Glossary
OLAP
Meaning and details
OLAP, online analytical processing, is reading for analysis. Instead of touching one row at a time, analytical queries scan millions of rows to aggregate, slice and compare: revenue by cohort, retention by signup month,...
OLAP meaning
The analytical query workload: scanning and aggregating large historical datasets, usually in columnar storage.
What is OLAP?
OLAP, online analytical processing, is reading for analysis. Instead of touching one row at a time, analytical queries scan millions of rows to aggregate, slice and compare: revenue by cohort, retention by signup month, usage by plan. The storage that serves this well is columnar, compressed and batch-oriented, the opposite shape of a row-store transactional database.
Why it matters to an indie builder: your application database is built for writes and point reads. The moment you ask it a real business question over a year of history, you hurt: the query is slow, and worse, it evicts the cache your checkout flow depends on. Analytics belongs on a copy, and the modern indie answer is a columnar engine like DuckDB (embedded, free) or a managed warehouse like MotherDuck or BigQuery.
A concrete example: churn by monthly cohort over 40 million event rows. On your production Postgres, that scan takes 45 seconds and triples page latency while it runs. Export the same data to DuckDB: the columnar engine reads only the two columns involved, the query lands in 0.4 seconds on a laptop, and production never notices. This exact pattern, a nightly export into DuckDB queried from a notebook or Metabase, is the 2026 indie analytics stack.
Common mistakes: running month-end reports on the primary database (the classic self-inflicted outage), building heavy ETL before a cron-and-SQL pipeline has proven the questions you actually ask, skipping incremental exports until datasets outgrow memory, and buying warehouse capacity for data that fits in a compressed columnar file on your own machine. The analytics blocks on IndieBoot list the tools that make each step cheap.
A note on when to graduate: if queries stay under a second and run a few times a day, DuckDB on your laptop is the whole stack. Move to a managed warehouse when data outgrows memory, when more than two people need simultaneous access, or when dashboards need to refresh on a schedule nobody babysits. Until one of those bites, the warehouse bill is paying for a problem you do not have.