A consultant identifies that a query is scanning a very large dataset on every dashboard load. What pre-aggregation or dataset design changes would most improve performance?