Kafka is often a slide deck decision. Three questions are enough: is daily volume actually millions+, do you need days of replay and retention, do you have a team to run the cluster?
For small teams, RabbitMQ and transactional outbox cover most cases. Kafka adds cluster ops, lag monitoring, consumer group rebalances, schema evolution debates. Wrong tool choice means hidden headcount just to keep Kafka alive.
Throughput slides don't show operational cost. Choose based on volume and team capacity.