If your enterprise does not upgrade its applications and operating systems except when they buy a new PC, then the critical need is to run Windows Update regularly and keep your anti-virus protection up-to-date. Failed memory, drives, keyboards, mice and monitors can be replaced, but if anything else goes wrong, it is time to buy a new machine. Period.
Because this is the rare exception, however, your enterprise should rely on a three-year lifecycle if your applications or your operating systems are updated as often as the vendor ships a new version. During the those three years, be prepared to increase memory 50- to 100 percent with every OS upgrade and never buy the minimum recommended memory or disk space. (No matter what Microsoft says, 128MB is not enough for Windows XP -- start with 256MB.) Stagger machine purchases so that no more than one-third of the machines are replaced annually. To reduce support issues, make each annual purchase from a single vendor and keep variations from a base configuration to a minimum.
You can live with a four or even five-year lifecycle if your application upgrade needs are minimal but don't expect to get away without adding memory -- and possibly disk space. Don't expect any workstation to last through two OS upgrades. (For instance, no workstation that was sold with Windows NT is sufficiently robust to run Windows XP.)
For security reasons, when buying new, always buy the Professional version of the OS. (Yes, they might be more vulnerable to attack but they are also easier to secure and are more likely to get vendor patches sooner.)
When buying new, you don't have to buy the top-of-the-line model, but it is a good idea to pay attention to speed, RAM, disk space, and the monitor. Most everything else is driven by the needs of specialty applications, or worker mobility. When thinking about speed, get the "next fastest" processor which has the smallest incremental cost for the upgrade. (It will have a longer useful life without having to pay the R&D costs of the "latest & greatest.") Same way with disk space and memory. (You can always add memory and disks later -- as long as you don't waste expansion space with multiple smaller increments. One 256MB DIMM is always more prudent than two 128 MB DIMMS.)
Don't skimp by keeping old monitors, keyboards, or mice around, and don't settle on modest low-end monitors. Your workforce will be much happier (read: more productive) if each new computer has a 17-inch (or better) mid-range monitor and a new keyboard and mouse.
I have yet to see a system upgrade which will really cost less than a new system. Sure, you can get a new motherboard for a four- or five-year-old system, but then you are depending upon old (slower) drives, slower memory, old expansion cards, etc. In short, you are shifting the bottleneck from the processor to the peripherals, but the bottleneck is still there. Further, such upgrades soon turn into a large number of different systems with different drivers from unknown sources -- not to mention different configurations. If the enterprise needs to distribute a single desktop configuration to a significant number of systems, you soon have a support nightmare.
In the end, pick a lifecycle that you can live with, and purchase machines annually over the course of that lifecycle. If each annual purchase is made up of nearly identical machines from a single vendor, support is limited to three or four platforms. The only decision then is "Do I need a three-year lifecycle, or can I live with four? Or five?" Any lifecycle shorter than three years and your enterprise is wasting its money. Longer than five and your enterprise is wasting its time.
C. Marc Wagner
Services Development Specialist
UITS, Student Technology Centers
Indiana University, Bloomington