Trust
"The data can't support that" — why that's a feature
The most useful thing a data system can say is sometimes "I don't know." It's also the rarest. Most tools are built to always return a number — and a number always looks like an answer, even when it's resting on three Mondays of data and a hopeful average.
What "grounded" actually means
A grounded answer has two properties. First, every figure in it comes from your historic data — not from the model's general knowledge of how restaurants tend to behave. Second, the answer shows its working: the covers, the labour hours, the attach rates it used, so you can check it.
Why admitting limits is the valuable part
Consider the question "is Wine revenue high enough on Mondays to justify the extra Service shift?" If your sample only contains three Mondays that carried that shift, there simply isn't enough signal to answer reliably. A system optimised to always respond will average those three and hand you a confident-sounding figure. That's not an answer — it's a liability, because you might roster against it.
Exactabase is built to flag this instead: "Only 3 Mondays in the sample carried that shift — too few to be reliable." It then suggests revisiting the question once more data has accumulated and there are enough Mondays to answer it properly.
A tool that's honest about what it can't see is one you can actually trust with what it can.
How we keep it honest
This isn't a claim you have to take on trust either. Every answer can be traced back to the rows it came from, and we validate Exactabase against a structured set of test questions with hand-verified results — so you can see exactly where it excels and where it holds back.
Grounded answers and a willingness to say "not enough data" are the same discipline, pointed in two directions. Together they're what makes the numbers worth acting on.