AI Front Office FAQ
Answers to common setup and operating questions about AI Front Office.
For what the app is and how to set it up from scratch, see the
AI Front Office Overview.
Setup
What do I need to turn on before the AI receptionist will answer?
Two things: the master Enable AI Receptionist switch and at least one channel — Web Chat Channel, Inbound SMS Channel, or Voice Channel (phone). Both are off by default. Turn them on at Apps → AI Front Office → Settings. If you enable only the master switch but no channel, the receptionist has nowhere to listen and nothing happens.
What is shadow mode, and should I leave it on at first?
Shadow mode is on by default. The AI receptionist runs end-to-end and logs every conversation, but the booking requests it creates are marked as shadow and hidden from dispatchers in the default queue view. Leave it on during your first few days so you can read transcripts and evaluate answer quality before any real bookings reach your team.
What is after-hours-only mode?
After-hours-only mode is on by default. When it is on, the AI receptionist only handles inbound contacts outside your configured business hours. During business hours, it immediately routes the visitor to a dispatcher so your human staff stays in the loop. Default business hours are Monday–Friday 8 AM–6 PM Pacific, Saturday and Sunday closed. Set your own hours at Settings → Business Hours.
What does “Dispatcher Approves All Bookings” mean?
It is on by default. Every booking request enters the queue in Pending Review regardless of its confidence score. When you turn it off, high-confidence bookings that also pass the geocode-confidence floor and service-area check can be auto-approved without dispatcher review — but only if the receptionist’s autonomy dial (on the AI Console tab) is set to Auto.
Permissions and access
Who can see the dispatcher queue, and who can change settings?
Two role lists control access, both configurable at Settings → Roles:
- Allowed Roles (default: Manager, Admin, Super Admin) — can view booking requests, conversations, outcome events, and the Playground.
- Administrative Roles (default: Admin, Super Admin) — can open Settings, Win-Back, Call Coach, Knowledge Base, Routing Rules, and Analytics.
Platform admins and app admins always have full access regardless of these lists.
Day-to-day
How does a booking request get routed to a specific dispatcher?
Routing rules match on three dimensions — service type, zip code, and urgency — evaluated in priority order (highest priority first). The first rule that matches assigns the booking to that rule’s dispatcher. An empty field on a rule acts as a wildcard. If no rule matches, the booking stays unassigned and the entire dispatcher team sees it.
What are the Speed-to-Lead warning and breach thresholds?
The default SLA window is 5 minutes (configurable at Settings → Speed-to-Lead SLA). A warning fires when 25 % of the window remains (75 % elapsed). A breach fires when the window is fully expired. By default, the SLA clock pauses outside business hours. Turn Speed-to-Lead follows business hours off for a 24/7 wall-clock SLA.
How does the AI first-touch backstop work?
When Speed-to-Lead AI first touch is enabled and no human has made first contact by the SLA warning threshold, the AI sends the first touch itself. The default channel is SMS; you can switch it to Voice if outbound voice is also enabled. SMS marks the SLA as met. A voice call is scheduled but still subject to quiet-hours and rate-limit checks at dial time.
How do I embed the chat widget on my own website?
Go to Settings → Embed Widget, generate an embed key, and paste the provided <script> tag into your site’s HTML. The key is rotatable — generating a new one immediately invalidates the old one. To restrict which sites can host the widget, add their origins to Widget Allowed Origins; an empty list allows any origin.
When something looks wrong
I sent a message in the chat widget but the AI did not answer — why?
The most common causes:
- After-hours-only mode is on and you are inside business hours. The AI defers to humans during open hours. Check your configured business hours.
- The AI Receptionist master switch or the channel switch is off. Both must be enabled.
- Shadow mode is on. The AI is answering, but its bookings are hidden from dispatchers. Open the Conversations tab to see transcripts.
- Rate limit. The chat widget allows 60 messages per visitor per hour and 300 messages per business per hour.
The AI keeps telling me it can only help with service or booking questions — what is happening?
A topic gate runs before the main AI to filter off-scope and jailbreak messages. It is graduated: the first two off-topic messages get a polite redirect, the third escalates to a dispatcher, and the fifth is a hard stop. Any on-topic message resets the counter. You cannot turn the gate off — it protects the AI from spending tokens on conversations it was not built for.
A booking request I saw earlier is gone from the queue — where did it go?
Three possibilities:
- It was a shadow booking. Shadow-mode bookings are hidden from the default Pending Review filter. Switch to the Shadow status tab to see them.
- It expired. If Booking auto-expiry is set to a number of days (default is Never), pending bookings that no dispatcher has actioned are automatically expired after that period.
- Another dispatcher already approved or declined it. Check the Approved or Declined status tabs, or search by the customer name.
Why did the AI escalate to a dispatcher mid-conversation?
Escalation happens for several reasons: the conversation hit the token cap (50,000 tokens per conversation), the per-call cost cap was reached (default 200¢ for voice), the visitor tripped the topic gate three or more times, or sentiment-based escalation detected 2 consecutive negative visitor turns (configurable). In each case the visitor is told a dispatcher will follow up.
Win-Back and outreach
How do I make sure win-back outreach is TCPA-compliant?
The platform enforces four layers:
- Consent audit trail — each enrollment records when and how consent was captured. The delivery job refuses to send if consent is missing.
- Suppression list — re-checked at send time, not just at enrollment. A recipient who opts out between enrollment and delivery is blocked.
- STOP-keyword opt-out — a STOP reply from the recipient adds them to the suppression list automatically.
- Quiet hours — outbound voice calls are never placed during the tenant’s configured quiet window (default 9 PM – 8 AM local time). Set your own window at Settings → Outbound voice quiet hours.
Outbound voice also enforces a per-phone rate cap: by default, 3 attempts per phone number across all trigger kinds inside a 7-day rolling window.
What audience types can a win-back campaign target?
Five audience builders are available: Dormant customers (converted bookings older than 90 days with no follow-up), Declined estimates (Field Service estimates declined or expired within 180 days), Overdue maintenance (last completed work order older than 300 days), Abandoned intakes (conversations abandoned or closed with no booking, older than 24 hours), and Unscheduled recurring service (service agreements with a pending entry in the next 30 days that never generated a work order). All lookback windows are defaults.
Call Coach
What does Call Coach score, and how is the overall score calculated?
Call Coach scores each conversation on five axes, each rated 0–100: Greeting (friendly, identifies the business), Diagnosis quality (right qualifying questions for the service type), Booking attempt (tried to schedule or quote), Objection handling (addressed price or availability concerns), and Escalation appropriateness (escalated only when warranted). It also flags whether a booking opportunity was missed. The overall score is the average of the five numeric axes.
Licensing and limits
Does AI Front Office require a license?
No. AI Front Office is a pay-as-you-go app — it is available to every tenant with no license decision. Billing is per active front desk; zero front desks costs nothing.
What limits should I know about?
| Limit | Default | Configurable? |
|---|---|---|
| LLM tokens per conversation | 50,000 | Yes |
| Voice-call cost cap | 200¢ per call | Yes |
| Geocode confidence for auto-book | 0.85 (0–1 scale) | Yes |
| Daily job capacity (capacity-aware booking) | 12 jobs/day (floor of 1) | Yes |
| Chat widget messages | 60/visitor/hour, 300/business/hour | No |
| Online booking submissions | 30/visitor/hour, 200/business/hour | No |
| Lead webhook requests | 500/business/hour | No |
| Bulk actions (approve/decline/export) | 100 records per submit | No |
| Outbound voice attempts per phone | 3 per 7-day window | Yes |
| Sentiment escalation streak | 2 consecutive negative turns | Yes |
| Speed-to-Lead SLA window | 5 minutes | Yes |
When the token cap or cost cap is hit mid-conversation, the AI escalates to a dispatcher instead of burning more tokens.
More help
- AI Front Office Overview
- Ask AI — the assistant answers AI Front Office questions from these articles.