ClickHouse for product analytics: when it earns its keep
ClickHouse is the right answer when your event volume passes ~100M/month and you want sub-second analytical queries. Below that, Postgres is fine and your engineering time is better spent elsewhere.
ClickHouse is having a moment, mostly because Postgres isn't great at analytical workloads at scale. After shipping it on two products, here's the honest read.
It earns its keep above ~100M events per month, when analytical queries (funnels, retention, cohort analysis) start taking minutes on Postgres and seconds on ClickHouse. The columnar storage and vectorised execution genuinely change what's possible at the dashboard layer.
Below that scale, the operational tax doesn't pay off. ClickHouse is a different mental model than Postgres — joins are expensive, eventual consistency on inserts is real, and the SQL dialect has gotchas. If you can serve your analytical queries from a Postgres replica, do that. If you can't, ClickHouse is the right answer.