The first sign that automation worked may be a larger backlog.
Not a smaller one.
Across different industries, I have seen the same pattern: when a team is slow or expensive to engage, people quietly stop asking it for help. The backlog looks manageable because demand is being filtered by friction.
Then the workflow improves. Triage becomes faster. Analysis becomes cheaper. A request that was not worth two weeks of coordination becomes worth one prompt and a review. The queue fills again — not necessarily because productivity failed, but because the organization can finally see work that users had learned not to request.
This changes the management problem. The goal is not to automate every newly visible request. It is to decide which demand deserves capacity, which should become self-service, and which should remain unserved.
Leverage does not always empty the queue; sometimes it reveals the queue that friction was hiding.