# P13 — The Full Business, Explained Like a Partner

**Updated:** 30 August 2026 · Read this FIRST, before every other document.
**Who this is for:** Tasfin, starting from zero knowledge of voice AI. Easy language, short sentences, no jargon without explanation.

Think of this document as me — your business partner — sitting with you over coffee and explaining the whole thing, start to end. One thing to say before anything else: **this project has no product build. We write no software.** We configure other companies' voice-AI platforms and sell the result as a monthly service. That is why it fits the portfolio as "opportunistic pilot — anytime (no build)": it never competes for a dev build slot, only for your configuration hours.

---

## Part 1 — What is this business?

### 1.1 The story in 60 seconds

Every local business — a clinic, a salon, a restaurant, a real-estate office — lives and dies by its phone. And every one of them misses calls: at lunch, after hours, when the receptionist is with a customer, on Fridays. A missed call is usually a missed booking, and the caller does not call back — they call the next place on Google.

In the last two years, "voice AI" became good enough to answer those calls. Not a robotic menu ("press 1 for..."), but a natural voice that says *"Thanks for calling Smile Dental, how can I help?"*, understands the caller, **books the appointment into the real calendar**, answers common questions, takes a message, and hands off to a human when needed. It works 24/7, never sick, never rude.

Companies like **Vapi, Retell AI, Synthflow and Bland** sell the underlying machinery per minute — roughly **$0.07–0.15 per call minute** all-in. But a salon owner in Dhaka, Sharjah or Texas cannot use these platforms: they are developer tools with dashboards full of "webhooks", "LLM temperature" and "endpointing". Someone has to configure the assistant, write the prompts, connect the calendar, test the calls and watch the quality.

**We are that someone.** We configure the receptionist on the platform, connect it to the client's calendar and CRM, and charge the client **$99–199/month, setup included**. The platform bills us pennies per minute; the client pays us a flat, predictable fee for a working receptionist. The gap between the two is our margin.

### 1.2 A picture of where we sit

```
Caller                         The platform (rented)                Client's world
┌──────────┐   ┌───────────────────────────────────────┐   ┌─────────────────────┐
│ Customer │ → │ Phone number → Voice AI assistant     │ → │ Calendar (bookings) │
│ calls or │   │ (Retell/Vapi-class infrastructure —   │   │ CRM (leads)         │
│ WhatsApp │   │  WE configure it, THEY run it)        │   │ WhatsApp/email      │
└──────────┘   └───────────────────────────────────────┘   │ (summaries, alerts) │
                          ↓ when needed                     └─────────────────────┘
                   Transfer to the owner's real phone
```

Important: we do **not** build or host any of the voice technology. The platform runs the calls. We are the configuration, integration and quality layer — the same relationship an agency has with Meta when it runs ads: Meta owns the machine, the agency owns the client and the result.

### 1.3 Why us? Why now?

- **Why us:** ThemeTags already serves exactly these businesses. We built a restaurant system and a CRM for local SMB clients; the digital-marketing arm talks to local business owners every week. The first three pilots come from names already in our phone. Plus: a Dhaka cost base means the ongoing care that makes this service good (weekly call reviews, prompt tuning) costs us a fraction of what it costs a US competitor like Smith.ai ($95–800/mo) or Slang.ai ($399–599/mo, restaurants only).
- **Why now:** the platforms matured (sub-second response, natural interruption handling, native calendar booking tools), per-minute prices fell to where a $99–199/mo retail price can carry them **if we cap minutes** (Part 4 — this is the one number that decides everything), and US SMB awareness is high because AI receptionists are being advertised at them daily.
- **Why "opportunistic":** no code, no build slot, ~2 weeks of part-time configuration per pilot. If it works, it becomes a compounding monthly-revenue line and a wedge for the CRM/portal products. If it fails, we lose a few founder-weekends, not a build quarter.

---

## Part 2 — The missed-call money math (the pitch, and why SMBs pay)

This is the arithmetic we will show every prospect. Numbers are **illustrative estimates** — we fill in the client's real numbers during the pitch.

| Line | Salon (BD/Gulf) | Dental clinic (US) |
|---|---|---|
| Calls received per day | ~15 | ~30 |
| Share missed (after-hours, busy, lunch) — industry claims range 20–40% | ~25% → 4/day | ~25% → 7/day |
| Missed calls that were bookings | half → 2/day | half → 3.5/day |
| Average ticket | $15 | $180 |
| **Lost revenue per month (26 days)** | **≈ $780** | **≈ $16,000** |
| Our fee | $99/mo | $199/mo |

Even if the real miss rate is half our estimate, the service pays for itself many times over. The US clinic case is absurdly favourable — which is why US clients via the LLC are the margin engine, while BD clients are the learning lab. The sales hook is not a spreadsheet, though — it is the **missed-call audit**: we call the prospect's business after hours, record what happens (ring... ring... voicemail or nothing), and play it back to the owner. See `SALES_PLAYBOOK.md`.

---

## Part 3 — How do we earn, exactly?

One stream, deliberately simple:

| What we sell | Price | What the client gets |
|---|---|---|
| **Managed AI receptionist** | **$99–199/mo, setup included** (tiers by included minutes and features — `PRICING.md`) | A dedicated number (or their existing number forwarded), a configured voice assistant for their vertical, calendar booking, lead capture to their CRM/WhatsApp, daily summaries, escalation to a human, and us watching quality every week |

What the money covers on our side, per client per month (planning estimates, researched 30 Aug 2026 — `PRICING.md` has the full table):

| Cost | Estimate |
|---|---|
| Platform + AI minutes (~250–800 min/mo depending on tier) | $25–85 |
| Phone number rental | $2–5 |
| Our ongoing labour (QA reviews, tweaks) | target ≤ 2h/mo |

Gross margin target: **≥ 55% after infrastructure** at every tier. The margin killer is not the per-minute cost — it is **our time**. Which is why the portfolio kill criterion is binding and quoted verbatim everywhere it matters: **delivery time >10h/client/mo (service margin death) → drop or productize config into templates.** At $99/mo, 10 hours of monthly labour is $9.90/hour — worse than any freelance work. The whole operating design (templates, checklists, weekly batch QA) exists to keep steady-state labour under 2 hours per client per month.

### Why zero build is a feature, not a compromise

1. **The platforms already won.** Vapi/Retell-class infrastructure has hundreds of engineers on latency, telephony and speech. Building our own stack would be a multi-year, funded-startup project to arrive at parity with something we can rent for $0.07/min.
2. **The value is local anyway.** The client is not buying "voice AI" — they are buying *their* receptionist: their menu, their doctors' schedules, their tone, their escalation rules, in their language. That is configuration and relationship work — agency work. Our kind of work.
3. **Productization path exists without code.** If the pilot works, we don't build software — we build a **template library**: one battle-tested config per vertical (clinic, salon, restaurant, real-estate) that turns a 2-week setup into a 2-day setup. That library is the answer to the kill criterion and the real asset we accumulate (`SPEC.md` D4).

---

## Part 4 — Margin discipline (the honest part)

The research produced one loud warning, and I would rather you hear it now than in month three:

**At realistic all-in platform costs (~$0.10–0.16/min), an *uncapped* $99/mo client is underwater.** A busy restaurant doing 25 call-minutes a day consumes ~750 min/mo ≈ $75–120 of pure infrastructure — the entire fee, before one minute of our labour. The managed competitors know this: Smith.ai bills per call ($1.60–1.90), Goodcall caps callers per tier, Slang.ai charges restaurants $399+.

So the pricing rules (binding, detailed in `PRICING.md`):

1. **Every tier includes a minute cap** ($99 → ~250 min; $149 → ~500; $199 → ~800) with an overage rate. No unlimited plans, ever.
2. **Margin floor: infrastructure ≤ 40% of the monthly fee.** A client that persistently breaches it moves up a tier or gets a re-quote — scripted, not awkward, because it means the receptionist is working.
3. **Labour floor: ≤ 2h/client/mo steady state**, tracked in `PROGRESS.md`; the 10h kill line is the hard stop.
4. **Premium voices and premium LLMs are Pro-tier features** — the budget stack (fast small model + standard voice) is genuinely good in 2026 and costs a third as much.

---

## Part 5 — What could go wrong?

| Risk | Truth | Our protection |
|---|---|---|
| "Customers hate talking to robots" | Partly true — they hate *bad* ones: slow, interrupting, looping | We configure barge-in (caller can interrupt naturally), instant human transfer on request or frustration, and we QA 20 scripted calls before any client goes live (`TEST_PLAN.md`) |
| Minutes eat the margin | Real — see Part 4 | Caps + overage + margin floor rule; monthly per-client cost report |
| Our time eats the margin | The classic managed-service death | Template library from day 1; the 10h/client/mo kill line is binding |
| BD/UAE telephony is restricted | **True and verified** — UAE blocks VoIP-style numbers for this use; BD numbers are heavily regulated. `COMPATIBILITY.md` has the honest matrix | BD/Gulf pilots run on **WhatsApp calling and/or call-forwarding designs**, not local VoIP numbers; US market has zero such friction — the LLC channel is the growth market |
| Bangla voice quality isn't there yet | Partly true — Bangla STT/TTS lags English and Arabic | Pilot honestly: English-first for US, English/mixed for Gulf; Bangla assessed in P2 with real test calls before we promise it to anyone (`COMPATIBILITY.md`) |
| Platform raises prices or dies | Possible — this market is consolidating | Primary + fallback platform chosen up front (`SPEC.md` D1); our assets (prompts, call flows, integrations) are written portably |
| Recording/consent law trouble | Serious — UAE is all-party-consent with criminal penalties; 12+ US states are two-party | Recording disclosure in the greeting, per-market rules in `SECURITY.md`, recordings off by default in UAE |
| Client cancels after novelty wears off | The retention metric exists precisely to test this | Success metric is retention, not signups: **3 pilots → 2 retained at full price.** If they don't retain, the market told us the truth cheaply |

---

## Part 6 — First steps (this week, founder only)

1. Read `FOUNDER_GUIDE.md` — after 25 minutes you can explain how a voice agent works and hold a pilot conversation without fear.
2. Pick the 3 pilot candidates from our existing client list (restaurant-system client + CRM client + one marketing client — warm names only). Item O1 in `SPEC.md`.
3. Open the platform account (task `F-01`), buy a US test number (~$1.15/mo), and make your first test call to your own assistant the same evening. Total cost to start: **under $20**.
4. Run the missed-call audit on the 3 candidates (call them after hours; record what happens).

No developers. No servers. No build slot. The clock that matters: 2 retained clients by ~P8 (end of October 2026) or we stop.

---

## Part 7 — Your reading path (in order)

| Order | Document | Why | Time |
|---|---|---|---|
| 1 | **This document** | The whole business in one sitting | 15 min |
| 2 | `FOUNDER_GUIDE.md` | How voice AI actually works — the latency chain, the jargon, the cost anatomy | 25 min |
| 3 | `SALES_PLAYBOOK.md` | The pilot pitch, the audit hook, the objection answers | 15 min |
| 4 | `PRICING.md` | The tiers, the unit-economics math, the margin floor | 15 min |
| 5 | `COMPETITOR_ANALYSIS.md` | Who charges what; where a Dhaka-cost agency wins | 10 min |
| 6 | `ROADMAP.md` + `TASKS.md` | The P1–P8 pilot calendar and your task list | 10 min |
| 7 | `SPEC.md` | Decisions D1–D10, vertical playbooks, open items O1–O6 | 25 min |
| — | Everything else (`DEPLOYMENT`, `TEST_PLAN`, `TESTING`, `API_CONTRACT`, `SECURITY`, `COMPATIBILITY`, `RELEASE_CHECKLIST`, `MARKETING`, `FEATURES`, `PROGRESS`) | Reference — used during pilot delivery, not before | as needed |

---

*Partner-to-partner closing note: this is the smallest bet in the portfolio — no build, warm leads, under $20 to start, and a binding kill line. The one discipline it demands is saying no: no uncapped plans, no custom builds for pilot clients, no more than 2 steady hours per client. If we keep those three "no"s, this is a quiet compounding revenue line and the cheapest market research we will ever do on US SMB sales through the LLC.*