omnisonant-design（全声设计）

Omnisonant Design Guide

Product: Omnisonant — Every channel, one voice.

Tagline: AI voice agents that handle calls so humans don't have to.

Repo: git@github.com:Forth-AI/omnisonant.git

Trigger: When designing, reviewing, or improving Omnisonant features, voice agent behaviors, or architecture.

1. The Paradigm Shift

Traditional Call Center Model

Customer calls → IVR maze → Hold music → Human agent → Manual CRM update
                    ↓           ↓            ↓               ↓
               Frustrating   Expensive    Inconsistent   Error-prone

Problems:

• Cost: $15-25/hour per agent
• Availability: Limited hours, timezone constraints
• Consistency: Agent quality varies
• Scale: Hiring/training bottleneck
• Data: Manual entry, lost context

Omnisonant Model

Customer calls → Voice agent → Instant resolution → Automatic CRM update
                     ↓               ↓                    ↓
                24/7 ready    Perfectly consistent    Zero manual work

Value:

• Cost: $0.10-0.20 per call (95%+ savings)
• Availability: 24/7/365, any timezone
• Consistency: Same quality every time
• Scale: Infinite parallel calls
• Data: Automatic logging, full transcripts

2. Target Audiences

Primary: SMB Owners

Profile: Small business owners who hate phone work

Buying motivation: "I want to fire my phones"

Design implication: Must be self-service, no technical setup required.

Secondary: Voice AI Resellers/Agencies

Profile: Agencies building voice solutions for clients

Buying motivation: "I want to sell this to my clients"

Design implication: Multi-tenant, white-label capable, reseller dashboard.

3. Three Core Use Cases

Use Case 1: Appointment Scheduling (Outbound)

Replaces: Staff calling to confirm/reschedule appointments

Example vertical: Dental clinic

Flow:

Agent calls patient → Confirms tomorrow's appointment →
  ✓ Confirmed: "Great, see you at 2pm"
  ↻ Reschedule: Opens calendar, finds slot, books
  ✕ Cancel: Marks cancelled, offers future booking
→ Updates calendar automatically

Voice Agent Script Pattern:

"Hi, this is Sarah from [Business Name].
I'm calling to confirm your appointment tomorrow at [time] with [provider].
Can you make it?"

[If yes] "Great! We'll see you then. Is there anything you need before your visit?"

[If reschedule] "No problem! Let me check what we have available.
How about [alternative 1] or [alternative 2]?"

[If cancel] "I understand. Would you like me to book a future appointment,
or shall I have someone call you later?"

Key metrics:

• Confirmation rate: Target 80%+
• Reschedule rate: Track, optimize
• No-show reduction: Target 50%+
• Call duration: Target <2 min

Design requirements:

• Calendar integration (Google, Outlook, practice management)
• Smart slot suggestion (based on availability + preferences)
• Reminder confirmation (SMS after call)
• Retry logic (voicemail, callback attempts)

Use Case 2: Lead Qualification (Outbound)

Replaces: SDRs making initial qualification calls

Example vertical: Real estate

Flow:

Lead submits form → Agent calls within 5 minutes →
  Qualifies: Budget, timeline, preferences →
    ✓ Hot lead: Books showing with human agent
    ~ Warm lead: Adds to nurture sequence
    ✕ Cold lead: Marks as not ready
→ Updates CRM with full notes

BANT Qualification Script:

"Hi [Name], this is Alex from [Agency].
You inquired about homes in [area]. Do you have a few minutes to chat?"

[Budget] "What price range are you looking at?"

[Authority] "Will anyone else be involved in the decision?"

[Need] "What's prompting your move? More space, new job, investment?"

[Timeline] "When are you hoping to move by?"

[If qualified] "Based on what you've told me, I think [Agent Name]
would be perfect to show you some properties. They're available
[time slots]. Which works for you?"

Key metrics:

• Contact rate: Target 60%+
• Qualification completion: Target 70%+
• Lead-to-meeting conversion: Target 30%+
• Speed to lead: Target <5 min

Design requirements:

• CRM integration (Salesforce, HubSpot, etc.)
• Lead scoring based on answers
• Intelligent routing to human agents
• A/B testing for scripts
• Time-of-day optimization

Use Case 3: Customer Support (Inbound)

Replaces: Tier-1 support agents handling common queries

Example vertical: E-commerce

Flow:

Customer calls → Agent identifies (phone/order#) →
  📦 Order status: Pulls from system, provides update
  ↩️ Return request: Creates RMA, sends label
  ❓ General question: Answers from knowledge base
  ⚠️ Complex issue: Escalates to human
→ Logs interaction, updates ticket

Support Script Pattern:

"Thank you for calling [Company]. This is your AI assistant.
How can I help you today?"

[Order status] "I'd be happy to check on your order.
Can I get your order number or the email address on the account?"
→ "I found your order. It shipped on [date] and should arrive by [date].
Would you like me to text you the tracking number?"

[Return] "I can help you start a return. Which item would you like to return?"
→ "I've created a return label for you. It's being sent to [email].
Is there anything else I can help with?"

[Escalation] "I want to make sure you get the best help for this.
Let me connect you with a specialist. Please hold for just a moment."

Key metrics:

• Resolution rate (no human needed): Target 70%+
• Customer satisfaction: Target 4+/5
• Average handle time: Target <3 min
• Escalation rate: Target <30%

Design requirements:

• Order system integration
• Return/refund workflow automation
• Knowledge base for FAQs
• Seamless escalation to human
• Post-call survey option

4. Voice Agent Design Principles

Principle 1: Sound Human, Be Honest

Right: "Hi, this is Sarah, an AI assistant calling from..."
Wrong: Pretending to be human without disclosure

Right: Natural speech patterns, appropriate pauses
Wrong: Robotic cadence, unnatural phrasing

Why: Trust requires transparency. Deception backfires.

Principle 2: Graceful Interruption Handling

Right: Stop talking when customer speaks, acknowledge, respond
Wrong: Keep talking over customer, ignore interruption

Right: "Oh, go ahead!" → listens → responds to what they said
Wrong: "Please wait for me to finish"

Why: Natural conversation requires turn-taking.

Principle 3: Fast and Focused

Right: Get to the point, respect their time
Wrong: Long introductions, excessive pleasantries

Right: "Hi, this is Sarah from Bright Smile confirming your appointment tomorrow at 2pm. Can you make it?"
Wrong: "Hello! How are you doing today? I hope you're having a wonderful day! I'm calling from..."

Why: People hate phone calls. Make them short.

Principle 4: Recover Gracefully

Right: "I didn't quite catch that. Could you repeat the date?"
Wrong: "Error. Invalid input. Please try again."

Right: "Hmm, I'm having trouble finding that order. Let me connect you with someone who can help."
Wrong: [Silence] or [Hang up]

Why: Errors happen. Recovery maintains trust.

Principle 5: Confirm Before Acting

Right: "So I'll book you for Thursday at 3pm with Dr. Chen. Does that sound right?"
Wrong: "Done. Goodbye." [Hangs up]

Right: Wait for confirmation before finalizing
Wrong: Assume and execute without verification

Why: Mistakes are costly. Confirmation is cheap.

Principle 6: End with Clear Next Steps

Right: "You'll get a text confirmation in a moment. Is there anything else?"
Wrong: "Okay, bye."

Right: Tell them what happens next
Wrong: Leave them wondering

Why: Closure creates confidence.

5. Voice & Personality Guidelines

Voice Selection Criteria

Personality Traits

For appointment scheduling:

• Friendly, efficient, respectful of time
• "I know you're busy, so I'll be quick"

For lead qualification:

• Curious, engaged, consultative
• "Tell me more about what you're looking for"

For customer support:

• Patient, helpful, solution-oriented
• "Let me take care of that for you"

Things Voice Agents Should NEVER Do

• Pretend to be human when directly asked
• Get frustrated or impatient
• Argue with the customer
• Share information about other customers
• Make promises outside their authority
• Continue calling after "stop calling me"

6. Technical Architecture Principles

Dual Pipeline Support

Design implication: Abstract voice pipeline so agents work on either.

Latency Budget

Ideal conversation turn:
  Customer speaks     → 500ms → Agent responds

Acceptable:
  Customer speaks     → 800ms → Agent responds

Frustrating:
  Customer speaks     → 1500ms+ → Agent responds

Design implication: Every millisecond matters. Optimize ruthlessly.

Tool Execution Model

Customer: "What's the status of my order?"
    ↓
Agent: [Thinking] "Let me check that for you"
    ↓
Tool call: lookupOrder({ phone: "+1..." })
    ↓
Agent: "Your order shipped yesterday and should arrive Friday."

Design implication: Tools must be fast (<1s) and reliable.

Fallback Strategy

Level 1: Agent handles completely
Level 2: Agent + tool call
Level 3: Agent transfers to human
Level 4: Agent takes message for callback

Design implication: Never dead-end. Always a path forward.

7. Value Proposition Checklist

Every feature must deliver on at least one:

✅ Cost Reduction

• Does this reduce cost per call?
• Does this reduce need for human agents?
• Is ROI measurable and significant?

✅ Availability Improvement

• Does this extend service hours?
• Does this handle more concurrent calls?
• Does this reduce wait times?

✅ Consistency Improvement

• Does this ensure same quality every call?
• Does this reduce human error?
• Does this improve compliance?

✅ Scale Enablement

• Does this remove hiring bottleneck?
• Does this handle demand spikes?
• Does this expand geographic reach?

Red flags (features that don't fit):

• "Requires human review for every call" ❌
• "Only works during business hours" ❌
• "Needs custom development per client" ❌
• "Improves metrics but costs more" ❌

8. Interface Patterns

Admin Dashboard

Primary actions:

1. View active calls (live monitoring)
2. Review call history + transcripts
3. Configure voice agents
4. Manage campaigns (outbound)
5. View analytics

Key UX requirements:

• Real-time call status visibility
• One-click access to any call transcript
• Easy agent script editing
• Clear performance metrics

Agent Builder

Primary actions:

1. Define agent persona (name, voice, personality)
2. Set greeting and conversation flow
3. Configure available tools
4. Test with sample calls
5. Deploy to phone number

Key UX requirements:

• Natural language prompt editing
• Voice preview (hear before deploy)
• Sandbox testing environment
• A/B testing support

Campaign Manager (Outbound)

Primary actions:

1. Upload/select contact list
2. Choose agent and script
3. Set calling schedule and rules
4. Monitor progress
5. Review results

Key UX requirements:

• Bulk contact management
• Scheduling controls (time windows, timezone handling)
• Real-time progress dashboard
• Export results to CRM

Web Demo Interface

Primary actions:

1. Click to start call
2. Speak with agent
3. See live transcript
4. Experience the product

Key UX requirements:

• One-click to start (no signup for demo)
• Visual audio feedback
• Live transcript display
• Mobile-friendly

9. Anti-Patterns (Omnisonant-Specific)

"Sounds Like a Robot"

Symptom: Unnatural speech, no personality, mechanical responses. Fix: Better prompts, voice selection, natural language patterns.

"IVR in Disguise"

Symptom: "Press 1 for...", rigid menu trees, no natural conversation. Fix: Open-ended listening, intent detection, flexible responses.

"Infinite Hold"

Symptom: Can't reach human when needed, escalation fails. Fix: Clear escalation paths, graceful handoffs, callback option.

"Amnesia Agent"

Symptom: Doesn't remember what was said earlier in call. Fix: Proper context management, conversation memory.

"Over-Promising Agent"

Symptom: Agent commits to things it can't deliver. Fix: Constrain agent authority, confirm before committing.

"The Interrogator"

Symptom: Feels like a survey, too many questions, no empathy. Fix: Conversational flow, acknowledge answers, show understanding.

"Uncanny Valley"

Symptom: Too human-like in a way that's creepy. Fix: Honest about being AI, consistent persona, appropriate boundaries.

10. Competitive Positioning

vs. Traditional Call Centers

Omnisonant advantage: 95%+ cost reduction with better consistency.

vs. IVR Systems

Omnisonant advantage: People hate IVR. They tolerate or even enjoy good AI.

vs. Vapi (Direct)

Omnisonant advantage: Vapi is infrastructure. Omnisonant is solution.

vs. Other Voice AI Platforms

Omnisonant advantage: "Omni" in the name is the promise.

11. Review Checklist

When reviewing Omnisonant designs:

Voice Quality

• Does it sound natural?
• Is there appropriate personality?
• Are interruptions handled well?
• Is latency acceptable (<800ms)?

Conversation Quality

• Does it get to the point quickly?
• Does it confirm before acting?
• Does it recover from errors gracefully?
• Does it end with clear next steps?

Business Value

• Does this reduce cost per call?
• Does this extend availability?
• Does this improve consistency?
• Does this enable scale?

Integration Quality

• Does data flow to CRM automatically?
• Are actions executed in real systems?
• Is escalation to humans seamless?
• Are transcripts accessible?

User Experience (Admin)

• Can they set up without technical help?
• Can they monitor calls in real-time?
• Can they make changes without coding?
• Can they measure ROI?

12. Key Metrics

Call Quality

• Resolution rate: Calls resolved without human (Target: 70%+)
• Customer satisfaction: Post-call rating (Target: 4+/5)
• Average handle time: Call duration (Target: <3 min)
• Error rate: Calls with issues (Target: <5%)

Business Impact

• Cost per call: All-in cost (Target: <$0.25)
• Conversion rate: Leads converted, appointments booked (varies)
• ROI: Savings vs. human agents (Target: 10x+)

Technical Performance

• Latency: Time to first response (Target: <500ms)
• Uptime: System availability (Target: 99.9%+)
• Accuracy: Speech recognition accuracy (Target: 95%+)

13. Feature Prioritization Framework

Must Have (P0)

• Core voice conversation capability
• At least one use case working end-to-end
• Basic analytics (calls, duration, outcomes)
• Phone number provisioning

Should Have (P1)

• All three use cases polished
• CRM integrations (top 3)
• Campaign management
• Transcript search

Nice to Have (P2)

• White-label support
• Custom voice training
• Advanced analytics
• Multi-language

Won't Build (v1)

• Video calling
• Chat/SMS (v2)
• Custom voice cloning
• On-premise deployment

14. The Omnisonant Promise

To SMB owners:

"Your phone rings, AI answers. Appointments get confirmed. Leads get qualified. Customers get helped. You get your time back. All for less than the cost of a part-time receptionist."

To resellers:

"Add voice AI to your service offering. White-label our platform. Your clients get cutting-edge technology. You get recurring revenue."

Every design decision should reinforce this promise.

15. Voice Agent Prompt Template

Use this structure for creating voice agents:

## Agent Identity
- Name: [Agent name, e.g., "Sarah"]
- Company: [Business name]
- Role: [What they do, e.g., "appointment coordinator"]

## Personality
[2-3 sentences describing tone, style, approach]

## Goal
[Primary objective of this call]

## Key Information to Gather/Share
1. [Item 1]
2. [Item 2]
3. [Item 3]

## Available Actions
- [Action 1: e.g., "Book appointment"]
- [Action 2: e.g., "Check availability"]
- [Action 3: e.g., "Transfer to human"]

## Constraints
- Never [constraint 1]
- Always [constraint 2]
- If [condition], then [action]

## Escalation Triggers
- [When to transfer to human]
- [When to offer callback]

## Closing
[How to end the call professionally]