AI in Singapore F&B: Dynamic markdown and specials boards

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

The problem this solves

Every food venue ends some days holding stock that will not survive the night: pastries baked for a rush that never came, the last portions of today's braise. The information needed to act usually exists by mid-afternoon, a forecast of tonight's covers against what is still on the pass, but acting means someone doing margin arithmetic mid-shift and improvising a discount and a handwritten sign, so mostly nothing happens until closing, when the food goes in the bin. When discounting does happen it is ad hoc: no consistent margin floor, no record of what worked, and no link to the channels (an in-store screen, a social story) that would actually move the stock before last order.

Nationally, the scale is not in doubt. NEA reports Singapore generated 790,000 tonnes of food waste in 2025, about 11% of total waste, holding in roughly the same 755,000–817,000 tonne band every year since 2021, with the recycling rate stuck at 18% for four years running (NEA, accessed 2026-07). Large commercial and industrial premises above set thresholds, including food manufacturers, caterers and shopping-mall F&B areas, are already required to segregate and report food waste under the Resource Sustainability Act, a practical starting point for any operator wanting a data trail to forecast from.

What it costs to ignore

No Singapore-specific quantified estimate for restaurant-level surplus was found. US industry benchmarks (not Singapore figures, read as directional only) put the average restaurant's food waste at 4–10% of the food it buys and 5–6% of revenue, and one widely cited analysis puts the US sector's annual lost profit from food waste at around USD 162 billion (independent press, US market, directional). Run the arithmetic for your own venue: a casual-dining restaurant turning over SGD 1.5 million a year losing even 2% of revenue to unsold surplus is writing off roughly SGD 30,000 annually, before disposal costs or the covers lost to a display case that already looks picked-over by 4pm.

What good looks like

• An early warning, not a closing-time discovery - a mid-afternoon view of tonight's forecast shortfall or surplus, in time to act before service ends rather than after.
• Timed markdowns and bundles, not a blanket discount - a specific proposal ("2 pastries for SGD 7 after 5pm") rather than a flat percentage off everything.
• One decision, two channels - the same proposal updates the in-store specials screen and goes out as a social post, instead of a person retyping it twice.
• A margin floor that actually holds - ask the system to discount a loss-leader below cost and it refuses; the rule sits outside the AI's judgement, not inside a prompt it could be argued out of.
• Worth knowing - the grocery-sector waste-reduction figures vendors quote (30–80% less waste, several points of margin) come from supermarket electronic-shelf-label deployments at a different scale to a single restaurant; treat them as directional, not a guarantee.

How it works

A forecast decides what is at risk, fixed rules decide the price and the timing, and the AI only writes the words and drives the channels.

• Mechanism: a covers/demand forecast is compared against on-hand inventory and prep quantities from the POS; a markdown rules engine picks candidate items, timing and price steps inside a pre-set margin floor; a language model drafts the offer copy in the house voice; a renderer pushes the same offer to an in-store digital screen and, optionally, schedules it to social channels.
• Data it draws on: sales history and the day's forecast, current inventory and prep sheets, item costs and margin floors, and a house list of items that may never be discounted (a signature dish, anything under a supplier price-protection agreement).
• How it decides: price, floor and timing are fixed rules, never left to model judgement; the model's job is copy and formatting only. This is markdown to clear surplus, explicitly not surge pricing: prices only ever move down from the standard menu price, never up, and the system has no mechanism to raise a price at peak demand.

Vendor landscape

Singapore-native gap. No Singapore-native vendor stitches together predictive end-of-day surplus forecasting, a rules-based margin floor, an in-store specials screen and a social push, sized for a single restaurant. The closest Singapore-facing piece is Yindii, founded in Bangkok and live in Singapore since August 2024, but it is a consumer marketplace app: a merchant lists a discounted "surprise bag" on Yindii's own storefront, a different thing from a system that predicts tonight's shortfall, proposes a bundle or drives an in-store screen. The AI-driven markdown engines that do exist, Wasteless and Invafresh, are grocery-chain enterprise tools built around electronic shelf labels at supermarket scale, and neither is marketed in Singapore. Digital-signage vendors, including Singapore-incorporated Blupepper, supply the in-store screen but carry no pricing intelligence of their own. The category is genuinely fragmented, in Singapore and globally: the pieces exist, but nobody has assembled them for a single restaurant.

None of these tools does the whole job. The surplus-marketplace apps sell to consumers off-premises rather than driving an in-store screen; the grocery markdown engines are built for chain-scale shelf-label estates; the signage vendors render whatever they are told but decide nothing.

Buy or build?

If the goal is simply to stop throwing food away, buy: list surplus on Yindii at SGD 39/year and accept that it is a consumer channel, not an in-store system. That alone addresses the waste question without a build. The forecast-plus-margin-floor-plus-specials-screen loop is a build, because nobody sells that combination for a single Singapore restaurant at any price. Start with fixed rules rather than a machine-learning forecast, a single restaurant's daily transaction volume is thin, and a simple rule (mark down category X by Y% after 5pm if forecast covers are under Z) will outperform a model with too little data to learn from. Keep a person able to override any proposed markdown, and keep the margin floor as a hard rule outside the AI's reach, not a suggestion inside a prompt.

Singapore-specific considerations

• Food safety: discounted food is still food. The obligation that food offered for sale must be sound and fit for consumption, and the storage and time-stamping requirements for prepared food, apply regardless of price; a markdown engine should never be allowed to extend how long an item stays on display.
• Reputational risk, and why this is not surge pricing: Wendy's 2024 announcement that it would explore time-of-day pricing, despite saying prices would only fall at quiet times and never rise at peak, triggered a #BoycottWendys backlash and a rival's "No urge to surge" campaign within days. A 2025 US industry survey found 62% of restaurant operators worried about consumer reaction to dynamic pricing. Message this as a savings or sustainability offer, a countdown discount, a "surplus special", never as "pricing" or "surge", and never let a price move upward.
• PDPA: a specials board and a generic social post process no personal data. Targeting individual loyalty members with personalised markdown offers would cross into personal data processing and need consent.
• Grants: none of the vendors above was confirmed on the PSG pre-approved list; digital signage is PSG-supportable in principle, up to 50% of qualifying cost, capped at SGD 30,000 per company per year, against listed vendors only. A custom predictive-markdown build is the kind of project AI Singapore's 100 Experiments (100E) can co-fund, up to SGD 150,000, matched.
• Language: English covers most of the market; Mandarin-language last-minute specials perform well on food-focused social platforms for local, price-sensitive buyers.

Sources

1. NEA, Food Waste Management (generation, recycling and disposal figures 2021–2025, Resource Sustainability Act reporting) - https://www.nea.gov.sg/our-services/waste-management/3r-programmes-and-resources/food-waste-management - government - accessed 2026-07-14
2. The Restaurant HQ, restaurant food waste statistics (US industry benchmarks) - https://www.therestauranthq.com/trends/restaurant-food-waste-statistics/ - independent/blog - accessed 2026-07-14
3. Forbes, "Restaurants Lose $162 Billion To Food Waste, New Report Finds" - https://www.forbes.com/sites/hankcardello/2026/03/25/restaurants-lose-162-billion-to-food-waste-new-report-finds/ - independent/press - accessed 2026-07-14
4. Too Good To Go - Wikipedia (markets and countries of operation) - https://en.wikipedia.org/wiki/Too_Good_To_Go - independent/reference - accessed 2026-07-14
5. Back of House, "This App is Helping Restaurants Monetize Food Surplus and Curb Food Waste" (Too Good To Go US fee structure) - https://backofhouse.io/resources/app-helping-restaurants-monetize-food-surplus-curb-food-waste-good-to-go - independent/trade - accessed 2026-07-14
6. Karma, company site - https://karma.life/ - vendor - accessed 2026-07-14
7. Yindii, for business - https://www.yindii.co/for-business - vendor - accessed 2026-07-14
8. Yindii, Singapore launch announcement - https://www.yindii.co/post/yindii-sea-s-largest-marketplace-for-surplus-food-is-coming-to-singapore-to-fight-food-waste - vendor - accessed 2026-07-14
9. Wasteless, AI markdowns - https://www.wasteless.com/ai-markdowns - vendor - accessed 2026-07-14
10. Tailwind Capital, "Tailwind Capital Merges Invafresh with Upshop" - https://www.tailwind.com/news/tailwind-capital-merges-invafresh-with-upshop/ - independent/press - accessed 2026-07-14
11. Blupepper, about us - https://blupepper.co/sg/about-us/ - vendor - accessed 2026-07-14
12. Juuno, "Best Digital Signage Software 2026" (pricing comparison) - https://juuno.co/blog/digital-signage-software - independent/blog - accessed 2026-07-14
13. QSR Magazine, "Why Traditional Dynamic Pricing Doesn't Work for Restaurants" - https://www.qsrmagazine.com/story/why-traditional-dynamic-pricing-doesnt-work-for-restaurants/ - trade/independent - accessed 2026-07-14
14. Restaurant Dive, "Wendy's backtracks on dynamic pricing after consumer backlash" - https://www.restaurantdive.com/news/wendys-backtracks-on-dynamic-pricing-after-consumer-backlash/708799/ - trade/independent - accessed 2026-07-14
15. Singapore Statutes Online, Sale of Food Act 1973 (general food-safety obligation) - https://sso.agc.gov.sg/Act/SFA1973 - legislation - accessed 2026-07-14
16. Enterprise Singapore, Productivity Solutions Grant (support rate and annual cap) - https://www.enterprisesg.gov.sg/financial-support/productivity-solutions-grant - government - accessed 2026-07-14
17. AI Singapore, 100 Experiments (100E) (funding cap) - https://aisingapore.org/innovation/100e/ - government-linked programme - accessed 2026-07-14

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].