How to prevent database deadlocks caused by heavy analytics queries during peak hours

Our main database began throwing timeout errors during afternoon traffic because heavy reporting queries were locking transaction tables. We first tried to optimize each query. The real fix was creating a separate read replica for all reporting tasks. Moving analytics traffic away, from the transactional server lowered our error rates to zero right away. Ignoring infrastructure isolation as the user base grows leads to production outages. How do your systems handle heavy reporting queries without disrupting the core user experience?