AI Front Office
AI receptionist, online booking, dispatcher routing, and B2C win-back for service businesses.
Meet the agent
This app ships with a production AI agent — permission-aware, tenant-scoped, audit-logged, and governed by the Agent Development Lifecycle.
AI Front Office Receptionist
Greets, qualifies, books appointments, and routes to a human dispatcher when needed.
Overview
AI Front Office handles inbound chat, SMS, and voice from customers of service businesses (a homeowner calling a plumbing co, a customer booking a salon appointment). The AI receptionist qualifies the request, books an appointment, and hands the booked work to the operations system (Field Service Suite or ServiceTitan). Includes Speed-to-Lead SLA tracking, online booking widget on a dedicated tenant subdomain, dispatcher review queue, B2C win-back campaigns with TCPA-compliant consent handling, and Call Coach AI scoring. AI Front Office is the B2C front desk for the tenant business's own customers.
An AI receptionist that answers web chat and SMS, turns conversations into scored booking requests, and keeps dispatchers in control with shadow and approval modes.
Highlights
Capabilities
AI Receptionist
-
Public web chat widget on the tenant subdomain and dedicated {slug}-book.workforce.mangoapps.com host
-
Embeddable iframe widget for the tenant's own website (embed key + origin allowlist)
-
Inbound SMS receptionist via Twilio
-
Pre-LLM topic gate that rejects off-scope questions and jailbreak attempts before spending a token
-
Answers common questions from the FAQ, documents and videos you add in the app's Knowledge Base (off until you switch it on)
-
Brandable widget — your logo, colors, welcome message, and submit-button label
-
Bilingual English + Spanish auto-detect (opt-in)
Receptionist Modes & Safety
-
After-hours-only mode (humans handle business hours; AI engages outside them)
-
Shadow mode (logs everything; booking requests kept out of the review queue and all alerts)
-
Dispatcher-approves-all mode (manual review of every booking)
-
Sentiment-based auto-escalation on sustained visitor frustration
-
Human takeover — dispatcher replies into an escalated web-chat / SMS thread
-
Instant dispatcher alert the moment the AI escalates a conversation to a human
-
Admin Playground (preview chat on the real agent + tools; side effects stubbed)
Booking & Dispatch
-
Online booking widget (static form, no AI conversation required)
-
Booking requests with confidence scoring + auto-book eligibility
-
Dispatcher routing rules (service-type / zip / urgency)
-
Service-area validation against the tenant's coverage zones
-
Capacity-aware booking against the day's committed Field Service job load (opt-in)
-
Auto-match a booking to an existing customer by phone, email, or address (auto-creates one when new)
-
One-click conversion to a Field Service work order (or ServiceTitan Job)
Speed-to-Lead & Autonomous Follow-up
-
Speed-to-Lead SLA tracking with warning + breach alerts (business-hours-aware clock)
-
AI first-touch backstop (SMS or voice) when no human responds by the SLA warning threshold
-
Missed-call text-back when an inbound call ends unanswered
-
Booking confirmation / decline SMS to the customer
-
Urgent-booking mobile push to dispatchers the moment an emergency job lands
-
Post-job thank-you text on work-order completion, with an optional review request
-
External lead webhook (POST /ai_front_office/leads, key-authenticated)
Win-Back & Call Coach
-
Audience builders (dormant customers, declined estimates, abandoned conversations — chats, texts and calls that ended without a booking)
-
AI-drafted SMS / email win-back outreach copy
-
Win-back campaign lifecycle (create / launch / pause / archive)
-
Call Coach AI scoring of each phone call on five fixed axes (greeting / diagnosis quality / booking attempt / objection handling / escalation appropriateness) plus a missed-opportunity flag
-
Manager scorecards against your own rubric (how chats and texts are scored — rubrics don't change the AI's scoring) + AI-generated practice drills
Compliance & Safety
-
TCPA consent audit trail + STOP-keyword opt-out with suppression list re-checked at send time
-
Outbound quiet-hours window (tenant-local, TCPA) — no AI calls, and win-back, post-job and non-urgent booking-confirmation texts wait until it ends
-
Per-phone outbound-call rate caps across all trigger kinds
-
Hallucination scan on AI responses (range-only quotes, no fabricated prices)
Reporting & Insights
-
Front-office analytics dashboard (booking rate, AI-resolution rate, escalation rate) with CSV / PDF export
-
Daily digest email — what the AI answered and booked yesterday (opt-in)
-
Win-back campaign analytics — messages sent, jobs booked, and estimated spend
Limits & Specs
-
Default Speed-to-Lead SLA: 5 minutes (admin-configurable)
-
Max LLM tokens per conversation: 50,000 — agent escalates on cap
-
Per-voice-call cost cap: 200¢ (admin-configurable)
-
Minimum geocode confidence for auto-book: 0.85
-
Channels: Web chat, SMS (Twilio), voice (Vapi), embed iframe, REST API
-
Pricing: Pay-as-you-go in packs of front desks that follow your count (no license required)
Use cases
FAQ
Web chat and inbound SMS work out of the box. Phone voice (a Vapi assistant on your mapped number) and in-widget WebRTC voice are off by default — they're turned on with the Voice — phone and Voice — browser mic settings, and need Vapi set up by MangoApps support plus a mapped phone number or web channel.
Shadow mode is on by default. The AI receptionist runs and logs every conversation, but the booking requests it creates are kept out of the pending-review queue and trigger no dispatcher alerts — you review them deliberately under the queue's Shadow filter — so you can evaluate answer quality before going live.
On by default, it lets your humans handle business-hours contacts while the AI receptionist only engages outside your configured business hours — so it augments the front office instead of replacing it.
Set a lead webhook key in settings and POST leads to /ai_front_office/leads (site forms, Zapier, lead-form bridges). They land in the dispatcher queue with the Speed-to-Lead clock running. A blank key disables the webhook (fail-closed).
Yes. With the Push approved bookings to ServiceTitan setting on and a connected ServiceTitan integration, approved bookings push to ServiceTitan as Jobs instead of creating Field Service Suite work orders (the two conversion paths are mutually exclusive). Without ServiceTitan, an approved booking converts to a Field Service Suite work order in one click.
Yes. The AI Front Office Assistant in Ask AI answers dispatchers' questions about the booking queue and the conversations the receptionist escalated, and approves, declines, takes, marks contacted or marks handled, each after a confirmation. AI Front Office admins also get the switches, the last-30-day figures and the Win-Back list. It is separate from the receptionist your customers talk to.
It's pay-as-you-go with no license decision — it's available to every tenant and billed monthly in packs of front desks that follow your count: the pack moves up as you add desks and down on the 1st when you have fewer, and the price per desk falls as packs grow. Zero front desks costs $0, so you only pay once a desk is live.
AI Front Office is the external front desk for a service business's own customers — a homeowner booking a plumber, a caller asking to schedule an appointment — turning inbound chat, SMS, and voice into booked work orders. Service Desk is the internal IT help desk for your own employees (ticketing, knowledge base, SLA tracking). Different audiences, different jobs; you can run both.
Yes. Win-back and outbound messages check a consent/suppression list that is re-verified at send time (the list can change between enrollment and send), honor STOP-keyword opt-outs, and respect a tenant-local quiet-hours window (default 21:00–08:00). Outbound voice adds a per-phone attempt cap (default 3) across all trigger kinds inside a rolling window (default 7 days).
There's a hard cap of 50,000 LLM tokens per conversation and a per-voice-call cost cap (default 200¢, admin-configurable) — hit either and the agent escalates to a dispatcher instead of burning more. A pre-LLM topic gate rejects off-scope and jailbreak prompts before spending a token, and a hallucination scan keeps AI replies to range-only quotes with no fabricated prices.
