Bidirectional Twilio ↔ Gemini Live
Twilio Media Streams hands audio directly into a Gemini Live WebSocket. Audio flows both directions in real time on a single persistent connection — no IVR tree, no script.
A Proposal for Roofing Operators · $5M–$20M Revenue
A live, conversational AI receptionist that answers every call.

The market for "AI receptionists" is now flooded with products that take a message and call it intelligence. MARK is not one of those products.
MARK is a live, real-time conversational AI built on a Twilio ↔ Gemini Live bidirectional voice pipeline. When a homeowner calls, MARK answers in under two rings, holds an actual back-and-forth conversation, lets the caller interrupt and redirect mid-sentence, books the inspection on a real calendar, writes the lead into your CRM before the caller hangs up, and texts the confirmation while the field rep gets a dispatch alert.
For a roofing operator doing $5M–$20M in annual revenue, the math is simple: one missed storm-day call is $10K–$15K of pipeline. One missed storm week is $1.5M–$2M of pipeline.
MARK Professional. ~1% of a $10M operation's revenue. Pays for itself in a single recovered call most months — and 50–100x over in any storm event.
This proposal benchmarks MARK only against other live, conversational AI phone agents — systems that hold real-time, back-and-forth conversations. We do not benchmark against human answering services, IVR phone trees, voicemail-to-text bots, or generic web chatbots. Those are different product categories at different price points.
"We are effectively turning away four out of five new customers when the weather turns."

Phone numbers entered wrong. Addresses missing. Notes that say 'call back about roof.' Reports go from useful to fiction in a single quarter.
Every inspection booked manually consumes 6–10 minutes plus a confirmation call plus a reminder text. At 40/week, that's a full clerical salary on logistics alone.
Storms roll through at 11 PM Sunday. Your office is closed. Your competitor with an AI receptionist is open. The homeowner is on whoever's schedule answers first.
Each one is already running on the FNS Tech stack — not a roadmap promise.
Twilio Media Streams hands audio directly into a Gemini Live WebSocket. Audio flows both directions in real time on a single persistent connection — no IVR tree, no script.
The caller can interrupt mid-sentence. MARK stops, listens, and pivots — the single most important quality marker of a real conversational AI vs. a script-reading bot.
While MARK is talking, it's writing the customer record, job, address, and lead source to your CRM via function-calling tools. Lead is fully entered before the caller hangs up.
MARK checks live calendar availability, confirms a slot in conversation, and writes the appointment during the call. Caller receives a confirmation text in their hand.
Pinned-preferences system stores your voice, vocabulary, FAQs, and personality. MARK doesn't sound like a generic robot — it sounds like your receptionist.
First sentence reaches the caller's ear while the rest is still being composed. Response gaps feel natural — not the 4–8 second dead silence most AI agents leave on the line.
Triple-tier routing: Gemini fast-path first, OpenClaw HTTP API fallback, Claude long-thinking fallback. If one stack has a regional outage at 2 AM, the call doesn't drop.
Caller texts a photo of damage mid-call. Gemini Vision analyzes it and MARK references it on the call. No SMB-tier competitor offers this natively.
A dedicated supervisor monitors heartbeat. If MARK crashes or wedges, the watchdog restarts with exponential backoff. 24/7 uptime in practice, not just promise.
Optional add-on: full custom site, deployed to Vercel, in under a week. Part of the FNS Tech stack — not a third-party reseller integration.
The pattern: cheap-end vendors don't do real barge-in or in-call CRM writes. Mid-tier does some of it. Enterprise matches MARK on individual features but not on the combination — and none have multi-LLM redundancy or native vision intake at any price.

Tropical storm just passed. Phone rings.
Thanks for calling Apex Roofing — this is Mark. Are you calling about damage from tonight's storm?
[Twilio Media Stream → Gemini Live · sub-second pickup · persona = Apex Roofing]
Yeah my whole back side got hit, there's shingles in the —
Got it. I'm sorry to hear that. Let's get you on the schedule first thing. Can I get the property address?
[Native barge-in — caller cut MARK off mid-question, MARK pivoted instantly]
1247 Maple Ridge Drive, Lakeland.
Got it. Insurance claim or out of pocket?
[In-call CRM write — customer record + property address created in Supabase]
Insurance, I haven't called them yet.
Perfect — we work with all the major carriers, and we can do the inspection before you file so we have photos for the adjuster. I have tomorrow at 10 AM or 2 PM, which works better?
[Live calendar check ran while MARK was speaking — slots returned in milliseconds]
10.
Locked in. I'll text you a confirmation now and your inspector's name. By the way — if you can send a quick photo of the worst spot to this number, I can pass it along to your inspector tonight so he knows what to expect.
[Appointment written · SMS confirmation dispatched · inspector alert queued · vision intake offered]
Lead in CRM. Appointment on calendar. Confirmation texted. Field rep alerted. Owner notified. Photo received and tagged.Zero human involvement.
| Vendor | Price | True barge-in | In-call CRM writes | Live calendar | Roofing-tuned | Photo intake | Multi-LLM redundancy |
|---|---|---|---|---|---|---|---|
| MARK | ⌜$1,497 → $5,997/mo⌟ ✓ Best-in-class value | Instant interrupt | Real-time logging | Direct scheduling | Deep industry model | Seamless capture | Triple-stack failover |
| Synthflow | ⌜$29 → $1,250/mo⌟ | Partial | Webhook | Cal.com | — | — | — |
| Retell AI | ⌜$0.07–$0.31/min⌟ | Partial | Webhook | Via integration | — | — | — |
| Vapi | ⌜$0.05/min + comp.⌟ | Partial | Webhook | Via integration | — | — | — |
| Bland.ai | ⌜Per-min, custom⌟ | Partial | Webhook | Limited | — | — | — |
| Air AI | ⌜~$3K–$10K/mo⌟ | Yes | Configurable | Configurable | — | — | — |
| AgentZap | ⌜$109 → $899/mo⌟ | Partial | Via integration | Via integration | Deep industry model | — | — |
| Goodcall | ⌜$66 → $208/mo⌟ | — | Limited | Limited | — | — | — |
| Rosie | ⌜$49 → $299/mo⌟ | — | Limited | Basic | — | — | — |
| Smith.ai | ⌜$292+/mo⌟ | Partial | Via integration | Via integration | — | — | — |
| Dialzara | ⌜$29+/mo⌟ | — | Limited | Basic | — | — | — |
| My AI Front Desk | ⌜$65 → $119/mo⌟ | — | Limited | Basic | — | — | — |
| OnCallClerk | ⌜$49+/mo⌟ | — | Limited | Basic | Trades-tuned | — | — |
↔ swipe to see full comparison
Sources · NextPhone AI Receptionist Cost 2026; Synthflow, AgentZap, Goodcall, Retell AI, Smith.ai, Dialzara, My AI Front Desk, Rosie public pricing pages; Ringly Voice AI Pricing 2026; CloudTalk Voice AI Cost 2026; Roofing Contractor magazine 2026 AI adoption report.
All tiers include unlimited calls, FNS Tech engineer-direct support, and the full live conversational stack.
$5M–$8M single-location operators · 100–300 calls/month
An essential command center for operators seeking foundational, always-on call capture.
Position · Above the SMB pack — Goodcall Scale ($208), Rosie Growth ($299), AgentZap mid (~$400).
$8M–$15M operators · 300–700 calls/month · multi-crew dispatch
Our most popular tier — the sweet spot for the $10M roofer driving growth and unlocking strategic advantage.
Position · Above AgentZap top tier ($899) and below boutique custom Retell/Vapi builds ($4K–$8K).
$15M–$20M+ multi-branch operators
The ultimate solution for complete operational dominance and future-proofing your enterprise.
Position · Squarely against Air AI enterprise and custom Retell/Vapi builds at $5K–$10K+/mo.

A two-person front desk covering business hours only is ~$4,000–$8,000/mo loaded (salary + benefits + overhead). MARK Core is less than the cheaper end of that and runs 24/7. MARK Professional matches the upper end and does work no human receptionist can do — vision intake, sub-second pickup, multi-LLM failover, bilingual without a second hire.
Assumes 30% miss rate · 30% recovery · $2997/mo Pro tier
Review 5 of your recent inbound calls. Draft conversation flows. Provision CRM access. Map integration touchpoints.
Voice tuning. Knowledge base build (services, pricing, service areas, FAQs). CRM tested in sandbox. Pinned preferences set.
MARK handles after-hours and overflow calls. Office staff handles business hours. Baseline metrics captured.
MARK is primary receptionist. Office staff on escalation only. Full reporting live.
Weekly tuning calls during the first 90 days. Monthly thereafter.


We are not reselling somebody else's platform with our logo on it. The voice pipeline, the CRM tool layer, the orchestrator, the watchdog — all our code. When you call support, you reach the engineer who wrote it.
Gemini Live for real-time voice. Claude for long-thinking reasoning. ElevenLabs for TTS fallback. Supabase for the data layer.
We built the persona system because every roofer's voice is different. We built storm-mode scaling because that's where the money is.
No offshore support tickets. No tier-1 script readers. The engineer on the line built the thing.
If your captured-lead count doesn't increase by at least 30% in the first 90 days, we refund the setup fee.
We listen to 5 of your recent inbound calls and walk you through what MARK would have done differently on each.
Using your actual call volume, close rate, and average job size — not the industry-mid-case numbers above.
Written, scoped, and priced for your operation.
Implementation begins within 5 business days.
The phone is the difference between a million-dollar storm week and a million-dollar storm week your competitor had.Let's make sure yours always gets answered.