Chaos

If you have never read the original Chaos report from the Standish Group you should stop looking at this screen and immediately read it. When you have finished, come back here and read on.

For those of you who read the report sixteen years ago and have forgotten it, this report, among other things, identifies reasons for failure of software projects. The original report in 1994 showed 16% of software projects succeeded, 53% were completed but with cost and time overruns and reduced deliverables and 31% were complete failures. In 2004 the Standish Group reported that this had improved to 29% succeeded and only 18% failed.

Yikes!!! In 10 years, as an industry, we have gone from 31% total failure to 18% total failure. I am really happy that we are not building airplanes!

One of the primary reasons for the limited improvement of the failure rate that we have been able to achieve is the continued long project timelines. The Chaos report states, “Research at The Standish Group also indicates that smaller time frames… will increase the success rate.” On my scale, a project with a timeline of more than a year has almost no chance of success, while a project with a timeline of a few weeks has a very good chance of success.

What about a data warehouse project? Usually these are measured in months, often exceeding a year. How is the success rate? According to Bill Inmon’s estimate there is a failure rate of 70% – 80%. Other studies have reported failure rates as high as 90%. Given the overall failure rate of software projects from the Standish Group, it seems like data warehouse projects are the primary source of project failures in the IT industry.

Of course, one can respond that building a data warehouse is a big job and requires a lot of time. Where is most of the time spent? Requirements definition, schema design, data normalization, optimizing indexing strategy and designing all of the secondary structure needed for access to information.

Now, I would like to challenge the other database vendors to participate in benchmarks where these factors are evaluated rather than the essentially meaningless TPC-H tests. I think most people would agree that any success is better than a large failure and these are the factors that will determine whether your data warehouse project will need sixty days to succeed or sixty weeks to fail.