Every CRM vendor says their platform is “AI-powered.” Most mean it has a subject line suggestion tool.
This post is about something different. A real implementation — multi-location franchise, GoHighLevel CRM, AI generating personalized SMS and email at every stage of the pipeline. Here’s exactly how it works.
The Problem With Templates
Templates feel personal until you receive one. The giveaway is always the same: the message could have been sent to anyone.
- “Hi, this is an automated message from [Business Name]. We noticed you recently inquired about our services. Please click the link below to schedule your visit.”
- Day 0 and Day 1 SMS — identical
- No mention of what they asked about
- “Hi Sarah — thanks for coming in Thursday. You mentioned you were looking at the infant program starting in September. I’ve held a spot through end of week.”
- References specific service, manager they met, and real next step
- Reads like a person wrote it
That difference isn’t magic. It’s inputs.
What the AI Actually Reads
Before generating any message, the system pulls four data points from the contact record:
1. Customer segment — four distinct groups, each with different pain points and timelines
2. Manager visit notes — what the customer said, asked about, any concerns
3. Pipeline stage — where they are in the decision process right now
4. Location-specific details — branch name, staff member, current availability
Those four inputs go into a structured prompt. The AI returns a draft. The system sends it. The customer never knows AI was involved.
This isn’t chatbot territory. The AI writes the draft. The system sends it. No customer is talking to AI.
The Pipeline — Stage by Stage
At every transition point, the system generates a new message — not from a template, but from the current state of the contact record.
Each of the four customer segments gets a different messaging track entirely. Different urgency framing. Different CTAs. Different assumed objections.
Behavioral Branching
Personalization at send is one layer. What happens next matters just as much.
| Signal | System Response |
|---|---|
| Replied to SMS or email | Flag for staff takeover — AI hands off immediately |
| Email opened, no reply | Switch to SMS-first for the next touchpoint |
| Link clicked, no booking | Trigger targeted follow-up referencing the page they visited |
| Two consecutive unopened emails | Flag contact for phone call — remove from automated sequence |
The system isn’t just sending messages. It’s reading responses — or the absence of them — and adjusting channel and timing accordingly.
The Numbers
The cost framing matters. AI API calls run fractions of a cent per message. One additional conversion — at any meaningful customer LTV — covers a year of API costs.
One additional conversion typically covers 12+ months of API costs. This isn’t a technology investment with a long payback period — it pays for itself on the first win.
What Actually Makes It Work
The AI isn’t the hard part. The data is.
For this to generate useful messages, the CRM needs to be capturing the right inputs consistently. Visit notes have to be logged. Segments have to be assigned. Stage transitions have to be accurate.
This is why it’s an operations problem before it’s an AI problem. The pipeline structure, custom fields, and staff workflow all have to be in place first. The AI layer drops in on top of a system that already works.
Build the infrastructure. Then make it smart.
Frequently Asked Questions
Running on a stack that grew by accident?
Tools added one at a time, never architected together. That's the problem I solve. Book 45 minutes and I'll map what moves, what stays, and what makes sense for your operation.
Book a Discovery Call