We built VibeTokens' AI intake chat to solve a specific problem: initial consultation calls were eating my calendar, and most of them covered the same ground.
"What do you do?" "What does it cost?" "How long does it take?" "What would you build for a business like mine?"
Forty-five minutes per call. Multiple calls per week. Most of them were pre-qualification conversations, not real consultations.
Now an AI handles all of that. Here's exactly how we built it and what we learned.
The Problem We Were Solving
Every service business has a version of this problem.
Prospects want information before they commit to a call. Owners spend time on calls that don't convert. There's a gap between "I'm interested" and "I'm ready to talk specifics" that currently requires human time.
An AI intake system fills that gap. It answers the common questions, qualifies the prospect, and schedules the right kind of call — or routes them to the right resource — without requiring my time.
The Architecture
Our intake system has five components:
1. The trigger: A chat widget on the VibeTokens website. Initiates after a visitor has spent 45 seconds on the page — long enough to have read something, not so long they're already leaving.
2. The AI conversation model: We built on top of an LLM (Claude in our case) with a detailed system prompt that defines the AI's role, its knowledge of our services, pricing ranges, timelines, and process.
3. The knowledge base: Structured content about everything VibeTokens does. Services, typical client profiles, what we need from a client to scope a project, common questions and our answers to them.
4. The qualification logic: The system is programmed to gather specific information — business type, the problem they're trying to solve, their budget range, timeline. This determines what kind of follow-up makes sense.
5. The handoff: Qualified prospects are offered a direct calendar link to book a 30-minute scoping call. Unqualified or informational inquiries get routed to relevant content. Contact info is captured either way.
The System Prompt
This is the most important piece. The system prompt is what makes an AI tool feel like your assistant versus a generic chatbot.
Ours defines:
- Who Murph is and VibeTokens' background
- The specific services we offer and what makes them different
- Pricing ranges (general, not exact — "most websites are in the $X-$X range")
- What we need to know to give a useful scoping conversation
- How to handle questions we can't answer ("I'll have Murph reach out about that directly")
- The tone — direct, no-BS, not salesy
Writing a good system prompt takes iteration. Our current version is the eighth draft. The first seven were too generic, too salesy, or didn't handle edge cases well enough.
What It Actually Does in a Conversation
A typical interaction:
Visitor: "I need a new website for my landscaping company."
AI: "Good to know. What's not working with your current site — or are you starting from scratch?"
Visitor: "Current site is old, doesn't work on mobile, and I'm not getting any leads from it."
AI: "That's exactly what we fix. For a landscaping company, our typical build includes [specific elements relevant to service businesses]. What's the current situation — how many services do you offer, do you serve multiple locations?"
...and so on. It gathers context. It explains our process. It gives a ballpark on timeline and pricing. Then:
AI: "Based on what you've described, a scoping call would take about 30 minutes. Murph can look at your current site, discuss what we'd build, and give you a specific quote. Want to grab a time this week?"
That's a pre-qualified prospect booking their own call. Without any of my time until the call happens.
What It Doesn't Do
Important: the AI doesn't close deals. It doesn't make commitments on pricing. It doesn't have the final say on scope.
It's a qualified-lead-to-calendar-booking machine. The actual relationship, the specific scoping, and the close happen on the call with me.
If someone asks for a specific price, the AI gives a range and notes that Murph will give an exact quote after the scoping call. It doesn't invent numbers or make commitments.
The Build (What You'd Need)
To build something similar:
Platform options:
- Custom build with a chatbot framework (Botpress, Voiceflow) connected to an LLM API
- Simpler: Tidio or Intercom with AI features and a detailed knowledge base
- More advanced: Custom Next.js widget with direct API calls to Claude or OpenAI
Knowledge base:
- Write out every common question and your ideal answer
- Document your services, pricing ranges, process, and typical client profile
- Include what you need from a prospect to give them a useful conversation
Calendar integration:
- Calendly, Cal.com, or your existing booking tool
- The AI provides a direct link when the prospect is ready to book
CRM integration:
- Every conversation gets logged
- Contact info is captured and pushed to your CRM automatically
What It Changed
Before: I was spending 8-10 hours per week on pre-qualification calls. After: I spend 30-45 minutes on those calls total.
The ones that do get to the calendar are better qualified. They've already been through the basics. They know roughly what things cost. They understand the process. The call is a specific scoping conversation, not a "what do you do" conversation.
That time recovery is worth more than the tool cost by an order of magnitude.
Could You Build This for Your Business?
Yes, if you have clearly-defined services and a consistent intake process.
It works best for businesses where:
- The initial inquiry usually covers similar ground
- There's enough volume that intake calls are a real time drain
- You have a natural next step (call, consultation, estimate) to route qualified prospects to
It doesn't work well for businesses where every inquiry is genuinely unique and requires human judgment from the first moment.
If you want to explore building one, that's what we do. We can usually scope and build an intake system in a week.
