Innovation is usually thought to be a good thing, but often it's not. In fact, innovation can be a really bad thing, which is one reason why so many innovative products disappear without trace. They are simply more trouble than they are worth.

Rob Wilmot, who was once parachuted in to save ICL, Britain's biggest computer corporation, put it most succinctly. "Added value," he said, "is the same as subtracted value."

Every company on earth is apparently devoted to "adding value" in order to enhance its products, to make them stand out from the crowd. What they really want to do is find a way to make you buy their product instead of somebody else's. The issues that actually drive the corporate agenda in the IT industries therefore include: how do we develop something that makes it hard for users to switch to a competitor, or for a competitor to clone us and compete with us? Their weapons include patents, copyrights and other intellectual property rights, and especially the proprietary control of the hardware, software and peripheral interfaces that affect compatibility.

One of the most dramatic examples of this type of innovation was the launch of the IBM PS/2 personal computer and OS/2 operating system (co-developed with Microsoft) in 1987. IBM had controlled the data processing market for decades, but gave away its monopoly power by bringing in Intel and Microsoft to provide key components for the IBM PC. The PS/2 was an attempt to wrest that control back.

Of course, the PS/2's new "magic ingredient", the Micro Channel Architecture (MCA) expansion bus, was promoted as being technically superior to the existing bus. As, indeed, it was. OS/2 was also far superior to the PC-standard MS-DOS. On these grounds, the PS/2's "innovations" won the support of many technical experts who were either too clueless to see what was going on, or had an inadequate understanding of IBM's motives and modus operandi. However, they worried customers, who mostly stuck with existing PC standards that were good enough.

By 1993, IBM's PC operation was stuffing the distribution channels with PCs no one wanted to buy, and losing a $1bn a year. It appointed a new manager, Bruce Claflin, who gutted the company, moved the development lab, killed the PS/1, PS/2 and ValuePoint lines, and dropped the MCA bus, turning the division profitable in a year. Eventually, IBM went back to producing what were still called "IBM compatible" PCs. The IBM PC Company never recovered its old prominence on the desktop, but at least it succeeded in notebook computing with its ThinkPad line.

Which isn't to say that the ThinkPad was not innovative. However, the innovations came in things like colour and finish, screen size, the new TrackPoint pointing device and short-lived "butterfly keyboard", bundled software, price (low by IBM standards), marketing and support. The ThinkPad innovated in areas that were valued by customers, and customers were therefore prepared to pay for them. However, it did it without departing too far from accepted industry standards, which would have made customers reject it as "incompatible". Lesson learned.

The problem with pointless technical innovation is that it is trivially easy to do, especially in computing, but hugely expensive for users.

Innovation is easy because once you have a platform, hindsight can always find ways to improve it. And the longer a platform sticks around - and that can mean five, 10, 20 years or more - the easier it becomes to see better ways of doing things. But if those innovations introduce incompatibilities, they become increasingly expensive for the market to absorb.

Obviously, the companies that either made or used PCs didn't think ISA (Industry Standard Architecture - a politically correct name for the old IBM PC AT bus) was the greatest expansion bus ever invented. However, the fact that it was a common standard meant that, in terms of "the whole product", it was still better than MCA. In purely practical terms, the availability of cheap expansion cards, software, support and other factors -- especially the millions of PCs already in use -- outweighed MCA's theoretical advantages. (See the column published on January 25 for an explanation of "the whole product" concept.)

If we could start again from scratch, with hindsight, we might well decide to adopt the MCA bus, or something similar. Since we are not starting from scratch, we have to consider the switching costs. And unless the benefits are much bigger than the switching costs, we are not going to switch. Certainly we are not going to switch every six months, or every time some manufacturer brings an "improved" but incompatible system to market.

Of course, markets do switch. The IT industry has changed dramatically over the past 40 years, moving from large centralised machines (mainframes with dumb terminals) to decentralised client/server systems (mainframes, minicomputers, and other servers talking to PCs and other smart terminals), and it is still changing. The games console industry makes a platform switch every five years or so.

Backwards compatibility helps. Users are more likely to adopt innovations if they can keep the benefits of their existing platform - hardware, software, data, or all three - rather than throw away their past investments. But even that is possible. For example, the switch from vinyl LPs to CDs has taken two decades, but it does show that even an entrenched, ubiquitous standard can be displaced by an incompatible newcomer.

The key to understanding such changes is part of the whole product model, because "the whole product" also includes "the future product", or reasonable assumptions of what the future product might be.

So suppose, for example, that a company introduces an innovative product which departs from the market standard. Suppose also that the innovation delivers enough benefits to cover the switching costs. This still does not mean that the market will switch. It won't if users believe that the standard platform will be developed to provide equivalent benefits.

To take an IT example, most users would not switch from using Intel processors to using Motorola processors just to get a speed increase, or to move from a 16-bit to a 32-bit platform. When it was selling the 68000 line used in the Apple Macintosh, Atari ST and Commodore Amiga, Motorola seriously thought they would. But switching would have meant users changing their hardware and software and converting their data - a hugely expensive proposition. What users actually did was defer the benefits until Intel, as expected, delivered equivalent chips compatible with market standards.

Intel offers "road maps" that provide a detailed guide to where the company is going, but users might have had a different view of "the future product". They might have decided that not only were there benefits to switching, but that similar benefits could not be delivered in a reasonable time on the existing standard. And if switching made sense for them, they might well conclude that other users would also come to the same conclusion.

In this contrary vision, "the future product" of the standard system would include a decline in market share, and a loss of support, making the current product less attractive. This would not only make it a better idea to switch, but encourage users to switch sooner rather than later.

This may sound like a theoretical point, but it's the kind of process that underlies changes from one platform to another. For example, people are not willing to pay as much for PlayStations and PlayStation games once there is the perception that the market will switch to the PlayStation 2. The wary defer their buying decisions, and the new product casts a pall over the old one.

Similarly, the real value of Lotus Development's office software fell from £500 to about £5 once it became clear that Microsoft Office was the winning platform. The existing product didn't change by a single byte, but the future product suddenly had a negative value. The cost of switching to Microsoft Office was far less than the possible future cost of having to retrieve important data from dead file formats.

This is one of the most difficult problems for computer companies to handle, because even if they are wrong, perceptions can become self-fulfilling prophecies. One of the many reasons that Apple lost the desktop wars was the conclusion arrived at by every rational person: that Apple was bound to lose. One day, Microsoft could face a similar problem with GNU/Linux. So not only must it maintain Windows' dominance, it has to maintain the perception of future dominance. In this case, of course, the answer is Microsoft.net.

· Note: this column belatedly answers the main objections that readers emailed in response to the earlier one about "the whole product" concept on January 25. Like that, it draws heavily on work by Stan Liebowitz, Professor and Associate Dean of the School of Management, University of Texas at Dallas, and others, including Geoffrey Moore, managing director of the Chasm Group. All the mistakes are, as usual, my own.