Customers & Automation
SMS Auto-Respond
Last updated 19 May 2026
SMS Auto-Respond
When a customer texts your dedicated SMS number, AI classifies what they're asking about and either drafts a reply (you review and send) or auto-sends (under your configured per-intent rules). Velocity caps, a kill switch, and per-intent mode controls keep you in charge.
Where to find it — Web: Settings → Communications → SMS → AI auto-respond. Owner-only, PRO required.
TL;DR
- 6 intent categories: pricing, hours, availability, booking link, general, complaint.
- Per-intent modes: off, draft (default), or auto-send. General is draft-only; complaint always escalates.
- Velocity caps prevent the AI flooding (20/hour, 100/day defaults).
- Kill switch immediately disables all auto-send. Re-enabling doesn't restore your modes — you re-enable each deliberately.
- PRO only. Configuration is web-only; drafts appear in your inbox on both web and mobile.
What the customer experiences
A customer texts your dedicated number:
"How much is a balayage?"
The AI classifies this as pricing intent. If your pricing intent is on auto-send, the AI checks your service list and replies:
"Balayage starts from £140 for a quick refresh, £180 for full balayage. Book at {link}"
If your pricing intent is on draft, the AI generates the same reply but holds it as a draft above your composer in the inbox. You review, edit if needed, and Send. The customer doesn't see anything until you do.
If the customer messages something the AI can't confidently classify or handle (complex history question, complaint, unusual request), it falls back to draft mode regardless of your settings — the message lands in your inbox for you to handle personally.
The 6 intents
| Intent | Examples | Modes available |
|---|---|---|
| Pricing | "How much for…?", "Cost of…?" | Off / Draft / Auto-send |
| Hours | "What time are you open Saturday?", "Are you open Sunday?" | Off / Draft / Auto-send |
| Availability | "Any time Thursday afternoon?", "Got slots tomorrow?" | Off / Draft / Auto-send |
| Booking link | "Send me your booking link", "Where do I book?" | Off / Draft / Auto-send |
| General | Greetings, chat, "Just checking", "Thanks!" | Off / Draft only — never auto-sends |
| Complaint | Negative feedback, problems | Always escalates to inbox — no AI reply |
The general intent is constrained at the schema level: it can only be off or draft, never auto-send. By design — a chat-style message doesn't have a clear ask, so the AI shouldn't reply on its own.
Complaints are always handled by you. The classifier flags them and escalates; no AI reply is generated.
Per-intent modes
For each of the four configurable intents (pricing, hours, availability, booking link), you pick one of three modes:
| Mode | Behaviour |
|---|---|
| Off | AI doesn't process this intent. The message lands in your inbox unhandled. |
| Draft | AI generates a reply, holds it as a draft above your composer. You review and Send. |
| Auto-send | AI generates and sends the reply directly. Logged in the inbox so you can see what went out. |
Auto-send confidence thresholds
Auto-send only fires when the AI is confident. Per-intent thresholds:
- Pricing, hours, availability: ≥85% confidence
- Booking link: ≥90% confidence (higher because misclassification here sends a real URL)
Below the threshold, the message degrades to draft regardless of your auto-send setting.
Velocity caps
To prevent AI flooding (e.g. a misclassification that fires 100 replies in 10 minutes), every venue has two velocity caps:
- Hourly cap: 20 auto-sends per hour (default)
- Daily cap: 100 auto-sends per day (default)
Once a cap is hit, auto-send degrades to draft for the remainder of the window. The inbox surfaces a status note ("Auto-send paused — daily cap reached") so you know what's happening.
You can adjust caps in the same settings page.
Kill switch
A single toggle at the top of the SMS auto-respond settings. Tap to immediately disable auto-send for every intent.
Important: re-enabling the kill switch doesn't automatically restore your per-intent modes. Auto-send for each intent stays off until you deliberately switch it back on. This is intentional — if something went wrong enough that you killed the switch, you should make a fresh decision about each intent before re-enabling.
Brand voice and Business Memory
Every AI reply uses:
- Your venue's selected brand voice persona (HOST / CONCIERGE / CASUAL / ENTHUSIAST)
- Your Business Memory facts — specialities, walk-in policy, atmosphere, etc.
- Your venue's avoid list — words and phrases the AI is forbidden from using
The avoid list is configured at Settings → Brand voice. Add words you don't want the AI to use ("sweetie", "darling", "babe").
Validation and safety
Every AI reply is validated before sending:
- Slot citations — if the reply mentions specific times (availability intent), those times must match real availability. Fabricated times are hard-failed and the reply degrades to draft.
- Canary token — a hidden marker in the prompt. If the AI echoes it back, the prompt may have leaked and the reply is forced to draft.
- British English enforcement — colour, customise, organise (filtered post-generation).
- No AI tells — em dashes, "delve", "leverage", "seamless", "robust" are filtered post-generation.
- Character cap — 320 characters hard ceiling (covers 2 SMS segments).
- PII scrubbing — telemetry never includes customer names, phones, or emails.
If a validation fails twice (the system retries at lower temperature once), the reply degrades to draft.
Quiet hours
If your venue has quiet hours configured, auto-send pauses during that window. The AI doesn't fire at 3am even if it would normally auto-send. Drafts are still generated and held — they're just not sent until you wake up.
Drafts in the inbox
When a reply lands as a draft (either by intent setting or by degraded auto-send), it appears in the inbox as an AI draft chip above the composer with three actions:
- Apply — drops the draft into the composer ready to send or edit
- Edit — type to refine before sending
- Discard — dismiss the draft
A staleness banner appears if the draft is older than 30 minutes (the conversation may have moved on).
Preview on sample threads
Before turning on auto-send for any intent, use Preview on sample threads to test the AI's output against 3 canned scenarios (pricing question, hours question, general greeting). Each preview consumes 1 Spark; rate-limited to 3 previews/hour per venue to prevent abuse.
Cost
| What | Cost |
|---|---|
| AI classification + reply generation | 1 Spark per reply |
| Preview sample | 1 Spark per sample (3-per-hour cap) |
Out of Sparks? The system writes a marker draft ("(Out of AI credits — reply manually)") so you know exactly what happened. Top up to resume auto-replies.
On the FREE plan
SMS Auto-Respond is PRO only. FREE venues see the upgrade prompt on the settings page. Inbound SMS still reaches your inbox — it just doesn't get an AI draft or auto-reply.
Common mistakes
| Problem | What to check |
|---|---|
| AI replied to a non-booking message | The classifier occasionally flags ambiguous messages. Use the inbox to send a corrective follow-up. The system learns from patterns over time. |
| AI mis-quoted a price | Confirm your service list has the right price. The AI cites prices from there directly. If the AI gave a range that's not what you intended, update Business Memory with the specifics. |
| Auto-send isn't firing | Check (in this order): kill switch off? Intent set to auto-send? Quiet hours not active? Velocity cap not hit? Wallet has Sparks? Confidence above threshold? |
| Draft is stuck saying "(Out of AI credits — reply manually)" | Wallet is empty. Top up from Settings → Billing to restore AI replies. |
| Customer says they got 2 AI replies | Shouldn't happen — idempotency by inbound message ID prevents this. If you see it, contact support with the conversation ID. |
| Need to disable for one event (e.g. close for holiday) | Hit the kill switch. Re-enable each intent deliberately when you reopen. |
FAQ
What does SMS Auto-Respond do?
When a customer texts your dedicated SMS number with a question, AI classifies the intent (pricing / hours / availability / booking link / general / complaint) and either drafts a reply for your review or auto-sends, depending on your per-intent mode. Complaints always escalate to you; general chat always drafts.
How is this different from the AI Booking Enquiry Auto-Responder?
Booking Enquiry Auto-Responder is a specific case — it handles SMS that's clearly a booking question and offers slot times. SMS Auto-Respond is broader — it handles pricing, hours, generic questions, and booking-link sharing. The booking-enquiry path runs first; if it's not a booking enquiry, SMS Auto-Respond picks it up.
What's the kill switch?
A one-tap toggle that immediately disables auto-send for every intent. Useful when something's going wrong or you want to pause AI handling for a busy stretch. Re-enabling the kill switch doesn't automatically restore your per-intent modes — you re-enable each one deliberately.
What happens when AI confidence is low?
Each intent has its own confidence threshold for auto-send (85-90%). Below that, the message degrades to a draft and surfaces in your inbox for review. Complaints, low confidence, validation failures, and canary-token leaks all force the message into draft mode.
Does it handle SMS in New Zealand?
SMS Auto-Respond requires a dedicated SMS number, which is AU/GB only. NZ doesn't have the SMS infrastructure. NZ venues use Email Enquiry Auto-Draft for the email-side equivalent.
Will my customers know they're talking to AI?
The AI never reveals it's AI by default. If a customer asks directly ("Is this a bot?"), the reply is "I'm here to help with bookings. How can I help?" Configure an AI disclosure suffix on your brand voice page if you want a regulatory disclosure appended.