Nobody builds costing software for a two-person restaurant

Restaurant costing software is built for operations with a general manager, an inventory count and enough volume to justify a subscription stack. Below that line there is no tier. Dr. Nikki Siso, founder of Conscious Kitchen in Westlake Hills outside Austin, evaluated a product at around $150 a month, found it more than her budget allowed, and opened a restaurant whose build-out came in at $85,000 against a $30,000 plan.

This is the one category in Artificial Reality's Season 1 research where the software problem is genuinely solved and the operator still could not use it.

What did the operator actually ask for?

Five requirements, all stated by Siso rather than inferred:

• Convert reliably between units — pound to cup, cup to tablespoon, tablespoon to teaspoon.
• Divide by yield, so a pie becomes twelve slice costs.
• Update when ingredient prices change, without rebuilding the model.
• Produce accurate, trustworthy output — the general-purpose attempt duplicated rows and returned wrong values.
• Be priced for a small restaurant on a tight budget.

Artificial Reality adds three, as analysis rather than operator testimony: handle raw and small-batch production, where yields vary more than in standardized commercial cooking; account for waste and trim, which is where theoretical and actual food cost diverge; and require little enough maintenance to survive a two-person kitchen, because a costing model that goes stale is worse than a spreadsheet, since it is trusted.

Why the general-purpose substitute failed

She tried to build it herself and it did not hold. The task was arithmetic: a pound of almonds at $24, one cup per recipe, twelve slices per pie, maintainable as prices move.

"I tried to create that full chart breakdown so that I can adjust as the price of almonds goes up and down. I can adjust the price, and then it adjusts it."

"I had one hell of a time getting it to actually fill in the spaces on the Excel chart. For some reason it did not want to fill in the blanks. Or it would just duplicate and it would be the wrong values."

Galina Fendikevich put it to her that AI had created more work than doing it by hand. Siso's answer: "Yes." That failure inverts the usual story — the same operator used a general-purpose model to review a forty-five-page commercial lease, source commercial refrigeration and develop herbal formulations, and could not get it to divide a pie into twelve slices reliably.

Is the tier below the subscription stack a real market?

It is a fair question and the honest answer is unresolved. Low willingness to pay, high support burden, and a high failure rate among the customers. Against that: the two-person restaurant, the food truck and the bakery have the same arithmetic problem as a multi-site group, and getting food cost wrong is most likely to close a business at exactly this size.

It may be that the answer is not a cheaper version of the existing platforms but a genuinely reliable spreadsheet template — the thing Siso was trying to build when the model gave her duplicated rows and wrong values. That is a smaller product than a platform, and it is the one with documented demand attached to it.

Costing and forecasting are not the same product

Siso separates them, and the distinction is the sharpest thinking in this part of the interview.

"Forecasting was a complete guessing game ... this might not be something that can be resolved, because every restaurant's very different. We are not a Flower Child that has 50 orders for lunch or even more. We're very nuanced, very small batches. So I don't know if forecasting could ever be perfected — but budgeting I could have had a lot more support with. I didn't know the cost of all the fridges and the sinks and the insurance."

She does not believe demand for a novel small-batch concept can be predicted, and she is probably right. She does believe someone should have been able to tell her what a commercial fridge costs. One of those is a hard modelling problem. The other is a list — and the gap between the $30,000 plan and the $85,000 outcome is what the missing list costs.

What to build first

Build the boring half. A setup budgeting tool that prices refrigeration, sinks, insurance and permits for a specific city has a checkable answer, no sales history requirement and a documented gap between a $30,000 plan and an $85,000 outcome to point at. Nobody in this operator's account offered her one.

If you are already in costing, the product decision is narrower: publish a price for a two-person operation, and prove the conversion chain end to end — pound to cup to tablespoon, divided by yield, updating everywhere when one ingredient price moves. That is the exact arithmetic that broke the free alternative, which means it is also the demo that wins against it.

The question for a founder in this category: is your cheapest tier priced for the restaurant that most needs to know its food cost, or for the one that can already afford to be wrong?

Where this comes from

S1E6: I trusted AI to build my restaurant, it cost me $30,000 — the full interview with Dr. Nikki Siso. Listen or watch: YouTube, Spotify or Apple Podcasts.