At Null Studio we build AI voice and automation systems for local businesses — plumbers, dentists, HVAC companies, law firms. And almost every engagement starts the same way: someone tells me their marketing isn't working, when the real problem is that their phone rings and nobody answers it.
This isn't a hot take. It's just math nobody runs on their own business. So before I show you the build, let's run it.
The math nobody runs
Pull up your phone system's call log (or your CRM, or just your gut estimate) and fill in four numbers:
missed calls per week ×
% with real intent (≈60%) ×
your close rate on inquiries ×
average ticket = weekly leak
Here's a worked example from a plumbing client: 15 missed calls a week × 60% real intent × 30% close rate × $300 average ticket ≈ $810/week, or roughly $3,200/month. That's not money lost to bad ads or a weak website. That's money lost after the marketing already worked — after someone Googled, found them, and dialed the phone.
Two structural facts make this worse than it looks:
- You can't staff your way out of it. The call comes in at 7pm, or while the technician is under a sink, or when the front desk is on another line. Local businesses miss something like 40% of their calls, and that's not a training problem, it's a coverage problem.
- Voicemail is where intent goes to die. Roughly 80% of people who hit voicemail don't leave one. They just scroll to the next result. Your competitor's number is one thumb-tap away.
Once I show a business owner this math against their own numbers, the missed-call problem stops being an abstract "we should look into that" and becomes the most urgent thing on their list. It's also why "fix the missed calls" has become the first project we recommend to almost every local-business client, ahead of anything flashier like a full AI receptionist.
The fix, layer by layer
I deploy this in three layers, in this order, because each one buys ROI before you build the next.
Layer 1 — Missed-call textback (ships in an afternoon)
The instant a call goes unanswered, the caller gets a text:
"Sorry we missed you! How can we help? Reply here or book: {link}"
That's it. No AI, no voice model, just a trigger and a template. But it catches the lead in the exact moment their intent is highest — while they're still holding the phone, before they've opened Google again.
I build this in GoHighLevel wired to the business's phone system. The part that trips people up isn't the automation logic, it's A2P 10DLC registration. If you skip it, US carriers will silently filter your texts — no error, no bounce, they just never arrive, and you won't find out until a client asks why nobody's replying to their "missed you" texts. I've written up the full technical build with the registration steps here: nabeelbaghoor.com/blog/missed-call-textback-ghl-retell.
Layer 2 — AI callback (this is where it actually converts)
Textback captures the lead. It doesn't convert them — a text thread with a plumbing company isn't how anyone books a $300 job. So sixty seconds after the missed call, I have an AI voice agent (built on Retell) call the lead back. It answers basic questions, quotes rough pricing, and books the job straight into the business's real calendar, mid-call.
This is the same voice-agent engine I use for AI appointment setters generally, just pointed specifically at the missed-call stream instead of outbound or ad-driven leads. Conversation converts in a way a text thread rarely does, especially for anything with real ticket size behind it.
Layer 3 — Full AI receptionist (the endgame)
The first two layers recover missed calls. The third layer means there's nothing to recover, because every call gets answered live, in seconds, 24/7. This is the heaviest build of the three and usually only makes sense once call volume justifies it — but it's where a lot of these projects end up eventually.
What it actually costs vs what it recovers
- Textback: basically free to run (SMS costs pennies). Setup runs a few hundred dollars once you include A2P registration.
- AI callback / receptionist: usage costs land around $0.07–$0.20 per call-minute, with the service typically landing $150–$600/month all-in depending on volume.
Against a $3,200/month leak, the whole stack pays for itself on the first one or two recovered jobs. That math is why I lead with this project instead of trying to sell a business owner on something more impressive-sounding first.
The details that decide whether it actually works
I've shipped this stack enough times now to have a short list of things that separate "works great" from "generates spam complaints and gets turned off in week two":
- Text from the number they called. If the reply comes from a different number, you orphan the conversation — the customer replies and it goes nowhere.
- Respect quiet hours. A "sorry we missed you!" text at 1am earns a spam report, not a booking. Queue overnight misses for the morning.
- Filter out existing customers. A regular calling about a routine matter doesn't need a lead-capture flow, and sending them one feels like a downgrade in service.
- One nudge, maximum. No reply in 30 minutes → one follow-up → stop. Don't be the business that texts someone four times about a missed call.
- Always leave a human escape hatch on the AI callback. "Let me have someone call you back" has to work, every time, no dead ends.
- Review transcripts weekly for the first month. You will rewrite the AI's scripts at least twice in the first few weeks. That's normal, and it's exactly why you review them.
Why I keep building this particular stack
The pattern that keeps showing up across clients: the business already paid for the lead (ads, SEO, word of mouth, a truck with their number on the side) and then loses it to something completely fixable — a phone that rings into a voicemail box nobody checks. Fixing that is unglamorous compared to "we built you an AI agent," but it's the highest-ROI thing I build for almost every local-business client, and it's usually the project that earns me the trust to build the bigger stuff afterward.
If you're weighing whether this is worth building for your own business (or a client's), the fastest gut check is the math at the top of this post. Run your own numbers before you build anything — it tells you exactly how urgent this actually is.
I run Null Studio, where we build this stack — textback, AI callback, receptionist, A2P registration, CRM wiring — as one project for local businesses.
Top comments (0)