Content is user-generated and unverified.

AI Assistant Routing & Specialization - Complete Implementation Cheat Sheet

Based on Mike Copeland's Assistable Workshop Training

Core Philosophy: Specialized AI Personalities

The Fundamental Principle

"You probably wouldn't have your front desk person be your sales person. This is where we separate it, right? We have personality traits that are kind of important for these roles." - Mike Copeland

Key Insight: Different business functions need different AI personalities, not one general-purpose receptionist bot.


1. AI Personality Differentiation

Sales Bot (SDR) Characteristics

Direct Quote: "You are the SDR. Now I'm doing SDR because I want the tonality. I want the AI to have what's, this is SDR is sales development representative."

Generated Personality Traits (from single-sentence prompt):

  • Consultative, not pushy
  • Naturally curious
  • Confident, but humble
  • Empathetic
  • Quick thinking
  • Persistent ← Key differentiator

Primary Tasks:

"Sales is set up to identify pain points, right? Uh, and collect and confirm their information, to move them to a point- book and get them booked. Or if they're ready to buy, giving them the next steps there."

Support Bot (CSM) Characteristics

Direct Quote: "You are the CSM [Customer Success Manager]"

Generated Personality Traits:

  • Approachable
  • Patient
  • Understanding
  • Strategic thinker
  • No "persistent" trait ← Critical difference

Primary Tasks:

"It's gonna ask about support questions, current status, offer resources, check-ins, all that kind of information language there."

The Critical Difference

"In an agency sales bot, you want somebody that's probably a little bit more persistent. I don't know that we want a persistent customer support person that's like very, uh, motivated to try to get someone to show up to a support call. Uh, from support, we want to try to handle the support. We're on the call or create a ticket, right? Um, we don't want to be filling up our calendars with support. We want to keep it busy with our, um, sales." - Mike Copeland


2. Technical Implementation: Contact Type Routing

Core Setup Requirements

System: Go High Level with LC Phone (IVR doesn't work on Twilio)

Step 1: Create Specialized Assistants

Sales Assistant: "You are the SDR for [Company Name]"
Support Assistant: "You are the CSM for [Company Name]"

Step 2: Active Tags Configuration

Direct Quote: "So what we did, we just, we have phone numbers that are linked to our assistance. That's going to give us the ability for voice conversation to happen. Um, this is going to give us the ability for chats to happen."

Implementation:

  • Sales Bot → Active Tag: "lead"
  • Support Bot → Active Tag: "customer"

Step 3: IVR Automation Setup

Core Logic: Route based on Contact Type field (standard GHL field)

Workflow Structure:

1. Check: Does contact have "AI_off" tag?
   - YES → Route to human
   - NO → Continue to routing

2. Check: What is Contact Type?
   - Lead → Connect to Sales AI
   - Customer → Connect to Support AI
   - Other → Define additional routing

Direct Quote: "Contact type is a default standard field. So it's going to be right here in contact details and you scroll down. That's like your first name, full name, email, phone. And there's also type. A lot of people aren't using this, uhm, but you should if you aren't. Uh, so if you have like when somebody purchases on the website, make sure to update the type to customer."


3. The AI Kill Switch System

Implementation

Tag Name: AI_off (must be exact)

Direct Quote: "We now have kill switch. What kill switch is, is making sure that AI is on or off. And so what this means is you can keep the tag, the active tags assigned to your contacts. And. If you wanted to shut off the AI, you still can by not having to remove the tags for all of those bots. Really helpful."

How It Works

"If I have the AI off tag. And I try responding... what you'll see here in our processing... Kill switch equals true. And our AI has not responded. Right? So... I'm gonna go back to our contact and we're gonna remove AI off tag. So the information can still. It'll be, uhm, gathered by the AI, but it's not gonna allow it to do anything."

Key Benefit: Conversation history is preserved when humans take over.


4. Advanced Prompt Engineering Techniques

The Identity Programming Method

Problem: AI often ignores rule-based instructions

Solution: Program behavior into identity rather than rules

Direct Quote: "Jordan has said this before. If you say, like, I don't want you to think about a pink elephant, people are going to immediately have the thought of a pink elephant. Um, and AI, if you give it instructions and say, do not do this, you're actually kind of putting it in its head of the potential that it, You could do that."

Implementation Example: Instead of: "Only ask one question at a time" Use: "Alex is known for only asking one question at a time and never repeats or..."

Result: "It didn't cling to that. Like, it, it saw that in its rules, but it really likes its identity. That it's, like, it's known for."

Professional Sales Methodology Integration

NEPQ Training: "Maybe we could say, uh, trained, um, on any PQ, which is, uh, if you guys are familiar with Jeremy Minor"

Result: "These questions are definitely deeper, more Jeremy Minor style."


5. Complex Flow Management (Advanced)

Python-Style Structured Prompts

When to Use: Complex discovery flows where prospects might skip ahead

Direct Quote: "So if I, if it goes through this second question and then you say, how much is it? It will go in and it will like answer whatever it is... And then it will go back to where it was in the flow here. Because it's indexing where it left off."

Warning: "This prompt does, there's three things that are happening in this... you'll see voice latency says two seconds compared to if we jump into this one... voice latency is .77. All right. Um, and that's because there's a lot of context in here."

Best Practice: "I would separate this in three different ones."


6. Workflow Automation Setup

Customer Replied Workflow

Purpose: Automatically update contact types based on behavior

Logic:

IF Contact Type = Lead:
  - Add "lead" tag
  - Remove "customer" tag

IF Contact Type = Customer:
  - Add "customer" tag  
  - Remove "lead" tag

Rationale: "You could have a customer that may be cancelled. So if they cancelled, and they're now a lead, um, well, then they should be talking to the sales bot again."

Human Notification System

For AI_off Contacts:

Trigger: Customer Replied + Has "AI_off" tag
Action: Send notification "Inbound message needs human response"

7. Quality Control & Optimization

Prompt Length Impact

Data Point: Simple prompt (.77s latency) vs Complex prompt (2s latency)

Best Practice: "So guideline for the prompt structure that is, like, perfect to match with the system is pretty. It's typical if you go in here and you click to create a system and you generate it. Like, we are trying to give it the best prompting that it will be."

Voice vs Chat Considerations

Platform Limitation: "IVR does not work on Twilio. It's only LC phone for this."

Transfer Capability: "But what you can do is you can set up, the call transfers... so this is where we go back to... you can have like the receptionist that can handle sales or support or do a flowbuilder, um, and you can have a transfer the call to your sales AI or your support AI"


8. Common Implementation Mistakes

Over-Engineering Warning

Direct Quote: "Please don't- don't hit us with your JSON prompts or Python prompts. Uhm, yeah, just don't bog us down with that. If you- if you go into that territory, like, try that at your own discretion there. You're going in the deep end. And you- . . You don't have to have it for most things."

Prompt Terminology

Critical Point: "Important thing that I'm seeing still a lot of users doing is putting in contact, client, prospect, other things for the name of the person that it's talking to. If you look at our prompt when we generate it, it never says that, it's always user."

Correct Usage: Always use "user" not "contact," "client," or "prospect"

Emoji Usage

Don't Do This: "I've also seen people putting emojis in their prompts. You don't need to do that."


9. Business Impact Measurement

Key Performance Indicators

  • Conversation Routing Accuracy: Verify in server logs which bot handled each conversation
  • Flow Completion Rates: Track prospects completing discovery vs. jumping ahead
  • Human Handoff Success: Monitor AI_off tag usage and resolution times
  • Voice Latency: Monitor response times with different prompt complexities

Server Log Analysis

What to Look For: "You'll also see we have a different ID than the one we had before because we're now on our support bot."

Verification Process: Check conversation logs to confirm proper routing based on contact type.


10. Step-by-Step Implementation Guide

Phase 1: Basic Setup (Day 1)

  1. Create two assistants: Sales (SDR) and Support (CSM)
  2. Set up active tags: "lead" and "customer"
  3. Configure phone numbers for each assistant
  4. Test basic routing with manual contact type changes

Phase 2: Automation (Week 1)

  1. Build IVR workflow for automatic routing
  2. Create "Customer Replied" workflow for tag management
  3. Implement AI_off kill switch system
  4. Set up human notification workflows

Phase 3: Optimization (Week 2-4)

  1. Refine bot personalities based on conversation analysis
  2. Add transfer tools between bots
  3. Implement discovery flow sequences
  4. Monitor and optimize prompt performance

Phase 4: Advanced Features (Month 2+)

  1. Add specialized bots for specific use cases
  2. Implement complex flow management if needed
  3. Create client-specific bot variations
  4. Build comprehensive analytics dashboard

11. Practical Application Examples

For Agencies

Sales Bot Use Cases:

  • New business inquiries
  • Service capability assessment
  • Prospect qualification and booking

Support Bot Use Cases:

  • Existing client questions
  • Campaign performance inquiries
  • Technical support requests

For Service Businesses

Sales Bot Focus: "In an, in an agency sales bot, you want somebody that's probably a little bit more persistent"

Support Bot Focus: Problem resolution and resource provision without sales pressure


12. Advanced Routing Scenarios

Time-Based Routing

Example: "So you could do it during business hours. But you could say if, um, assigned, or I guess we have to go contact details. Assign user, um, is empty. Right? Then we could have it go to the AI. But you could say if it's not, Empty. Then it could go to, you know, maybe the assigned person."

Spam Prevention

Implementation: "You can use this to stop inbound spam calls. Umm, I mean, you can have AI that does, like, whatever. The, like, if it's tagged as spam, you could just have when they call in, if they're tagged as spam, just to, like, just not even connect."


13. Success Metrics & ROI

Efficiency Gains

  • Reduced founder involvement in routing decisions
  • 24/7 coverage with appropriate expertise
  • Faster response times for different inquiry types

Quality Improvements

  • More relevant responses based on inquiry type
  • Better conversion rates through specialized approaches
  • Improved customer satisfaction through appropriate service levels

Scalability Benefits

  • Handle multiple conversations simultaneously
  • Consistent service quality regardless of volume
  • Systematic approach to lead qualification and customer support

14. Transfer Tools Configuration (Advanced)

Setting Up Inter-Bot Transfers

Direct Quote: "So the way you do that, you go create new, you create transfer, uhh, transfer call, and you're gonna get this window."

Configuration Steps:

  1. Create Transfer Tool: Tools → Create New → Transfer Call
  2. Department Setup: "I have Alex with sales, Ellie with support, and I have the numbers here"
  3. Advanced Options: "If you don't see the ability for this, you might just have a phone number here. It's this button right here... it's gonna give you transfer destination prompt"

Warm Transfer Setup:

"I also set up a warm transfer, and I set in one sentence, explain to the caller why you have the user on the line."

Structure:

Transfer Destination: Department/Person Name
Phone Number: Specific bot's number
Transfer Type: Cold or Warm transfer
Context: Brief explanation for recipient

15. Phone Number Management & Registration

Number Registration

Best Practice: "If I want to verify this to make sure that we're- just getting better call answering and stuff, you can go here, free registry, make sure that you get your numbers registered there."

Voice Call Settings

Critical Configuration: "In your voice calls, you have inbound calling. And so you want to make sure if you don't have your, if you have ring user assigned to call number, it could bypass the AI and maybe you want it to do that. This is kind of a situational thing."

Decision Point: Choose whether calls bypass AI when user is assigned vs. always going to AI first.


16. Competitive Positioning & Migration Strategies

ZappyChat Migration Advantage

Market Opportunity: "Zappy chat was acquired right close by. Closebot does not have a migration for zappy chat. It's a full rebuild."

Solution Offered: "I have a less than five minute fix to keep anybody that was on zappy chat. Convert it over to us, keeping everything intact, so nothing breaks."

Competitive Edge: "Because it's a complete rebuild from close bot and they're charging them usage for every single text message."

Value Proposition

Service Pricing: "We are going to be offering Dunfree Buildouts... It's only going to be, like, $1,000. dollars. I think $9.97."


17. Advanced Routing Scenarios & Edge Cases

Assigned User Logic

Implementation: "You could say if, um, assigned, or I guess we have to go contact details. Assign user, um, is empty. Right? Then we could have it go to the AI. But you could say if it's not, Empty. Then it could go to, you know, maybe the assigned person."

Business Hours Routing

Example Logic: Route to AI during business hours, to voicemail or different flow after hours.

Outbound Campaign Integration

Advanced Use: "Let's say you had a, an, an AI that was. Your database reactivation AI. Like you have an outbound AI that's doing promotions and specials. Uh, and they're like, yeah, I'm interested and it's like, call me and then they. Call you can have it go specifically to that, like promotional AI bot."

Spam Prevention Automation

Workflow Setup:

Trigger: Tag "spam" added
Action: Add "AI_off" tag
Result: AI won't respond, prevents waste of resources

Alternative: "You could also have it just off contact type. Right. And the, um, if contact type is spam to remove all the tags that you have that are active on your contacts."


18. Troubleshooting Common Technical Issues

Phone Number Connection Problems

Common Issue: Numbers not connecting or routing properly

Troubleshooting Steps:

  1. Verify phone number is properly linked to correct assistant
  2. Check wallet balance for voice minutes
  3. Confirm IVR workflow is using correct phone number
  4. Test with manual contact type changes first

Latency and Performance Issues

Monitoring: "Prompt length matters... you'll see voice latency says two seconds compared to if we jump into this one, we click. This little rocket here, voice latency is .77"

Optimization: Keep prompts concise, use knowledge base instead of long context in prompts.

Contact Type Field Issues

Common Mistake: Not updating contact type when customers purchase Solution: "When somebody purchases on the website, make sure to update the type to customer"


19. Quality Assurance & Testing Protocols

Server Log Analysis

What to Monitor: "Here's our server logs. And so it says, uhm, the tag was a lead. And so it matched. us to the lead bot. You'll also see we added kill switch."

Key Indicators:

  • Which bot ID handled the conversation
  • Kill switch status (true/false)
  • Tag matching accuracy
  • Contact type routing verification

Conversation Flow Testing

Process:

  1. Test with different contact types manually
  2. Verify proper bot personality engagement
  3. Check transfer functionality between bots
  4. Validate AI_off tag behavior
  5. Monitor conversation quality and completion rates

20. Support Resources & Community

Available Support Channels

Training Schedule: "Q&A calls are 9 a.m. and 4 p.m. Mountain Time on Monday, Tuesdays. Thursdays and Fridays"

Team Resources:

  • Aiden: Technical Q&A calls
  • Andrew Hagan: "Former director of Zappy Chats coming into the picture... lead AI advisor... gonna be here. To help you guys with, um, prompting and structure"
  • Bernie: Assistable community support

Documentation Access

AI Support Docs: "We have AI support docs... This is something that you can eye frame into a custom menu link in your agency if you want to."

Branded Options: "If you guys want it branded, we can give you a branded one... you can put in your own articles, um, your logos and your stuff"

Communication Guidelines

Important: "If you are personally messaging Jordan, I just urge you guys, don't do that anymore, um, because we need to know what he knows, and if you- he's still trying to support folks on his own. Um, it slows down development and progress for you guys"

Proper Channels: Use school group, tag team members, create support tickets.


21. Business Model Applications

Agency Implementation

Multi-Client Management: Different bots for different client types or services Scalability: "You probably wouldn't have your front desk person be your sales person" - Same logic applies to AI

Service Business Applications

Lead Generation: Persistent sales bots for new prospects Customer Retention: Patient support bots for existing clients Operational Efficiency: Reduce founder involvement in routine routing


Final Implementation Checklist

  • Created specialized bot personalities (Sales SDR vs Support CSM)
  • Configured active tags for routing (lead/customer)
  • Set up IVR workflow for automatic routing
  • Implemented AI_off kill switch system
  • Created customer replied workflow for tag management
  • Set up human notification systems
  • Tested routing with different contact types
  • Configured transfer tools between bots
  • Monitored server logs for routing verification
  • Optimized prompt length for latency
  • Established performance measurement protocols

Remember: "You just saw me just do one sentence here and get a pretty good instructions here. A good format. Is this the finished product, Johnny? I would say it's not the finished product, but it's, this is probably 75% of the way there for a client." - Mike Copeland

This systematic approach ensures specialized AI experiences that match human business roles while maintaining the efficiency and scalability benefits of AI automation.

Content is user-generated and unverified.
    AI Assistant Routing & Specialization - Complete Implementation Cheat Sheet | Claude