Data model
A data model is the governed tables a data team reshapes raw data into, which it tests, documents, and owns.
A data model rests on four parts: tables that each hold one kind of record, keys (the column that identifies each row), a grain that sets what one row stands for, and joins that match two tables on a key they share. The data team sets and tests each part, so an agent does not have to guess how tables join.
The data model carries spend and revenue by channel, but return on ad spend (ROAS), a calculation on top, sits in a dashboard, a spreadsheet, or a saved query until a semantic layer holds it once. A semantic layer built on raw tables repeats their errors in every answer, so build the data model first.
For an AI agent, build it finer than your reporting has ever used. A dashboard can run on totals, but an agent needs each order, each ad, each customer to find trends, segment customers, and forecast.
Example
For an online brand, the minimum is marketing, customer, product, and order tables. Each order row carries the channel the attribution rule credits, so an agent can answer which channel brought in the customers who later bought from a second product category.