AI in Singapore F&B: Photo-to-floor-plan onboarding

Last updated: 2026-07 FX reference: 1 USD = SGD 1.28, retrieved 2026-07

Stub research page. This one is lighter than most of the catalogue: no formal market study behind it yet, just floor evidence from RAS 2026 (Singapore's Restaurant Association trade show) plus reasoning from the fuller UC21 research already in this catalogue. Treat it as a working thesis, due a proper pass once someone commits to building it.

The problem this solves

Every reservation platform onboarding still starts the same way: someone from the vendor visits the venue, measures the room, and manually redraws the floor plan into the booking system before a single table can be assigned. The floor-plan research already in this catalogue (see UC21) found that SevenRooms, OpenTable, Chope, TableCheck and Tablein all manage seating assignment inside a fixed floor plan, and that fixed floor plan gets built by hand at setup, not derived from anything the operator already has.

An operator with a phone can photograph their own dining room in minutes. The obvious question that follows: why does that photo need to become a person's visit and a line item on the invoice, rather than an input a model reads directly?

What it costs to ignore

No published figure for a Singapore-specific onboarding-visit cost was found. Directionally: an onboarding visit that requires travel, measurement and manual redrawing is a real line item, and vendors that quote a setup fee on top of a monthly subscription commonly land in the hundreds of Singapore dollars. Call it roughly SGD 700 as an illustrative outlay to eliminate, not a sourced figure.

What good looks like

Photo in, plan out: the operator photographs the room from a couple of angles; a vision model proposes table positions, shapes and approximate capacities.
Human-in-the-loop adjust: the operator drags tables to correct anything the model got wrong before it becomes the working floor plan. The model proposes, it never auto-commits.
No visit required: onboarding drops from a scheduled site visit to a five-minute phone job, done whenever the operator gets to it.
Worth knowing: this is an onboarding tool, not a redesign tool. It captures what is already in the room; it doesn't recommend a better layout. That's UC21's job, and it's the natural next step once a venue has a digitised floor plan to score.

How it works

A vision model reads room geometry from photos; a person confirms it before it goes live.

Mechanism: a vision model detects table footprints, approximate seat counts and walkways from one or more photos of the dining room, and proposes a draft floor plan on a grid.
Data it draws on: the photos themselves; no booking or POS history is needed for this step, which keeps the regulatory surface low. There's no guest data in a photo of an empty dining room.
How it decides: the model proposes, the operator disposes. Every table position is editable by drag before the plan is saved, so a misread corner booth or a folded-away table doesn't silently become the system of record.

Vendor landscape

Singapore-native gap. No vendor observed at RAS 2026 offers photo-to-floor-plan capture. The reservation platforms in this catalogue's UC21 research (SevenRooms, OpenTable, Chope, TableCheck, Tablein) all still rely on the floor plan being drawn by hand at onboarding, whether by the venue or by the vendor's own team. CAD tools (SmartDraw, RoomSketcher) draw a floor plan well but start from a blank canvas, not a photo, and have no link to a booking system either way.

Buy or build?

Nothing off the shelf does this today, so it's a build if anyone wants to move on it. The upside is that the model itself is a bounded, well-scoped computer vision task (detect rectangular and round shapes on a floor, estimate scale from known reference objects like chairs), and the output feeds directly into the booking-diary and floor-plan-scoring work already mapped out elsewhere in this catalogue, rather than starting a new data model from scratch.

Singapore-specific considerations

PDPA: minimal exposure at this stage. A photo of an empty dining room carries no personal data; that changes if photos are taken during service with guests in frame, so capture should happen pre- or post-service.
Grants: untested. No vendor is pre-approved for this specific pattern, so budget it as a cash-funded build unless a broader digitalisation grant happens to cover it.

Related in this catalogue: UC21, floor-plan and covers optimiser is the natural next step once a venue has a digitised plan, scoring it against the actual booking mix. UC02, booking and waitlist web chat is the other place a freshly captured floor plan gets used day to day, once tables need to be assigned to bookings.

Sources

RAS 2026 show-floor observation, Singapore, informal vendor survey, not exhaustive; first-hand; July 2026.
This catalogue, UC21 floor-plan and covers optimiser research (vendor landscape and pricing detail); internal; 2026-07.

This is part of a series on AI use cases for Singapore F&B operators, refreshed every two months. If you'd like to discuss applying any of this to your restaurant, get in touch at [email protected].