when do you finally pull the trigger on a distributed tracing system?

Lately, I've been trying to figure out how requests move between our microservices. Tracking down problems feels like a detective job, but half the clues are missing. We're getting to the point where just SSHing into servers and checking logs isn't enough. When did you all decide it was time to spend money on something like Jaeger or Zipkin? What were the big issues that made the extra work worthwhile?