A Restaurant Stacked AI on Everything Boring and Saved 20 Hours a Week

A Singapore restaurant called Grain is saving 15 to 20 hours a week by running AI across three things you would never put in a press release: dishwashing logistics, delivery routing, and shift scheduling. Not a chatbot. Not a social media tool. Dirty dishes and delivery windows. Published this week by Vulcan Post, the story is worth your attention not because Grain is doing something exotic, but because they are doing something most operators never think to do.

They stacked it. AI on one thing is a pilot. AI on everything tedious is a business model.

I ran auctions for twenty years. You learn fast that the money isn't in the marquee lots, the oil paintings and the silver. The money is in what you do with the other 400 items in the room, the ones everyone ignores. The operators who built real businesses figured out how to move the whole room, not just the chandelier. Grain figured out the same thing, and they did it with software instead of a gavel.

What Grain actually built

Grain is a Singapore food and beverage company that started as an online restaurant and catering business and has been running for more than a decade. Its co-founder and CEO, Yong Yi Sung, came from a management consulting background, which apparently gave him permission to look at F&B problems the way a consultant would: find the friction, measure it, fix it, repeat.

The flagship AI project came from the stewarding team, the people who manage dishes and kitchen equipment. The problem was predicting how much dirty ware was coming back from events and deliveries on any given day. Get it wrong and you either overstaff a shift or your team is buried. Grain built a forecasting system that predicts ware volumes and staffing needs, tracks maintenance, and logs routine tasks. The result: 15 to 20 hours saved per week, missed tasks and repeat defects down 20%, and issue resolution time cut by 50%.

Beyond that, the company now runs an AI autopilot that responds to customers and helps close sales, a "Tasting" dashboard that synthesizes customer feedback to improve menus, and an AI-powered delivery and operations tracker. Around 200 employees are actively using or building AI tools, and the company has rolled out more than 100 new tools and workflows.

That last number is the one that should stop you. Not 100 grand experiments. A hundred tools in daily use, across a food and beverage company, built mostly by the people doing the work.

The stacking principle

Here's the idea I'd ask you to hold onto, and I'm going to call it the stacking principle: a single AI tool saves you minutes. A dozen AI tools, each saving a few minutes on a task that happens every single day, saves you hours. Not metaphorical hours. Real hours you can reassign, recover, or just stop losing.

Sung says it plainly in the piece: "not using AI for the sake of it, but constantly asking whether there is a better, faster or simpler way to do something." That's the whole framework. Every repetitive task is a candidate. You don't need a big investment or a tech team to start. You need the habit of asking the question.

I've watched this play out in restaurants closer to home. Toast's data on how restaurant managers actually use AI during service shows the same pattern: the wins aren't dramatic, they're cumulative. A manager who uses AI to prep for a difficult conversation, check a vendor contract, and draft a shift note hasn't automated their job. They've bought back forty-five minutes they would have burned otherwise.

What this means for your restaurant

If you run a restaurant or a catering operation, the Grain story is directly translatable. You don't need their scale. You need their method.

Think about what your team does every week that is purely mechanical. Forecasting how many covers you'll do Thursday based on reservations and weather. Scheduling staff against expected volume. Tracking which equipment needs maintenance before it breaks during a Friday dinner rush. Writing the same vendor email for the third time this month. Summarizing comment cards so a manager actually reads them.

Every one of those is a candidate. None of them require a custom software build. Most of them can be started with tools that cost less per month than a single overtime shift.

Grain's CEO also made a point that I think owners underestimate: the company ran weekly learning sessions where employees shared what they had discovered. Someone found a use case, tested it, and passed it on. That's not an IT rollout. That's just making experimentation part of the culture instead of a one-time event. A social club in our coverage area did something similar and cut a three-day grant writing process to two hours. Same principle, totally different industry.

Under the hood, for the technically curious

The stewarding system Grain built is a demand forecasting model at its core. It takes historical ware return data, maps it against order volume and event schedules, and produces a staffing recommendation. This is not a large language model doing something clever. It's a prediction model doing something that Excel has technically been able to do for twenty years, just faster and with less manual data wrangling.

The interesting part is the maintenance tracking layer on top of it. That's closer to an asset management workflow with AI handling the logging and alerting. When the system reduces repeat defects by 20%, what it's actually doing is catching missed tasks before they cascade into equipment failures. That's a quality control loop, not magic.

The AI autopilot handling customer responses is almost certainly a large language model with a structured prompt and CRM integration. The Tasting feedback dashboard is a sentiment and theme extraction layer over customer comments. None of this is proprietary technology. All of it is available off the shelf. What Grain built is the habit and the workflow, not the underlying model.

What to do this week

First, make a list of the five most repetitive tasks your team does every week. Not the hardest problems. The most repetitive ones. Scheduling, vendor communication, feedback review, inventory forecasting, shift reporting. Pick one. Try handling it differently for two weeks.

Second, if you have not looked at what Toast's AI agents are doing for scheduling and marketing right now, spend twenty minutes there. It's the closest American restaurant analogue to what Grain is doing, and it's already built into a platform many of you are running.

Third, ask your team. Seriously. Grain's breakthroughs came from the stewarding team, not the C-suite. The person doing the repetitive task usually knows exactly how painful it is and has already wondered whether there's a better way. Give them thirty minutes in a meeting and ask.

Fourth, stop waiting for a single tool that solves everything. The stacking principle only works if you start stacking. One task, this week, then another one next month.

You can skip this one if

You run a solo operation with no recurring tasks and no staff, and your whole week is custom work that changes every time. The stacking principle needs repetition to compound. If nothing repeats, there's nothing to stack. Come back when you have a team.

The question I'd actually ask you

If I were sitting across from you right now, I'd ask you this: what does your team do every single week that a smarter version of your operation would have automated two years ago? Not what's possible in theory. What's the thing your people do on autopilot, that takes real time, that you've never once stopped to question?

Grain's stewarding team was forecasting dirty dishes by hand. They weren't embarrassed about it. They just fixed it. That's the whole story.

Sources

• Vulcan Post, "Grain is using AI on everything from dirty dishes to delivery and saving 20 hours a week," October 21, 2026: https://vulcanpost.com/914012/grain-restaurant-catering-ai/