A common justification for VoIP is toll-bypass. In this, it is generally assumed that VoIP calls traversing your corporate data network (the WAN, specifically) are either "free" or "cheaper than long-distance," and that difference will be more than the cost of the VoIP equipment and project.
This assumption is often made because data circuits are usually not metered, while long-distance voice service typically is -- meaning that you pay a per-minute usage charge for voice in addition to the assortment of fixed charges and fees, as compared with data circuits, where you pay for a certain amount of bandwidth whether you use it or not. People frequently calculate ROI by comparing the total long-distance bill (which is easily quantified) with the VoIP project costs, without taking into account the costs of the WAN transport itself.
In other words, it's typically difficult for people to actually quantify their costs for VoIP minutes because the data WAN circuits are a sunk cost -- you will pay for the bandwidth whether you use it or not. This is complicated by the fact that VoIP calls, like other data, vary in frequency and duration -- making it difficult to quantify utilization on any given circuit -- and further complicated by the fact that a given VoIP call can traverse zero, one, or many physical WAN circuits and can be transcoded at various points along the way. It might use 26Kbps of bandwidth on one circuit and 80Kbps on another. And, of course, your call detail records don't show the physical path.
All this leads to some simple, but practical, attempts to guess the cost of WAN transport. The simplest method is probably to take the size of your priority queue as a portion of the total cost of the circuit. Another option is collecting remote network monitoring (RMON) utilization data for the priority queue as a fraction of total utilization or total bandwidth and comparing that with the cost of the circuit. These methods are obviously flawed, for several reasons, but probably put you in the ballpark.
The point here is that whichever method is most appropriate for your situation, you don't want to underestimate this cost, because it might surprise you. In many cases, particularly domestic U.S., long-distance rates are so competitive that they may actually be cheaper than IP bandwidth. This means that even if you're deploying a fancy new softswitch and IP phones to all your users, you still may be better off routing calls to a long-distance carrier rather than over your WAN.
Somewhat more specifically, the ROI will look best when your anticipated VoIP usage fits easily in available space on your existing circuits. Then, you're really just using a sunk-cost resource a little more efficiently. But if you have enough traffic to justify it, an upgrade is worth investigating.
Tom Lancaster, CCIE# 8829 CNX# 1105, is a consultant with 15 years experience in the networking industry, and co-author of several books on networking, most recently, CCSPTM: Secure PIX and Secure VPN Study Guide published by Sybex.