How Do You Decide When to Stop Improving a Strategy?

There is always another filter, indicator, or rule that could potentially be added to a trading system. How do you know when further changes are likely to improve the method rather than make it unnecessarily complicated?