The most frequent example of that sort of reasoning is comparing computerized work with the identical work done by hand. For example, the CIO of an insurance company would show how much staff would have to be added to issue insurance policies without the system he's advocating.

This method is not only bad reasoning but also bad economics. What's really going on here? In effect, a CIO would calculate the ROI by assuming that all computers in the company have suddenly ceased to function. The ROI would be the ratio of the increased costs for a fictional staff divided by the actual costs for real IT spending. It's always a very large number, which gives the proponents a great deal of satisfaction.

I'm not using a ridiculous example to demonstrate foolishness. There are numerous books, including a cost-justification manual published by a major vendor, that suggest using such ratios to justify asking for real money. Some widely quoted cases using such reasoning come up with returns of more than 10,000%. But that's making false promises based on false premises. It's comparing the economics of a current technology with imaginary labor costs or an obsolete technology.

If you tried to cost-justify buying a 460-horsepower Mercedes automobile by comparing it with the cost of owning 460 horses, you'd be laughed at. It's obvious you have to compare the cost of one automotive technology - a Mercedes - with other cars that use the latest technology, such as a Lexus, a Lincoln or even a Honda. Yet in the IT world, I continue to find that IS managers, when judging computer applications, are comparing cars with horses. Top management is catching up to such deceptions and starting to demand proof that they will realize plenty of real cash in return for cash spent on IT.

The ROI of Two-Digit Negligence

The apologists' argument starts with the observation that the physical limitations of punched cards required conservation of column space on an 80-hole card. That limitation was finally overcome with the introduction of disk storage.

Next, the apologists say that in 1963 disk storage cost $10,600 per megabyte per year, making the cost of storing the extra two digits prohibitive. Clearly, the old-timers were wise to keep the size of the computer records to a minimum; a large corporation could avoid spending several hundred million dollars on costly disk space over 35 years.

Using that logic, the big fix-up bill now coming due seems to make perfectly good economic sense. The problem is that despite the smoothness of the exculpation claims, we're back to comparing a Mercedes with horses. The economics-made-me-do-it arguments don't hold up:

1. It isn't true that in 1963 the alternative to punched cards was mainframe disk memory. The fact is that with completely negligible exceptions, all computer records that didn't remain on cards were kept on magnetic tapes. The cost of a 2,400-ft. reel of 1,600 bit/in. magnetic tape in those days was less than $20, or a mere $0.0000069444 for two digits. That's less than $6 per megabyte for a one-time cost, not $10,600.

2. In fact, disk storage in 1963 didn't cost $10,600 per megabyte. The actual cost was about $2,000. The calculations of the payoffs from deferring year 2000 fixes restated 1963 costs in terms of 1995 inflation-adjusted dollars. That's achieved by applying to 1963 prices a multiplier concocted by the U.S. Bureau of Labor Statistics. That adjustment is based on a ratio of the prices of old technology compared with current prices. I have examined that ratio and found it to be a figment of a bureaucrat's imagination. It's like increasing the size of a weak horse fivefold so that it can run in competition with a much faster horse.

   Decisions can't be made in retrospect. Therefore, the act of omitting two digits from archival storage can be evaluated only in terms of actual 1963 cash economics and not by making comparisons with 1995 possibilities.

3. It's grossly misleading to justify delaying year 2000 conversion work because disks were expensive 35 years ago. For a typical corporation, software maintenance costs range from 20% to 30% of annual spending for software. That implies an average software life expectancy of less than five years. Each organization had at least six chances to make the necessary year 2000 repairs as a by-product of regular software maintenance, at a low cost, instead of deferring it until the bills accumulated.

Implications for CIOs

CIOs shouldn't go along, through temptation or ignorance, with these pseudo-ROI methods. They may have to devote more time to finding out how to distinguish between easy answers and sound financial analysis. Success as a CIO is likely to depend more on one's credibility as an investor who can deliver identifiable cash results than on one's talents as a financial spin doctor. Conjectures won't stand up.