Changing the code in your SP's? Reads, Writes, CPU, and Duration from a trace/profiler session, preferably over multiple runs, the first of which happens after clearing out caches and so on. If you want to look at global impact, overall perfmon counters - if you affect those, tuning was badly needed, but hopefully your normal workload variance is a much bigger source of difference than tuning will be, so you wouldn't see a statistically significant change.

Creating new indexes? If you're at a level where you need to show hard evidence that your chances are a net benefit, then you've got a much tougher row to hoe. In that case, you may need to demonstrate that, with whatever your company's business model weighting is, your new indexes have a net positive effect - i.e. the weighted cost of adding them, keeping them updated with every DML statement, maintaining them (REBUILD/REORGANIZE), storing them, backing them up, and so on and so forth is less than the weighted cost/benefit they have in every statement (not just the ones you're hoping to make better).