What is data granularity?
Data granularity is the level of detail at which data is recorded or summarized. Fine-grained data captures each event: one row per sales transaction, one reading per second from a vibration sensor, one line per call in a contact center. Coarse-grained data rolls those events up into totals or averages by day, month, store, or region. The grain of a table is what one row represents, and every table should have exactly one.
Aggregation runs in one direction. Fine data can always be summed into coarse data, but a monthly total cannot be split back into the days that produced it. Storage and processing costs rise with detail, which is why older data warehouses kept summaries and discarded the raw events.
Why data granularity matters
Models need the grain that matches the decision. A grocery chain forecasting replenishment needs store-item-day sales, because a region-week total hides the Friday spike at one location. The finance team reviewing the same business is content with region-month. If both teams share one dataset, the grain must be chosen for the finer need and summarized for the coarser one.
Mixed grain also produces wrong numbers. Joining daily sales to a monthly budget repeats the budget on every day of the month. Lakehouse designs keep raw events in a landing layer and build summaries on top, so both needs are served from one copy. The data model should state the grain of each table so that analysts and text-to-SQL assistants do not have to guess it.