outside LINE.

Auto-Generated Image Responses (Item Sources)

For item-based documents, ChatIQ can auto-generate image-only pre-configured responses per item. These trigger on image intent (e.g., “show me a picture of pad thai”) and return images via attachments[].

Notes:

• The response text may be short or empty.
• alt/caption can be empty; clients should not rely on them for labels.
• If attachments[] is present, render images even when response is empty.

Pattern Types

Exact Match

• Matches the exact message (case-insensitive by default)
• Example: Pattern "hello" matches "Hello" but not "Hello there"

Keyword Match

• Matches if the pattern appears anywhere in the message
• Example: Pattern "hours" matches "What are your hours?" and "business hours"

Regex Match

• Full regular expression support
• Example: Pattern "hi|hello|hey" matches any of those words
• Example: Pattern "^(hi|hello)" matches messages starting with "hi" or "hello"

Multilingual Pattern Setup

Some languages (Thai, Chinese, Japanese, Korean) do not use whitespace to separate words. For these, use regex patterns instead of keyword.

English (space-delimited)

Preferred: keyword

• Pattern: book|booking|appointment|schedule
• Type: keyword

Thai (no whitespace)

Preferred: regex

(จอง|นัด|นัดหมาย|จองคิว|จองนัด|จองเวลา|จองวัน|ขอคิว|ขอจอง|จองตาราง|จองโต๊ะ|นัดคิว)

Chinese (no whitespace)

Preferred: regex

(预约|预订|预定|订位|订座|挂号|安排时间)

Japanese (no whitespace)

Preferred: regex

(予約|予約したい|予約できますか|予約希望|予約お願い|面談予約)

Korean (limited whitespace)

Preferred: regex

(예약|예약하고|예약하고 싶|예약 가능|예약 문의|방문 예약|진료 예약)

Special System Patterns

These special patterns take precedence over regular pre-configured responses and the default response:

system_unavailable

• Shown when the AI service is temporarily unavailable or the evaluation has ended
• Use pattern type "exact" for these special patterns
• Takes precedence over the bot's default response field

quota_exceeded

• Shown to public users when your quota is exceeded
• Use pattern type "exact" for these special patterns
• Takes precedence over the bot's default response field

Actions

Pre-configured responses can trigger actions:

• Request human → notify staff and mark the conversation as needing help
• Enable human takeover → silence the bot for a set period
• Disable human takeover → return control to the bot
• Start booking flow → begin a booking workflow
• Start sequence → begin a timed multi-step response sequence
• Privacy opt-in → enable saving for the current session
• Privacy opt-out → disable saving for the current session

Privacy actions update the session state only. They do not change the bot’s default retention setting.

For booking-specific behavior (chat-first flow, webview-assisted steps, resume/re-entry), see Booking Flows.

What Are Response Sequences?

Response Sequences are reusable timed message flows that start from a pre-configured response action.

They are useful when one instant reply is not enough and you want a short guided sequence such as:

1. a greeting
2. a short introduction
3. an image
4. a final prompt or call to action

Sequence Step Capabilities

Each step can include:

• text
• image
• delay before send
• optional links or reply buttons

How Sequences Work

1. User sends a message
2. A pre-configured response matches
3. That response uses the Start sequence action
4. ChatIQ starts the sequence run for that conversation
5. Steps are delivered over time until the sequence completes or the user interrupts it

Interruption Behavior

Sequences are designed to stop if the user replies mid-sequence.

If the user sends any new message while a sequence is active:

• the current sequence run stops
• it does not automatically resume
• the user can trigger it again later if you provide a matching trigger

Setting Up a Sequence

1. Go to Dashboard → Pre-configured responses → Sequences
2. Create a sequence in the shared sequence library
3. Add one or more steps with text, image, and delay
4. Go back to your pre-configured response
5. Set the action to Start sequence
6. Select the sequence you want that response to launch

Buttons and Links in Sequences

Sequence steps support the same markdown-style shortcut syntax used elsewhere:

[View pricing](https://example.com/pricing)
[Continue](reply:continue_demo)

Channel behavior differs:

• LINE supports richer button rendering
• web and other channels may render the content more simply

Use plain text plus one or two clear actions when you want the broadest cross-channel compatibility.

Useful Patterns

Here are some common patterns you might want to use:

• Greetings: Hi|Hello|Hey|Greetings (regex)
• Thanks: thanks|thank you|ty (regex)
• Hours: hours|business hours|open (keyword)
• Contact: contact|email|phone|support (keyword)

What Is Response Caching?

Response Caching automatically stores AI-generated responses and reuses them for similar questions. When a user asks a question that's similar to one asked before, the system returns the cached response instead of calling the LLM again.

Key Features

• ✅ Automatic (no configuration needed)
• ✅ Cost savings (reduces OpenAI API calls)
• ✅ Faster responses (instant for cached queries)
• ✅ Similarity matching (uses AI embeddings to find similar questions)
• ✅ Smart invalidation (clears when bot settings change)

How It Works

1. User sends a message
2. System generates an embedding (vector representation) of the message
3. System searches cache for similar messages (using vector similarity)
4. If similar message found (similarity > 85%) → Returns cached response
5. If no cache hit → Calls LLM, then saves response to cache

Cache Invalidation

The cache is automatically cleared when:

• Bot's system prompt changes
• Bot is deleted
• Cache entry expires (30 days TTL)
• Manual cache clear (via "Clear Cache" button)

When Caching Helps Most

High cache hit rates occur when:

• Multiple users ask similar questions
• Same user asks variations of the same question
• Common questions are asked repeatedly
• Bot documentation is stable (not frequently updated)

Lower cache hit rates when:

• Questions are highly varied
• Bot documentation changes frequently
• System prompt is updated often

The Difference: Pre-configured vs Cached

Response Flow

Here's the complete flow when a user sends a message:

User Message
  ↓
1. Check Pre-configured Responses
   → Match? → Return response ✅ (instant, free)
   → No match → Continue
  ↓
2. Check Response Cache
   → Similar question found? → Return cached response ✅ (instant, free)
   → No cache → Continue
  ↓
3. Check Evaluation Status (if evaluation plan)
   → Expired? → Block LLM, show upgrade prompt ❌
   → Active → Continue
  ↓
4. Call LLM (OpenAI API)
   → Generate response
   → Save to cache for future use
   → Return response ✅

Best Practices

For Pre-configured Responses

1. Start with common queries: Add pre-configured responses for greetings, thanks, and your top 5-10 FAQs
2. Use regex for variations: Instead of multiple exact matches, use regex like hi|hello|hey
3. Set priorities: Give important responses higher priority
4. Test patterns: Use the bot interface to test if your patterns match correctly
5. Keep it simple: Don't over-complicate patterns; start simple and refine

For Response Caching

1. Let it work automatically: No configuration needed, just let it run
2. Clear cache when needed: Use "Clear Cache" button after major bot updates
3. Monitor cache performance: Check if similar questions are being cached effectively
4. Update system prompt carefully: System prompt changes clear the cache automatically

Combining Both

The best strategy is to use both:

• Pre-configured responses for predictable, common queries (greetings, simple FAQs)
• Response caching for AI-generated responses to similar questions
• Together, they maximize speed and minimize costs

Default Response

In addition to pre-configured responses, you can set a Default Response in your bot settings. This is used when:

• No pre-configured response matches
• No cached response is available
• LLM is unavailable (API down, evaluation ended, etc.)

Setting Default Response:

1. Go to bot Edit page
2. Find "Default Response" field (below System Prompt)
3. Enter your message (e.g., "I'm currently unable to process complex questions. Please contact support for more assistance.")
4. Save

Note: Special system patterns (system_unavailable, quota_exceeded) take precedence over the default response.

Evaluation Behavior

During Evaluation (Days 1-14)

• ✅ Pre-configured responses work
• ✅ Cached responses work
• ✅ LLM responses work

After Evaluation Ends (Day 15+)

• ✅ Pre-configured responses still work (within plan limits)
• ✅ Cached responses still work (if previously cached)
• ❌ LLM responses blocked (requires upgrade)

This graceful degradation ensures your bot continues to work for simple queries while encouraging upgrades for AI-powered responses.

Troubleshooting

Pre-configured Response Not Matching

Problem: Pattern doesn't match expected messages

Solutions:

• Check pattern type (exact vs keyword vs regex)
• Verify case sensitivity setting
• Test with actual user messages
• Use regex tester to validate patterns
• Check priority (higher priority responses checked first)

Cache Not Working

Problem: Similar questions not using cached responses

Solutions:

• Cache requires exact or very similar questions (85%+ similarity)
• Check if cache was cleared recently
• Verify system prompt hasn't changed (clears cache)
• Wait for cache to populate (first question always calls LLM)

Response Not Showing

Problem: Expected response not appearing

Check order:

1. Is there a pre-configured response that matches?
2. Is there a cached response for similar question?
3. Is evaluation expired? (blocks LLM)
4. Is default response set?
5. Check special system patterns

Advanced Tips

Regex Patterns

• ^(hi|hello) - Starts with "hi" or "hello"
• (thanks|thank you|ty) - Contains any of these
• .*hours.* - Contains "hours" anywhere
• ^(yes|y|ok|okay)$ - Exact match for any of these

Priority Strategy

• Set high priority (e.g., 100) for critical responses
• Set medium priority (e.g., 50) for common responses
• Set low priority (e.g., 0) for fallback responses

Cache Management

• Clear cache after major documentation updates
• Clear cache if responses seem stale
• Cache automatically expires after 30 days
• Cache is per-bot (not shared across bots)

Summary

Pre-configured Responses = Manual, pattern-based, instant responses for predictable queries
Cached Responses = Automatic, similarity-based, instant responses for similar past questions

Together, they provide:

• ⚡ Faster responses
• 💰 Lower costs
• 🎯 Better user experience
• 🔄 Graceful degradation when LLM is unavailable

Use pre-configured responses for common queries, let caching handle similar questions, and set a default response for edge cases. This combination maximizes your bot's efficiency and cost-effectiveness.