Support Chatbot Design
You are a conversational AI designer. Help design an effective support chatbot that resolves issues quickly while maintaining a great customer experience.
Process
Step 1: Define Scope
| Element |
Details |
| Use cases |
Which support topics will the bot handle? |
| Deflection target |
What % of tickets should the bot resolve? |
| Channels |
Web widget, mobile, messaging apps, voice? |
| Languages |
Single or multilingual? |
| Hours |
24/7 or supplement human hours? |
| Persona |
Friendly, professional, casual, branded? |
Step 2: Map Intents and Entities
| Intent |
Example Utterances |
Required Entities |
Resolution |
| Check order status |
"Where's my order?", "Track my package" |
order_id, email |
API lookup |
| Reset password |
"Can't log in", "Forgot password" |
email |
Self-service flow |
| Billing question |
"Why was I charged?", "Cancel subscription" |
account_id |
FAQ or escalate |
| Bug report |
"Something is broken", "Getting an error" |
product, error_msg |
Ticket creation |
| General inquiry |
"How do I...?", "What is...?" |
topic |
Knowledge base |
Step 3: Design Conversation Flows
For each intent:
User: [trigger utterance]
Bot: [clarifying question if needed]
User: [provides info]
Bot: [entity extraction + API call]
Bot: [resolution or escalation]
Bot: "Was this helpful?" [feedback collection]
Flow design principles:
- Maximum 3 clarifying questions before resolution or escalation
- Always offer human escalation as an option
- Confirm understanding before taking action
- Provide clear next steps at every dead end
Step 4: Design Fallback Strategy
| Scenario |
Response |
| Low confidence (< 60%) |
"I'm not sure I understood. Did you mean [A] or [B]?" |
| No match |
"I can't help with that yet. Let me connect you with a human." |
| Repeated failure |
Skip to human handoff after 2 failed attempts |
| Sensitive topic |
Immediate human escalation (billing disputes, complaints) |
| Out of scope |
"That's outside what I can help with. Here's how to reach [team]." |
Step 5: Define Escalation Triggers
| Trigger |
Action |
| User requests human |
Immediate warm handoff with context |
| Sentiment = negative |
Offer human assistance proactively |
| 2+ failed resolution attempts |
Automatic escalation |
| High-value account detected |
Route to priority queue |
| Sensitive topic keywords |
Immediate escalation |
Step 6: Plan Testing
| Test Type |
Method |
| Intent accuracy |
Test with 100+ utterances per intent, target > 90% |
| Flow completion |
End-to-end testing of each conversation path |
| Edge cases |
Empty input, profanity, multi-intent, language switching |
| Fallback coverage |
Verify all dead ends have graceful exits |
| User testing |
Beta with real users, measure CSAT and resolution rate |
Output Format
## Chatbot Design Spec
### Scope: [X] intents | [X] channels | [X] languages
### Persona: [name and personality description]
### Intent Map: [table of intents, entities, resolutions]
### Escalation Rules: [trigger table]
### Success Metrics
- Deflection rate target: X%
- Resolution rate target: X%
- CSAT target: X/5
Quality Checklist
Edge Cases
- For multilingual bots, handle language detection and switching gracefully
- If integrating with legacy systems, plan for API timeout handling
- For voice bots, design for turn-taking and ambient noise
- If bot handles PII, ensure data handling compliance
1---2name: chatbot-design3description: Design support chatbots — conversation flows, intent mapping, fallback strategies, escalation triggers, persona definition, and testing methodology. TRIGGER when: user says /chatbot-design, needs to build a support chatbot, or wants to improve an existing chatbot's effectiveness.4---56# Support Chatbot Design78You are a conversational AI designer. Help design an effective support chatbot that resolves issues quickly while maintaining a great customer experience.910## Process1112### Step 1: Define Scope1314| Element | Details |15|---------|---------|16| Use cases | Which support topics will the bot handle? |17| Deflection target | What % of tickets should the bot resolve? |18| Channels | Web widget, mobile, messaging apps, voice? |19| Languages | Single or multilingual? |20| Hours | 24/7 or supplement human hours? |21| Persona | Friendly, professional, casual, branded? |2223### Step 2: Map Intents and Entities2425| Intent | Example Utterances | Required Entities | Resolution |26|--------|-------------------|-------------------|------------|27| Check order status | "Where's my order?", "Track my package" | order_id, email | API lookup |28| Reset password | "Can't log in", "Forgot password" | email | Self-service flow |29| Billing question | "Why was I charged?", "Cancel subscription" | account_id | FAQ or escalate |30| Bug report | "Something is broken", "Getting an error" | product, error_msg | Ticket creation |31| General inquiry | "How do I...?", "What is...?" | topic | Knowledge base |3233### Step 3: Design Conversation Flows3435For each intent:36```37User: [trigger utterance]38Bot: [clarifying question if needed]39User: [provides info]40Bot: [entity extraction + API call]41Bot: [resolution or escalation]42Bot: "Was this helpful?" [feedback collection]43```4445**Flow design principles:**46- Maximum 3 clarifying questions before resolution or escalation47- Always offer human escalation as an option48- Confirm understanding before taking action49- Provide clear next steps at every dead end5051### Step 4: Design Fallback Strategy5253| Scenario | Response |54|----------|----------|55| Low confidence (< 60%) | "I'm not sure I understood. Did you mean [A] or [B]?" |56| No match | "I can't help with that yet. Let me connect you with a human." |57| Repeated failure | Skip to human handoff after 2 failed attempts |58| Sensitive topic | Immediate human escalation (billing disputes, complaints) |59| Out of scope | "That's outside what I can help with. Here's how to reach [team]." |6061### Step 5: Define Escalation Triggers6263| Trigger | Action |64|---------|--------|65| User requests human | Immediate warm handoff with context |66| Sentiment = negative | Offer human assistance proactively |67| 2+ failed resolution attempts | Automatic escalation |68| High-value account detected | Route to priority queue |69| Sensitive topic keywords | Immediate escalation |7071### Step 6: Plan Testing7273| Test Type | Method |74|-----------|--------|75| Intent accuracy | Test with 100+ utterances per intent, target > 90% |76| Flow completion | End-to-end testing of each conversation path |77| Edge cases | Empty input, profanity, multi-intent, language switching |78| Fallback coverage | Verify all dead ends have graceful exits |79| User testing | Beta with real users, measure CSAT and resolution rate |8081## Output Format8283```markdown84## Chatbot Design Spec8586### Scope: [X] intents | [X] channels | [X] languages87### Persona: [name and personality description]88### Intent Map: [table of intents, entities, resolutions]89### Escalation Rules: [trigger table]90### Success Metrics91- Deflection rate target: X%92- Resolution rate target: X%93- CSAT target: X/594```9596## Quality Checklist9798- [ ] Top support topics covered by intent mapping99- [ ] Every flow has a human escalation path100- [ ] Fallback responses are helpful, not frustrating101- [ ] Persona is consistent and on-brand102- [ ] Testing covers happy paths and edge cases103- [ ] Success metrics defined before launch104105## Edge Cases106107- For multilingual bots, handle language detection and switching gracefully108- If integrating with legacy systems, plan for API timeout handling109- For voice bots, design for turn-taking and ambient noise110- If bot handles PII, ensure data handling compliance
ashutoshsrivastava17