A restaurant you never walk into

Context

Alongside the back office, I took a new venture from an idea to the edge of launch: delivery-only restaurants. No dining room, no servers, no storefront. You rent a kitchen in a shared commercial building, and drivers pick up orders for DoorDash and Uber Eats. The customer never sees the building.

We planned two brands out of one kitchen:

Hidden Sichuan, the lead brand, serving mala xiang guo, a spicy Sichuan stir-fry. It’s bold, and it travels well in a delivery bag.
Hidden Namcheon, the second brand, serving Busan-style pork bone soup. It’s comfort food for a different time of day and a different customer, but it needs insulated packaging to survive delivery.

The problem

“Let’s open a restaurant” isn’t a reason. Most restaurants fail, and delivery-only restaurants have failed in large numbers. Before anything else, the venture needed a reason it should work for us specifically.

What I did

The reason

There were three parts to it.

We already make the hard part. Most restaurants live or die on the kitchen: you need a real chef, consistency is hard, and labor is expensive. But our company already makes marinated proteins in a factory, cooked consistently and at scale. So our kitchen only assembles. It doesn’t need a chef, which means it can be repeated, which means it could eventually be franchised.
It’s a second customer for products we already make. The same spicy pork that goes into a supermarket freezer bag goes into a delivery bowl. We weren’t building a supply chain. We were pointing the existing one at a new customer.
The goal is a brand, not a restaurant. The long-term plan was to prove the concept in one kitchen, then license the brand and product into other restaurants’ idle kitchens. They’ve already paid for rent and equipment, and they have quiet hours to fill. Eventually the brand becomes something that can be sold.

That also changed how the brands related to the company. I recommended moving the retail mala packaging from the JuicyBite name to Hidden Sichuan, as “Hidden Sichuan by JuicyBite,” so that Korkio and JuicyBite stay the manufacturing backbone and each consumer brand could stand on its own.

Testing the reason

I compared the plan against real companies instead of optimism.

Wow Bao was the closest working example: a small restaurant chain that turned into a frozen-product and licensed-kitchen business, now in roughly 800 delivery kitchens and thousands of grocery stores, run by a very small team.
Nathan’s Famous showed what the model looks like when it’s mature: a large share of the income comes from licensing the brand, not from selling hot dogs.
Peckwater Brands was the failure case. I broke down why it collapsed: brands nobody cared about, partners who couldn’t see what was happening, kitchens that didn’t make money for their hosts, money that ran out, and a structure too fragile to survive any of it.

I tested six hypotheses about why our version could work. Most held up. One didn’t: I had claimed that mala was an empty category in the US, the way bao had been for Wow Bao. It isn’t, and I wrote that correction into the report instead of quietly dropping it.

Getting it ready

Site. I scored six candidate locations against seven weighted criteria rather than going on instinct. Times Square came first and Long Island City second. When I added pickup orders, which avoid delivery-app commissions, I re-weighted the scorecard and checked that the ranking held.
Terms. I went into the negotiation with the shared-kitchen operator with twelve prepared questions, and came out with real numbers: rent for each site (Times Square cost about 30% more a month), delivery commissions by platform, and confirmation that a personal guarantee was required.
Trademarks. I prepared and filed two trademark applications with the US Patent and Trademark Office, in Korkio’s name, myself.
Structure. Both brands would operate under the existing company rather than as new legal entities. Simpler, cheaper, faster.
Rules. New York requires a certified food-protection supervisor on site whenever a kitchen is open, and the common national certificate doesn’t count. So two people would need the New York certificate to cover every shift.
Model. I built one workbook connecting menu, pricing, marketing, staffing and profit and loss, about 700 live formulas, so changing any assumption updated everything else.
Marketing. The first budget set marketing as a percentage of revenue, which tells you nothing. I rebuilt it from the bottom up, so every dollar traced back to orders per day, and organized it into layers: being found in the apps, creating demand outside them, converting, bringing people back, and expanding. The target was about 70 orders a day and $60,000 a month by month seven, with a ceiling on what we’d pay to acquire each customer.

The thing that would have broken the plan

Our student ambassador program was built around students sharing personal promo codes. Before we built it, I checked whether the delivery apps actually allowed that. They don’t let merchants issue their own codes. If we had found out at launch, we would have had to redesign the program under pressure. Instead, I redesigned it early around group orders, such as club catering, with rebates on the receipt.

Where it stood

By the end of the summer, the only things left were the final choice between the two sites, testing whether the food survived thirty minutes in a bag without leaking or going cold, the two certifications, and launch.

What I learned

Honestly testing an idea is better than defending it. The most useful page in the whole report was the one where I corrected my own claim.

A model’s job isn’t to be right. It’s to show you how the business would feel. When the marketing budget turned into “this many orders a day,” the numbers finally meant something. And because a single assumption, like the share of pickup orders, could flip the conclusion, I kept every assumption in its own cell where anyone could change it.

Finally, a brand gets more valuable as it gets narrower. One dish, done clearly, is something you can sell.