Capacity Planning

Our company’s CIO is looking to develop a method to accurately determine the health of our current NT network, and the best ways to improve upon it with the least amount of capital. What ideas do you have to determine the health of servers (e.g.: using perf.mon stats)? What stats should be collected for each server, and how frequently? What are the warning levels for servers that are nearing their capacity limit, or conversely, are being under-utilized? Are there any tools you know of that can collect these stats and analyze them with minimal data manipulation? Any tools or ideas for determining capacity in an NT environment are greatly appreciated.