A board question needs a number nobody can produce before the meeting. Someone exports from the ERP, builds a pivot table and emails it round. The follow-up needs a different cut, so the cycle starts again — and before long the second question stops being asked.
Where the answer comes from
You ask Dima in plain language and get a plain-language answer. But the model does not supply the number: it translates the question and writes the sentence, nothing more. The figure comes from the semantic layer — in your own measure names, by your own definitions: the screen shows scrap rate, not uretim_fire. Queries are read-only and planned before they run; they cannot change your data.
Every figure carries its source
Every answer says which cube it came from, from how many rows, and when. You can check a number without asking us; the source ships with the answer. And if you asked the wrong question, the second one is free: answers arrive fast enough to keep asking.
Your ERP, spreadsheets and databases stay where they are. Dima connects to them and builds a semantic layer over them with you, modelling how each measure is calculated in your own definitions. When the setup ends, the layer is yours.
