Scanned Bytes
The volume a query reads, which is what most analytical engines bill for and what almost every optimisation ultimately reduces.
In a separated storage and compute architecture, cost tracks data read rather than rows returned. A query returning a single number can be the most expensive one that day if it scanned the whole table to compute it.
That reframing makes the levers obvious. Partition pruning skips files. Columnar
formats skip columns, so SELECT * in an exploratory query costs many times a
targeted select. Clustering within partitions improves skipping further.
Materialising an expensive aggregation trades storage for repeated scan cost, and
is the right trade for anything queried hourly.
The organisational half matters more than the technical one. Cost per query, attributed to a team and visible to them, changes behaviour faster than any optimisation programme. The recurring discovery is a dashboard set to refresh every five minutes against a full table scan, built by someone who left, watched by nobody — and it is usually a material share of the bill.