A look at the private sector reveals that software debacles are routine. And the more ambitious the project, the higher the odds of disappointment. It may not be much consolation to taxpayers, but the F.B.I. has a lot of company. Software hell is a very crowded place.

Consider Ford Motor Company’s ambitious effort to write new software for buying supplies. Begun in 2000, the goal of the project, code-named Everest, was to replace Ford’s patchwork of internal purchasing systems with a uniform system that would run over the Internet. The new software was supposed to reduce paperwork, speed orders and slash costs. But the effort sank under its own complexity. When it was rolled out for testing in North America, suppliers rebelled; according to Automotive News, many found the new software to be slower and more cumbersome than the programs it was intended to replace. Last August, Ford abandoned Everest amid reports that the project was as much as $200 million over budget.

A McDonald’s program called Innovate was even more ambitious – and expensive. Started in 1999 with a budget of $1 billion, the network sought to automate pretty much the entire fast-food empire. Software systems would collect information from every restaurant – the number of burgers sold, the speed of customer service, even the temperature of the oil in the French fry vats – and deliver it in a neat bundle to the company’s executives, who would be able to adjust operations moment by moment.

Or so it was promised. Despite the grand goals, the project went nowhere. In late 2002, McDonald’s killed it, writing off the $170 million that had already been spent.

Research by the Standish Group, a software research and consulting firm, illustrates the troubled fates of most big software initiatives. In 1994, researchers found, only 16 percent were completed on time, on budget and fulfilling the original specifications. Nearly a third were canceled outright, and the remainder fell short of their objectives. More than half of the cost overruns amounted to at least 50 percent of the original budget. Of the projects that went off schedule, almost half took more than twice as long as originally planned. A follow-up survey in 2003, however, showed that corporate software projects were doing better; researchers found that the percentage of successful projects had risen to 34 percent.

(from The New York Times: Does Not Compute)

A minor news item in the larger scheme of things, but the very fact the maneuver surprised me means I haven’t internalized what a fad it is to donate code. It doesn’t appear to be a full VoiceXML stack, nor is it the acclaimed ViaVoice desktop engine, but it’s a start nonetheless…

The New York Times > Technology > Speech Code From I.B.M. to Become Open Source

After decades of research and development, speech recognition is moving toward mainstream use. Advances in statistical modeling, pattern-matching algorithms and processing power have enabled speech recognition to interpret a far broader vocabulary of words and phrases than in the past, though glitches remain.

The software for speech-recognition applications once had to be custom built, but now packages of reusable and standardized tools are becoming available. The speech software can now be added to a Web application so that programmers can use familiar tools and need little additional training.

“This whole speech world is going in the same direction as the rest of the information technology industry, and that should drastically reduce the cost of building speech applications,” said Mark Plakias, an analyst at Opus Research.

I.B.M. is donating code that it estimates cost the company $10 million to develop. One collection of speech software for handling basic words for dates, time and locations, like cities and states, will go to the Apache Software Foundation. The company is also contributing speech-editing tools to a second open-source group, the Eclipse Foundation.