Building BI-ready datasets

A BI-ready dataset is a table or view that a dashboard tool (Power BI, Tableau, Looker Studio, Metabase …) reads directly. The person using it will filter, group and sum its columns with clicks, not SQL, so the dataset has to make the right answer the easy one. A checklist:

• One declared grain, unique: one row per day × channel, one row per customer. Put it in the name or the description, and test it.
• Keys and readable attributes: the date as a real date, names rather than codes (channel_name, not channel_key), and the attributes users will filter by.
• Additive measures (orders, units, net_sales, returned_net_sales) rather than ratios. The tool can compute net_sales / orders at any level; a stored average order value would be summed or averaged incorrectly.
• Special cases resolved explicitly: guest checkouts labelled ('Guest checkout' instead of a NULL region), test accounts excluded, missing values as 0 where 0 is the truth.
• Clear, stable names: lower_snake_case, units in the name when ambiguous (_sar, _pct), a prefix that marks the object as published for reporting (rpt_).