m MessageBox
All systems green · > 99.9% Book a demo
On this page
M.09 · AI Solutions Delegation level · Genius

Quietly intelligent. Loudly accountable.

AI that classifies a 12-second voice note in under a second, routes the task, matches a lost earring to its owner, and turns a guest like or dislike into a profile update immediately — with a human still in the loop and a paper trail your auditor would smile at.

<1s classify a voice note <1s service-type linkage 0 hallucinated facts Every decision logged
Live · staff voice note · Suite 1204
AI inference Today · 14:31:08 GST
In · voice
0:12
Transcribed — “Hi, uh, the AC in twelve oh four isn't really cooling — guest's a bit annoyed, can someone come up soon?
AI · entity extraction 117 ms
Service type Engineering · HVAC 0.96
Area / room Suite 1204 0.99
Urgency High 0.88
Sentiment Guest annoyed 0.83
→ Routed & assigned
ENG-7741 · Suite 1204
● AI · auto-routed
Air-conditioning underperforming, master bedroom
Engineering · HVAC Mario R. · on shift SLA 20 min
Closed in 23 min · audit trail saved
Chapter 01 / The pipeline

From a sentence in a hallway,
to a routed, assigned ticket.

01 — Capture

Voice or chat, in.

A staff member taps the mic. A guest sends a message. A note arrives by email. We don't care which.

channel: voice
duration: 12s
size: 94kb
02 — Transcribe

Words, in realtime.

Streaming multilingual transcription handles accents, noise and context — tuned for real conversations.

model: gpt-4o-transcribe
mode: real-time
latency: ~400–900 ms
03 — Classify

Type, area, timing.

LLM extracts service type, location, urgency, sentiment — with a confidence score on each.

service: engineering
area: suite_1204
urgency: high · 0.88
04 — Match

To real IDs in DB.

The classified service type and area are matched to actual records in your tenant — not invented.

dept_id: DEPT_017
room_id: RM_1204
sla: 20m
05 — Route

To right human.

On-shift, in-zone, with the right skill tag. The task lands on a phone with the original audio attached for context.

assignee: mario_r
phone: +971 5··· 4127
eta: 2 min
2.5s end-to-end median
1 Proprietary Algorithm matching
100% Fallback
0 facts invented
Voice → Task
Mobile · Web
“Lobby flowers wilting, the white ones — replace before the wedding party arrives at four please.”
F&B · Florals Lobby By 16:00 · today
Chat → Task
WhatsApp · Webchat · LINE
“Hi from 906, the shower drain is a bit slow and we've got a dinner at 7 — could someone take a look?”
Engineering · Plumbing RM 906 By 19:00 · today
Chapter 02 / Lost & Found

From misplaced to traced.
By a model that's seen a thousand earrings.

— Coming soon

Photograph it. The model
does the describing.

Housekeeping snaps a photo. Within a second, the model writes a structured description — material, colour, type, distinctive marks — and indexes the item against every guest who's checked out in the last 30 days.

  • Auto-described. Brushed gold, single drop earring, pearl accent. Logged exactly the way you'd type it, only faster.
  • Auto-matched. Cross-referenced with departing guests, in-room photos, and prior queries — ranked by confidence.
  • Auto-notified. The likely owner gets a WhatsApp with a photo and a one-tap "yes, that's mine" reply.
  • Audit-grade. Every photo, every match, every notification — timestamped and signed.
SCAN · 0.94 CONFIDENCE · HIGH
LF-3318 Logged · 14:18 · Suite 906
TypeValuable
StorageVAL-01
DescriptionPearl Earring · single
Found byHK · Aishwarya
Likely owner · Mrs. Chen Suite 906 · departed Tue 11:04 · WhatsApp on file
Chapter 03 / Guest feedback analysis

Your therapist for
guest feedback.

★ ★ ★
Inhouse Feedback · 02:47
The guest is staying for his wedding anniversary. He is finding AC in the bedroom loud. He also felt that the breakfast buffet felt tired for the price, repeats the same eggs/pastries every morning. However, he liked the room and the welcome was thoughtful. He found the service from the front desk incredible — Lana especially was amazing. He said he would have come back if it wasn't for the AC and the buffet.
AI · 4 themes extracted
Room aesthetics2 mentions Positive No action
! HVAC · noiseBedroom AC Negative → Engineering · ENG-7892
Front desk · LanaNamed recognition Positive → Recognise · HR-RC-44
! Breakfast buffetVariety, price perception Negative → F&B review · FB-2241
— Coming soon

A 2 a.m. Feedback
becomes a 9 a.m. action list.

Feedbacks have consequences. The model breaks each one into themes, tags sentiment, and routes the negatives as tickets and the positives as recognitions. By the time you walk in, the work has names against it.

  • Theme extraction. No more "the breakfast was bad" buried in paragraph four — every theme surfaces.
  • Sentiment per theme. A single review can have four moods. We separate them.
  • Action routing. Negatives become tickets, with the verbatim quote attached. Positives become recognitions.
  • Patterns over time. When "AC noise" shows up in eight reviews this month, the model tells you before you ask.
Chapter 04 / How we build it

Domain-specific. Human-led.
Audit-shaped.

i. Principle 01

Always with a human in the loop.

The AI proposes; a human accepts, edits, or overrides. Every override becomes training data for tomorrow's model.

ii. Principle 02

An audit trail for every decision.

Every classification, every match, every route — logged with model version, input, output. Forever.

iii. Principle 03

No hallucinated facts.

If a service type or room ID isn't in your tenant database, we map against a fallback with the user seeing the actual details — we don't invent.

iv. Principle 04

Domain-specific, not general-purpose.

Hospitality language, accents, slang, abbreviations. Trained on actual hotel data — not the open web.

Models hosted in your region · EU Training data: your tenant only, by default
Chapter 05 / What's next

Lots of quick wins.
More to come. We've just started.

The AI roadmap, written down.

Updated weekly · 11 modules in pilot
Shipping now RM.01

Lost & Found AI matching

Photo → description → owner match → guest WhatsApp.

Shipping now RM.02

Monthly guest feedback analysis

TripAdvisor, Booking, in-stay surveys — themed and actioned.

Shipping this quarter RM.03

Voice & chat to task

Live in 88 properties. Multilingual. Sub-second latency.

Shipping this quarter RM.04

Suggested replies (guest chat)

Hospitality-toned drafts for guest messaging — review and send.

Shipping this quarter RM.05

Suggested Preference

Understands that guest is conveying a preference and does PMS updates automatically.

Shipping this quarter RM.06

Smart escalation

If a thread is "going wrong", a manager gets a heads-up immediately.

Next quarter RM.07

Checklist - Ticket Creation

Create a ticket based on the comments on a checklist automatically. Track corrective action properly.

Next quarter RM.08

Feedback creation using Chat

No more form filling. Just type in the guest feedback and feedback created automatically.

Later · 2026 RM.09

Lost & Found - Photo based creation

Click a photo. Enter the area. Create a Lost and Found ticket automatically.

— Show us your data

The AI doesn't get its own demo.
It gets built into yours.

A 30-minute walkthrough on your property data. We bring the models, you bring a real shift's tickets, we'll show you the difference live.
Book a slot Schedule a demo WhatsApp us · +1 601 843 9486 LINE chat us Or write to hello@messagebox.ai