It depends on the shape of your queries. For high-volume append-only events with wide aggregations, a columnar store like ClickHouse will outrun a row store by a wide margin. If you also need transactional writes, keep Postgres as the system of record and stream into the analytics store. Start with one, measure, and split only when the read patterns clearly diverge.

Go with a columnar store such as ClickHouse for the analytics reads, and keep Postgres for transactional data. Stream events between them. Revisit only if the query patterns diverge.