This was the first time my podcast invited people who don't build AI to talk about AI.
Jerry Liao has been in Singapore for 21 years and in food and beverage for ten. He runs First Bowl, a halal Chinese fast-food brand, and Shiok Burger, which does fusion burgers. Together they have about 30 outlets. Eric Bao joined as CEO of First Bowl in 2024 to take it from one to ten, on operations and digitisation.
They aren't AI founders. They run thirty-odd shops serving close to 10,000 meals a day. To them, AI isn't agent architecture. It's whether the staff roster can come out a day sooner, and whether they can see last month's gross margin without waiting until the 20th of this month.
Halal certification is a moat, not a sentiment
Jerry started with Lanzhou beef noodles and noticed that halal Chinese food in Singapore was almost empty.
Not because nobody wanted to do it, but because few dared. Certification means rebuilding the kitchen, the supply chain and the staff. No pork, no alcohol, every seasoning supplier re-vetted, even the pots and pans replaced. Many Chinese restaurant owners see the cost and stop there.
Jerry ran a different calculation. Muslims are about 15% of Singapore's population, but their choices shape everyone else's.
"When a group of friends goes out to eat and one of them eats halal, the whole table follows."
With certification you can serve anyone, and revenue goes up 30 to 40%. He calls it the minority deciding for the majority. Competitors want to copy it but can't easily. Plenty of restaurant brands arriving from China have looked at the halal requirement and backed off.
39 square metres, three people, 3,000 meals a day
Jerry doesn't chase big stores. His model is the extreme small store: 39 square metres, three staff, 3,000 meals a day. Sales per square metre and per person are the two numbers he watches. Many owners at their twentieth outlet ask where to open the next one. He asks how much more efficiency he can squeeze out of the twenty he has.
He's also candid about himself: good at zero to one, not at polishing something to the limit. So he handed First Bowl to Eric and went off to start Shiok Burger. His century-egg chilli chicken burger became a hit.
From paper delivery notes to digital purchase records
When Eric took over First Bowl, the first problem wasn't AI. It was data.
Three point-of-sale systems that didn't talk to each other. Paper delivery notes. Rosters in Excel. Reviews scattered across Google, Grab and Foodpanda. To see last month's margin, the owner waited until the accountant closed the books on the 20th.
He spent more than a year stitching it together: one POS, Resto, across stores; rostering on StaffAny so labour cost per store per hour is live; reviews on Momos, with Google reviews synced and AI-assisted replies.
None of it is exciting. But without it, he says, every AI application on top would be built on air.
The tool he spent the most time on was ccMonet. Before, supplier deliveries came with a paper note, and month-end reconciliation meant flipping through a stack line by line. Want to know if a category's cost is rising? Go find the paper. Now every purchase line is digital and traceable: cost, category, supplier, price history.
That data became the foundation for AI. He's also clear about the limits: ccMonet covers finance and purchasing, but rosters live in StaffAny and sales in Resto, and joining them up is still on him. His next step is to connect the three into his own data layer.
Knowing on the 2nd whether you made money
Eric's team uses a Claude team account, seven seats, paid by the company. He pulls cost data from ccMonet, hours from StaffAny and sales from Resto, gives it all to Claude and asks for a full P&L forecast for the month.
The final books still close on the 20th. But on the 2nd of each month he now has a forecast that's close.
"I don't need it to be 100% accurate. I need it to be fast. By the 20th you know everything, but the window to adjust has passed. If I know roughly on the 2nd, I can decide that day."
He also tracks sales per labour hour for each store. The report comes out on the 1st, and on the 2nd it goes to the operations managers: which store's productivity is slipping, and is it the roster or the foot traffic?
A 60-point spreadsheet becomes a 90-point system in 20 minutes
They wanted a store inspection system with A, B and C grades tied to bonuses. The operations manager gave Eric a rough spreadsheet.
He gave it to Claude. Twenty minutes later he had what he called a 90-point inspection standard: weightings, front of house versus kitchen, links to attendance and performance, links to bonuses.
"This used to take a strong HR person working with operations for a week. I had the whole thing in 20 minutes. The operations manager could announce it to everyone the next day."
He uses AI to review contracts too. Work that used to take ten people now takes two.
The bottleneck is the grain of the data
My biggest takeaway: when AI lands in physical businesses, the bottleneck isn't the technology. It's how fine-grained the data is.
The unglamorous year Eric spent unifying POS, digitising purchasing and rostering was really laying pipes. Without the pipes, the strongest model has nothing to calculate.
For AI founders who want to sell into restaurants, his pointer was specific: ordering and inventory management, and bill-of-materials cost control. Nobody has a great tool for those yet. They may end up building their own with AI.
If you build AI products, this is worth hearing twice. You get a real customer telling you what he thinks, what he uses, and what's still missing.
*This article draws from a podcast conversation with Jerry Liao and Eric Bao of First Bowl and Shiok Burger.