AI in Singapore F&B: Multilingual menu / allergen Q&A (QR)
Last updated: 2026-05 FX reference: 1 USD = SGD 1.28, retrieved 2026-05-28
See the mock-up for this use case
A static, on-brand design concept — illustrative data, not a live system.
The problem this solves
Language at the table is a live Singapore issue. A Stomp poll of 4,309 respondents found 55% would not patronise restaurants with single-language menus (Stomp, trade/independent). An English-only menu in a tourist area excludes Mandarin, Bahasa Melayu, Tamil and Japanese-speaking guests.
Allergen risk is rising and local. In Singaporean children, peanut has overtaken bird's nest as the top anaphylaxis trigger, and food is the most common anaphylaxis cause across all ages (KKH, academic). The SFA lets restaurants communicate allergens informally rather than label every dish (SFA, regulator), so the burden of an accurate answer falls on floor staff — who may not speak the guest's language.
A QR-at-table agent lets a guest ask, in their own language, about pork, gluten or spice level. The essential design rule is strict retrieval-only: it answers from a curated allergen table or says "please ask staff," and never improvises an ingredient.
What it costs to ignore
No credible quantified SGD estimate is available. The downside is tail-risk — one mis-stated "nut-free" answer leading to harm and liability — plus diffuse lost covers from language exclusion. AI-hallucination liability is now live globally, and disclaimers do not reliably protect operators where real harm results (trade). That asymmetry is the argument for retrieval-only.
What good looks like
- Wider addressable covers — removes the language barrier 55% of diners react to (Stomp); realistic floor: incremental and hard to isolate.
- Server time recovered — guests self-serve common questions; marginal at 60 seats, useful at peak.
- Consistent allergen answers — one grounded data source answers identically in every language (depends on data hygiene).
- Policy alignment — fits the STB × OpenAI direction on multilingual tourism service.
How it works
It answers only from verified allergen data, and is designed to defer to staff rather than guess.
- Mechanism: the agent retrieves answers from a curated allergen/ingredient table and only answers from that data; languages are handled by the model.
- Data it draws on: a structured per-dish ingredient and allergen table (the verified source of truth) and the menu.
- How it decides: if the question maps to a known dish and allergen, it answers from the table; if it isn't sure or the item isn't in the data, it says "please ask staff" rather than inferring an ingredient.
Typically built as a retrieval app over a carefully curated allergen dataset, reached by a QR code at the table.
Vendor landscape
Singapore-native gap. No SG-native vendor does AI retrieval-grounded multilingual allergen Q&A. SG QR-menu players (Qashier, Oddle, Novitee) offer multilingual *static* menus, not guardrailed conversational Q&A. The vendors that do conversational Q&A (Qrav, Chocochip) are overseas and publish no guardrail detail. Answering only from verified data, and deferring to staff otherwise, is exactly what's missing off-the-shelf.
| Vendor | Origin | SGD/month | Suited to |
|---|---|---|---|
| MenuTiger | Philippines/US | Free to ≈ SGD 152 | Cheap multilingual *static* menu + QR ordering |
| Qrav | India | ≈ SGD 15–77 | The closest functional match (but no published guardrails) |
| Qashier | Singapore | QR bundled in Growth plan | The SG-native QR/ordering plumbing |
| Oddle | Singapore | from SGD 215 | SG ordering/delivery (not allergen Q&A) |
A few facts to weigh: MenuTiger's "multilingual" is static translation, not Q&A. Qrav answers allergy questions but is shown asserting "100% nut-free" with no visible guardrail — the failure mode this use case exists to prevent. Qashier and Oddle are SG-native and keep data in-country, but their menus are static — they're the front end, not the answering brain.
Buy or build?
Buy the QR-menu and ordering plumbing from an SG-native provider (Qashier or Oddle) for PDPA-resident data, but build (or commission) the allergen Q&A layer yourself — retrieval-only over a curated allergen table, with a firm "please ask staff" fallback. No current vendor ships that guardrail, and for allergen questions the guardrail is the whole point.
Singapore-specific considerations
- PDPA: guest questions may contain health data (allergies are sensitive) — log minimally, retain little, host in-region or anonymise.
- Grants: PSG (70%, enhanced April 2026, SGD 30,000/year cap) likely covers the SG-native QR/POS plumbing; a bespoke AI Q&A layer fits AI Singapore 100E (up to SGD 150,000) better. Aligned with the STB × OpenAI multilingual direction.
- Integrations: Qashier, Oddle, Novitee for the QR/menu front end.
- Language: English, Mandarin, Bahasa Melayu, Tamil, plus Japanese for tourists.
Sources
- Stomp — Chinese-only menus backlash — https://www.stomp.sg/the-big-picture/chinese-only-menus-spark-backlash-some-eateries-add-english-translations-reach-wider-crowd — trade/independent — accessed 2026-05-28
- KKH — food allergy prevalence study — https://www.kkh.com.sg/news/announcements/kkh-study-to-assess-growing-prevalence-and-impact-of-food-allergies — academic — accessed 2026-05-28
- SFA — food allergy/intolerance guidance — https://www.sfa.gov.sg/food-safety-tips/food-risk-concerns/food-allergy-intolerance — regulator — accessed 2026-05-28
- Qrav — https://qrav.in/ — vendor — accessed 2026-05-28
- Qashier QR ordering — https://qashier.com/sg/qr-code-ordering/ — vendor (SG) — accessed 2026-05-28
- STB × OpenAI MOU — https://www.stb.gov.sg/about-stb/media-publications/media-centre/stb-and-openai-sign-first-of-its-kind-mou-to-prepare-tourism-sector-for-ai-driven-future/ — government — accessed 2026-05-28
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].