AI Front Office App Overview

AI Front Office

1. What it is

AI Front Office handles inbound customer communication for service businesses — a homeowner texting about a plumbing leak, a caller booking a salon appointment, a website visitor requesting an estimate. The AI receptionist qualifies the request, captures a booking, and hands it to your dispatcher or operations team.

• Enablement: AI Front Office is a pay-as-you-go app, available to every tenant with no license decision. Usage is billed per front desk. An admin enables channels and safety modes from Apps → AI Front Office → Settings.
• What it is not: AI Front Office is the B2C front desk for your customers. It is not Mango GTM, which handles B2B sales and marketing for your own pipeline.

2. Standing it up

1. Go to Apps → AI Front Office → Settings.
2. Set your business hours. The receptionist uses these to decide when to handle calls itself versus defer to humans. The default is Monday–Friday 8 AM–6 PM Pacific, Saturday–Sunday closed. Choose your timezone and adjust each day’s open and close times.
3. Review the safety toggles. Three are on by default:
   • After-hours only — the AI only handles inbound during after-hours; humans handle business hours.
   • Shadow mode — the AI runs end-to-end but its bookings are saved as Shadow: kept out of the pending-review queue and every dispatcher alert, and listed under Shadow in the queue’s Status filter. The receptionist tells the visitor their request is recorded instead of promising a call, and gives your contact details (the Customer contact card in Settings) so they can confirm a time.
   • Dispatcher approves every booking — once Shadow mode is off, every booking enters the dispatcher queue for review regardless of the AI’s confidence.
4. Enable at least one channel. Web chat is the lowest-risk starting point. SMS requires a Twilio number entered as the SMS number in the Phone numbers — voice & SMS card of Settings. Voice requires a Vapi assistant and a Twilio phone number binding (enter the IDs in the same card).
5. Turn on the AI Receptionist master switch (AI Receptionist (master switch)). Without it, no channel processes inbound messages.
6. Confirm it worked. Open the Playground tab and send a test message. The Playground uses the same AI agent and tools as production but does not create real bookings or page dispatchers.

If Field Service Suite is not enabled, the Settings page warns you. The receptionist’s lookup tools depend on Field Service data (customers, service areas, availability), and approved bookings convert to Field Service work orders.

3. How it fits together

Conversations and booking requests — An inbound message (chat, SMS, or voice) creates a conversation. As the AI qualifies the customer’s need, it may create a booking request — a structured intent before a work order exists. A conversation can end without a booking (the caller just had a question, or the AI answered from the knowledge base). A booking request can also arrive without a conversation, through the online booking widget or the lead webhook.

Example: a homeowner texts “my water heater is leaking.” The AI opens a conversation on the SMS channel, asks for the address and urgency, and creates a booking request with urgency Emergency and service type Water heater. The booking enters the dispatcher queue as Pending review.

Booking lifecycle — A booking request moves through: Pending review → Dispatcher approved (or Auto-approved) → Converted (a Field Service work order or ServiceTitan job is created). Dispatchers can also decline with a reason. Shadow-mode bookings are kept out of the pending-review queue and every dispatcher alert (they are listed under the queue’s Shadow filter) and are excluded from analytics.

Channels — Four inbound channels, each independently toggled:

• Web chat — public widget on your booking subdomain.
• SMS — routes inbound texts to the AI receptionist via Twilio.
• Voice — phone (Vapi) — inbound calls via Vapi and Twilio.
• Voice — browser mic — visitors tap “Talk now” on the chat widget for a browser-mic conversation.

Dispatcher routing rules — Rules that auto-assign incoming booking requests to specific dispatchers based on service type, ZIP code, and urgency. Rules are evaluated highest priority first; the first match wins. Empty fields on a rule act as wildcards.

Safety modes — Three graduated controls for rollout:

• After-hours only keeps humans in the loop during business hours and lets the AI handle after-hours intake.
• Shadow mode lets the AI run fully while keeping its bookings out of the review queue and every dispatcher alert — useful for evaluating quality before going live. Read them under the queue’s Shadow filter, and the conversations behind them under Shadow captures on the Conversations page. Escalations still reach a dispatcher.
• Dispatcher approves every booking requires a human to approve every booking before it converts to a work order.

Turn these off one at a time as you build confidence in the AI’s responses.

Autonomy dial — AI Front Office also participates in the platform-wide AI autonomy setting (the same Off / Approve / Auto dial used across AI solutions; it is the AI mode card at the top of Settings). A tenant that has never touched the dial keeps full autonomy by default. Dialing it down to Approve or Off means it takes no bookings of its own: inbound chats and texts are handed to a dispatcher, and on a phone call it can only pass the caller to a person. On Approve, the outbound calls, follow-up texts and win-back it would send wait on the AI Console under “AI actions to approve” for a person to confirm; on Off it sends none.

4. Running it

Reviewing the dispatcher queue

1. Go to Apps → AI Front Office → Booking Requests.
2. The default view shows Pending review bookings, sorted by urgency (emergency first). Switch the sort to Newest or Speed-to-Lead alerts first using the dropdown. The Status filter lists the rest: Awaiting first contact, Auto-approved, Dispatcher approved, Converted, Declined, Expired, Shadow, and All statuses.
3. Open a booking to see the full conversation transcript, activity log, SLA status, similar prior bookings, and any outbound call attempts.
4. Click Approve to create a Field Service work order (or push to ServiceTitan if that integration is enabled). The work order has no technician yet: it waits in the Unassigned column of Field Service’s Dispatch board for the office to assign one. Click Decline and select a reason. If you have reached the customer but are not ready to decide, click Mark contacted: it records first contact and clears the booking’s Speed-to-Lead alerts, and the booking stays in the queue until it is approved or declined, or until it expires if Auto-expire un-actioned bookings after (days) is set.
5. Dispatchers can also create bookings manually via New booking request for walk-ins, phone calls, or leads that arrived outside the AI channels.

Bulk decline is available from the list view — select bookings, choose a decline reason, and submit. Bulk actions are capped at 100 records per submit.

Reviewing conversations

Go to Apps → AI Front Office → Conversations to see every intake conversation. Each row shows the channel, state, visitor info, and whether a booking was created. Open a conversation to read the full transcript, see the escalation reason (if any), and view linked scorecards. Once you have dealt with a conversation the AI handed to a person, click Mark handled: it stops counting as waiting and every dispatcher’s alert for it clears. It is marked booked if a real booking exists for it; otherwise it is closed, and a closed chat or text thread reopens when the customer writes again.

Monitoring Speed-to-Lead

When Speed-to-Lead monitor is on, the system tracks time from booking creation to first dispatcher contact. The default SLA window is 5 minutes (Speed-to-Lead (minutes) under Limits and timing). A warning fires at 25% remaining; a breach fires at 100%. Notifications go to the assigned dispatcher or, if none is assigned, to users in the configured dispatcher roles. By default, the SLA clock pauses outside business hours (an after-hours booking deadlines shortly after next open). Turn off Speed-to-Lead counts business hours only for 24/7 wall-clock SLA.

Using Call Coach

When Call Coach scoring is on, the AI scores each phone call the receptionist answers on five fixed axes — greeting, diagnosis quality, booking attempt, objection handling, and escalation appropriateness — and flags a missed booking opportunity. Chats and texts are not scored automatically: an AI Front Office admin scores them (open a conversation and click Add admin score). An admin scorecard starts from the active rubric’s criteria, or from greeting, diagnosis, booking attempt, and objection handling when no rubric is active. Admins author rubrics from Call Coach → More actions (⋮) → Manage rubrics; only one rubric per business can be active at a time. Rubrics set the criteria admins score against — they don’t change how the AI scores, and rubric weights are not applied to any score yet. Scorecards are created by the AI or by an admin, with an overall score from 0 to 100.

Managing Win-Back campaigns

When Win-Back campaigns is on, the Win-Back tab appears. Campaigns target dormant customers, declined estimates, abandoned intakes, overdue maintenance, or unscheduled recurring service. Each campaign sends outreach via SMS, email, or voice. Enrollments follow strict TCPA consent handling — no message is sent until consent is captured, and every send checks the platform suppression list. Campaign statuses are draft, active, paused, completed, or archived.

Asking Ask AI about the queue

Dispatchers can ask Ask AI about AI Front Office — the AI Front Office Assistant answers from the queue itself: how many booking requests are waiting for review, which are emergencies or past their Speed-to-Lead SLA, one customer’s booking request, and which conversations are waiting for a dispatcher. It can also approve, decline (with a reason), take, mark contacted or mark a conversation handled — each shows a confirmation card first, and runs the same action as the button on the page. AI Front Office admins can also ask about AI mode and the switches, the last-30-day figures and Win-Back campaigns. It is separate from the AI receptionist your customers talk to.

Using the Knowledge Base

When Knowledge Base is enabled, admins can author FAQ articles, upload documents (PDF, DOCX, TXT — up to 50 MB), and upload videos (up to 500 MB, auto-transcribed). The AI receptionist searches these entries before escalating to a dispatcher, so it can answer questions like “do you service this brand?” or “what’s the warranty?” without human help.

5. Settings

All settings are at Apps → AI Front Office → Settings unless noted otherwise.

Setting	Default	What it changes
AI mode	Auto	The Off / Approve / Auto dial in the card at the top of Settings, saved with that card’s own Save button. Below Auto the receptionist takes no bookings of its own: chats and texts go to a dispatcher, and on a call it can only pass the caller to a person. On Approve its outbound calls, follow-up texts and win-back wait on the AI Console under “AI actions to approve”.
AI Receptionist (master switch)	Off	Master switch for the AI receptionist across all channels. Nothing processes without it.
Web chat	Off	Enables the public web chat widget on your booking subdomain.
SMS	Off	Routes inbound SMS to the AI receptionist. Requires a Twilio SMS number in the Phone numbers — voice & SMS card.
Voice — phone (Vapi)	Off	Inbound phone calls via Vapi and Twilio. Requires the phone-number binding in the Phone numbers — voice & SMS card.
Voice — browser mic	Off	Visitors can start a browser-mic conversation from the chat widget. Requires a web channel binding.
After-hours only	On	AI receptionist only handles inbound during after-hours; humans handle business hours.
Shadow mode	On	AI runs end-to-end, but its bookings are kept out of the review queue and every dispatcher alert (listed under the queue’s Shadow filter), and the receptionist tells the visitor their request is recorded.
Dispatcher approves every booking	On	Once Shadow mode is off, every booking enters the queue for review regardless of AI confidence.
Speed-to-Lead monitor	Off	Tracks time-to-first-contact and alerts dispatchers when the SLA is at risk.
Speed-to-Lead (minutes)	5	The SLA window in minutes. Warning at 25% remaining, breach at 100%.
Speed-to-Lead counts business hours only	On	SLA clock pauses outside business hours. Turn off for 24/7 operations.
Speed-to-Lead AI first touch	Off	When no human contacts the lead by the warning threshold, the AI sends the first-touch SMS or call itself.
Online booking widget	Off	Public booking form on your tenant subdomain (no AI conversation required).
Win-Back campaigns	Off	SMS, email, or voice outreach to dormant customers, declined estimates, and more.
Call Coach scoring	Off	The AI scores each phone call on greeting, diagnosis quality, booking attempt, objection handling, and escalation appropriateness; AI Front Office admins score chats and texts against your rubric.
Knowledge Base	Off	Gives the AI receptionist a tool to search your authored FAQ and document content before escalating.
Bilingual (English + Spanish)	Off	The receptionist detects the caller’s language and replies in kind. When off, it stays in English.
Mobile push for emergency / same-day bookings	On	Push notification to dispatchers when an emergency or same-day booking lands. Skipped for shadow bookings.
Escalation alerts	On	Alert dispatchers the moment the AI escalates a conversation to a human.
Sentiment-based escalation	On	Escalates to a dispatcher when the visitor shows sustained frustration.
Sentiment escalation streak	2	Consecutive negative visitor turns before sentiment escalation fires.
Daily digest email	Off	Daily email to admins summarizing what the AI answered, booked, and escalated the previous day. Skips quiet days.
Capacity-aware booking	Off	Availability windows and auto-booking respect the day’s committed job load vs. the daily capacity ceiling. No effect for ServiceTitan-sync tenants.
Daily job capacity (jobs/day)	12	Ceiling for capacity-aware booking (jobs per day).
Outbound voice — master switch	Off	Master switch for AI-placed outbound calls (win-back voice, decline follow-up, abandonment recovery).
Outbound voice — auto follow-up on declined estimates	Off	Schedule a follow-up call when a booking is declined because the customer declined the estimate.
Outbound voice — auto recovery on abandoned conversations	Off	Schedule a recovery call when a chat or SMS conversation is abandoned and a phone number is on file.
Missed-call text-back	Off	SMS the caller when an inbound call ends unanswered, so the lead isn’t lost (needs the Twilio number’s call-status callback — ask MangoApps support).
Booking confirmation SMS	Off	SMS the customer when their booking is approved, declined, or expires unreviewed.
Post-job happy text	Off	Thank-you SMS to the customer when their work order completes. Adds a review ask when a review link is configured.
Push approved bookings to ServiceTitan instead of Field Service Suite	Off	Push approved bookings to ServiceTitan as jobs instead of creating Field Service work orders. Requires an active ServiceTitan integration.
Allowed roles	Manager, Admin, Super Admin	Roles that can access AI Front Office (review queue, conversations, outcome events, Playground). Set in the Role access card.
Administrative roles	Admin, Super Admin	Roles that can configure AI Front Office (settings, routing rules, win-back, call coach). Set in the Role access card.

6. More help

• AI Front Office FAQ — specific setup, operating, and troubleshooting questions.
• Ask AI — the AI Front Office Assistant answers dispatchers from the queue itself and acts on it after a confirmation (see “Asking Ask AI about the queue” above); questions about how the app works are answered from these articles.