Under the hood
Why a graph beats a spreadsheet for restaurant questions
Your sales and labour data already exists, and you can already open it in a spreadsheet. So why model it as a graph? Because the questions worth asking are almost never about a single row — they're about how rows connect.
Spreadsheets are great at rows, bad at relationships
A sales export is one row per line item: a Carbonara sold on a check, with its major group, family name, order type and time. A labour export is one row per shift: a Chef on from 11:39 to 20:07 at a given rate. Each file is fine on its own.
The moment your question spans both — "did the kitchen roster match the covers it actually had to produce?" — you're joining two datasets on time windows, then grouping by position, then by menu family, then comparing. Every new angle is another pivot, another VLOOKUP, another tab. The data supports the answer; the tool just makes it slow to reach.
A graph stores the connections as first-class things
Instead of tables you join on demand, a graph stores the relationships directly:
(Menu item) → sold in → (Order type)
(Order type) → happened during → (Service round)
(Service round) → staffed by → (Shift / position)
Once the data is shaped this way, a question like "which positions were on during rounds where Sides barely sold?" is a short walk along edges that already exist — not a multi-step rebuild. The relationships you'd otherwise reconstruct by hand are simply there.
Why this matters for asking in plain English
This is also what makes natural-language questions reliable. When an LLM translates "why was SPLH low last Tuesday?" into a query, a graph gives it clean paths to follow between covers, menu mix, timing and labour. The model isn't guessing how to join five tables correctly — it's traversing relationships that were defined up front. That's a big part of why the answers stay grounded rather than invented →.
Spreadsheets store the data. A graph stores the questions you'll want to ask of it.
None of this changes what you send us. You export the same sales and labour files you already produce; the graph is something we build on our side. You just get to ask better questions of it.