Deployment record
In the tracker →- Organisation
- Angela Moznerova's restaurant
- System
- AI Finance Tool
- Stage
- Pilot
- Function
- costing model
- Built on
- Claude, Python
- 01Angela Moznerova, a fractional CFO and restaurant owner, developed an AI finance tool to streamline cost analysis for her restaurant, which previously struggled with inconsistent invoices and undocumented recipes.
- 02The AI tool converts invoices and recipe units into a cost model, saving Moznerova about four hours a week in manual work.
- 03The system flags significant discrepancies in ingredient usage, helping to prevent costly pricing errors in menu items.
Show all 5Show less
- 04Moznerova identified a potential vendor price gap worth up to $2,100 annually but retains control over vendor selection to ensure quality.
- 05Small businesses often face barriers to adopting such tools due to costs and a lack of trust in untraceable numbers, prompting Moznerova to design her system for transparency.
Follow Deployed on Google
For years, Angela Moznerova could not justify the cost of working out what each plate at her restaurant cost. Moznerova, a fractional CFO who also owns a restaurant in Brooklyn, told Deployed by AIM Media House in an exclusive interview that two things stood in the way: the invoices and the recipes.
Her small vendors each send invoices differently, some without digital copies, and the restaurant had no documented recipes. Excel meant doing the work herself or paying an analyst.
The inventory module in her point-of-sale system required entering every ingredient, unit and recipe by hand, and did not read invoices.
"I couldn't justify the time or the cost," she said.
Using AI as her junior analyst, Moznerova built the first working version of a costing model in about four hours, and it now saves her about four hours a week.
It turns invoices into a single price book and converts recipes, including units such as "a pinch," into a plate-by-plate cost model. Her chef fills in and checks recipes in a shared Google Sheet.
Where AI Stops
Moznerova divided the work between AI and code. Claude reads the messy inputs, turns invoices and recipe units into clean tables, drafts each Python step from her prompts, and runs a daily sales skill.
Anything that produces a dollar figure stays in Python, with a test that has to pass and a saved run history, so each run can be compared with the last. Any plate cost that moves more than 15% is flagged.
She set that rule after seeing where AI went wrong. Fuzzy matching paired "Sage" with "Sausage," and an AI estimate of a wine price was off, so she replaced both with an explicit name map and the invoice price.
"AI can read, draft and suggest, but it doesn't get to make up a number," she said. Pricing and vendor decisions stay with her and her chef.
The Errors the Checks Caught
The model flagged a pasta plate whose recipe sheet listed 3 oz of pepper. Across delivery sales alone, that would have used 21 lb, against 5 lb on the invoices loaded into the system.
The week's store, brand and commerce deployments. Fridays.
Her chef weighed a portion: about 3 grams of pepper, and 30 grams of pecorino rather than the 5 oz on the sheet.
As written, the dish looked like a 33% food-cost problem. With the real weights, it is about 13% of the delivery net price. Moznerova said a repricing or menu decision based on the sheet would have been wrong.
The biggest rework came from a batch recipe whose name said 60 liters, which the model took as the yield. That was the pot size. The real yield was 15 liters, which moved one ragu from under $1 to about $4 a plate.
"Speed raises the stakes: a wrong input becomes a polished number in minutes," she said.
Decisions So Far
The tool is still in progress, and savings have not been measured yet. Every DoorDash and Grubhub order carries a fixed driver cost of about $6 to $7, leaving small orders with almost nothing.
She has set Grubhub's minimum at $35, the highest it allows, and asked DoorDash for $45, roughly where an order starts to earn the restaurant's target margin. She will review results over four weeks.
The system also found a vendor price gap worth up to $1,400 to $2,100 a year, but she does not let it pick vendors, because the cheaper vendor does not always have the item.
Her software costs are an AI subscription plus free tiers of other tools, compared with roughly $350 to $750 a month for larger paid restaurant tools. Updating recipe costs takes about 20 to 25 minutes a week.
What Holds Small Businesses Back
Moznerova said small businesses are held back first by cost and then by trust. Most have only an accountant or outside finance support, and owners are wary of numbers they cannot trace.
"That's why I built mine so every number can be checked," she said.
She would test the system with two or three other owners before building further. Her advice to CFOs still in pilot mode is to pick one decision someone makes every week, standardize the data for it in one location, and write down in advance what would make them scale or stop.
Last updated
