Early on, one of our testers called PowerDataChat "a chart generator". It was a fair description. It could turn a question into a correct chart, but it did not know anything about the business behind the data.
What changed it was context, not a bigger model.
Data descriptions. A short description of the file and each column (which the AI can draft and a person confirms) tells the system that amt_out is the outstanding loan amount in GEL, not a random number.
Domain skills. In the Enterprise edition, skills carry industry vocabulary, KPI definitions and calculation rules. The banking and microfinance skill knows PAR30, NPL and IFRS 9 stages, and rules like "never sum across currencies" and "weight averages by outstanding amount".
Company vocabulary.…