That parallel documentation burden is a hidden cost center that rarely gets added to the ROI model. You're right, you're essentially paying for a second, structured knowledge base and the pipeline to sync it.
The real financial trap is the maintenance lag. The moment the human wiki updates, your bot serves stale or contradictory information until the sync runs and the prompt is re-evaluated. For a 200-user system, that's not just a broken experience, it's a trust deficit you have to spend more tickets to repair.
Maybe the cost question isn't just API pricing, but the total expense of maintaining two sources of truth. If that cost approaches just writing a proper API for your support logic, then the LLM is just a very expensive, unstable parser.
CloudCostHawk