itpp-infrastructure/projects/front-desk-voice-agent.md

Sho'Nuff Front-Desk — AI Voice Lead Qualification & Booking

Saved: 2026-08-24 Status: Future Project — Competitive Teardown + Build Path Category: SaaS Product / Revenue Owner: IT Pro Partner (Sho'Nuff) Inspiration: Qexo.ai competitive teardown (2026-08-24)

---

The Competitive Trigger

Qexo.ai is AI voice agents for lead qualification + appointment booking, aimed at home-services / SMB verticals (solar, HVAC, home services, real estate, insurance, clinics). It is the third tool in a row we've teardown'd (after tranx.io and dozier.io) shaped as a single-purpose vertical tool on generic AI plumbing: narrow SMB wedge + clean before/after loop + evidence/confidence. The pattern is now a signal, not a coincidence.

Qexo's four-step loop:

1. Capture    — web/form intake catches the lead while intent is hot
                message + contact + consent attached to ONE record
2. Understand — "lead intelligence" scores intent / urgency / service-fit
                0–100 qualification score, flags missing details
3. Call & Book — with consent, voice agent calls OUT, gets context,
                books against live calendar, confirms address/access notes
4. Manage     — unified lead workspace (inquiry + transcript +
                qualification + appointment), human override

Why Qexo is clever (and what to steal):

• Full-context handoff — the outbound call inherits the web intake, never starts over. The caller already knows what the lead asked for.
• Transparent AI + consent — kills the FTC/legal objection up front. Voice agents calling out is the single most legally exposed move in this space; Qexo neutralizes it by design.
• "Not another inbox" — explicitly positioned as NOT a CRM. A unified lead workspace, not a login you'll ignore.

---

The Honest Read: We Already Own 90% of This

The gap is productization, not capability. We have every component running today:

Component	Existing Asset	Status
Outbound voice calling	shonuff-voice-caller, twilio-voice-calling skills (Twilio + ElevenLabs)	✅ Live
Inbound call handling	ai-receptionist.md (VoIPSimplicity Concierge)	📐 Designed, not built
Missed-call capture	missed-call-lead-recovery.md (Twilio SMS text-back)	📐 Designed, not built
Voice stack (STT→LLM→TTS)	voice-agent-deployment skill — Hermes Voice (xAI realtime) + Kokoro/faster-whisper open-source	✅ Live
Orchestration brain	Hermes Agent (personality, memory, tool routing)	✅ Live
Calendar	Rally family calendar + booking patterns	✅ Live

The missing 10% — the part that makes it a product rather than a demo:

1. A lead-record model that survives web → voice → calendar — one object holding intake message, contact, consent, qualification score, transcript, and appointment, all keyed to the same lead. This is Qexo's actual moat, and it's a schema problem, not an AI problem.
2. A qualification scorer — 0–100 intent/urgency/service-fit score with visible evidence (the lead's own words).

Neither is hard. Both are a weekend build on the infra we already run.

---

Build Path

Phase 1 — The Lead Record (the real moat)

Postgres table (single source of truth):

leads
  id, tenant_id, source (web/form/call/missed-call)
  contact_name, phone, email
  message (original inquiry, verbatim)
  consent_at (timestamp), consent_medium (form checkbox / verbal / none)
  qual_score (0-100), qual_evidence (json: flagged signals)
  service_fit, urgency, intent
  transcript (json, appended on call)
  appointment_id, appointment_status
  created_at, updated_at

Every downstream step (scorer, outbound call, calendar write) reads and writes this same record. The lead never starts over. This is the "full-context handoff" Qexo sells, reduced to a schema.

Phase 2 — Qualification Scorer

DeepSeek (via admin-ai) classifies the lead record into a 0–100 score with visible evidence:

• Intent — did they ask for a specific service, or just browse?
• Urgency — "as soon as possible" / "this week" / "just looking"
• Service-fit — does the inquiry match any offered service?
• Missing detail flags — no address, no timeframe, no budget signal

Output: a score + the exact phrases that drove it. The evidence layer is what makes it defensible against "AI made that up" — the customer sees the lead's own words backing the score.

Phase 3 — Outbound Call & Book

With consent on record, the voice agent calls out:

1. Pulls the lead record (never re-asks what the form already captured)
2. Confirms interest, fills the gaps the scorer flagged
3. Books against live calendar
4. Confirms address/access notes
5. Appends transcript + appointment to the lead record

Reuse shonuff-voice-caller (ElevenLabs professional male voice) + Twilio outbound. The open-source Kokoro/faster-whisper stack from voice-agent-deployment is the $0-cost alternative for beta.

Phase 4 — Unified Workspace ("Not another inbox")

Single view per lead: original inquiry + qualification score + transcript + appointment, human override everywhere. Not a CRM — a workspace where a lead either gets booked or gets a reason why not.

---

Positioning vs. What We Already Have

Product	Inbound	Outbound	Qualifies	Books	Key Differentiator
VoIPSimplicity Concierge (ai-receptionist.md)	✅ answers calls	❌	⚠️ basic	✅	Replaces IVR for existing VoIP customers
Missed-Call Recovery (missed-call-lead-recovery.md)	⚠️ missed calls	❌	❌	❌	SMS text-back within seconds
Sho'Nuff Front-Desk (this)	✅	✅ calls out	✅ scores	✅	Full-context handoff: web → score → outbound → calendar
Qexo.ai (competitor)	✅	✅	✅	✅	The benchmark we're matching

The Front-Desk is the top of the funnel the other two feed into. Missed-call recovery captures the lead; the Front-Desk qualifies and books it. These are three products on one voice stack, not three competing ideas.

---

Pricing (Premium, Value-Based — Never Undercut)

Modeled on Qexo's SMB wedge but priced like we own the infrastructure (we do):

Tier	Price/mo	Included
Solo	$99	1 voice number, web intake + scorer, 50 outbound calls, calendar booking
Pro	$299	3 numbers, 200 calls, multi-location, transcript archive, human-override console
Managed	$599+	White-label, agency resell, custom qualification rules, SLA

Undercut Qexo on unit economics, not headline price. Our marginal cost is near-zero (self-hosted voice stack, admin-ai tokens at cost). Qexo pays per-call infrastructure margins we don't.

---

Risks

1. FTC / TCPA on outbound AI calls — the single biggest exposure. Qexo's consent-first design is correct and mandatory. We mirror it: no outbound call without a recorded consent timestamp. debt-recovery-compliance skill already documents the TCPA/consent discipline; reuse it.
2. Voice quality at scale — Kokoro is good but not ElevenLabs. Start managed-tier on ElevenLabs, offer Kokoro for beta cost control.
3. Calendar write integrity — a wrong booking is a lost customer. The lead record must be the single writer; no side-channel calendar edits.

---

Open Questions

1. First vertical — solar/HVAC (Qexo's beachhead) vs. our existing warm markets (Debt Recovery Experts intake, VoIPSimplicity customers, Forefront Wireless)?
2. Calendar backend — Rally, or a dedicated booking calendar per tenant?
3. Consent capture — form checkbox (SMS/web) vs. recorded verbal consent (call). Both need a timestamped, auditable record.
4. Tenant model — multi-tenant from day one (agencies reselling to clients), or single-tenant until 3 paying customers?

---

Next Steps

1. Decide first vertical (open question 1) — this shapes every downstream choice
2. Build the lead-record schema (Phase 1) — the moat, and a pure schema task
3. Weekend spike: qualification scorer on 10 sample leads, verify evidence layer
4. Wire outbound call via existing shonuff-voice-caller + Twilio
5. Beta with 1–2 friendly businesses before any pricing commitment

---

Meta-Signal (worth remembering)

Three teardowns in a row — tranx.io, dozier.io, qexo.ai — are all single-purpose vertical tools on generic AI plumbing, each with the same shape: narrow SMB wedge, clean before/after loop, evidence/confidence layer. The pattern means the plumbing is commoditizing. The defensible layer is not the AI — it's the data model and the compliance posture (lead record + consent + calendar integrity). We already own the plumbing. The win is in the schema, the scorer evidence, and the consent design — not in out-building the AI.