InformationWeek
Defining The Business Value Of Technology
January 18, 1999
 Print this story |
continued...page 3 of 3
KnowledgePlan doesn't have an obvious way to estimate an entire year 2000 project. It does have a Major System category, which I used as a proxy. KnowledgePlan provides an estimate for the project-management overhead associated with a development project.
KnowledgePlan also doesn't include an estimate within its year 2000 model for putting an application into production. Yet the Hierarchy view, available after completing an estimate, does include a set of Delivery tasks, including installation, deployment, and user training. Evidently KnowledgePlan's Y2K project examples in the knowledge base don't include implementation.
KnowledgePlan allows more time than Cost Xpert for creating a test strategy; test cases; and unit-, system-, and integration testing--about 46% of the overall work effort. KnowledgePlan also includes a quality-assurance plan as part of the planning tasks, but it fails to include acceptance testing.
KnowledgePlan lets you tinker almost endlessly with the estimate produced automatically by the wizard. This can make a project estimate fit your environment more closely, but it may reduce its accuracy.
KnowledgePlan uses six sizing methods: four metrics (including lines of code and function points), components (a method for roughly estimating the number of code chunks added or modified), and by analogy with similar projects. But it won't let you use more than one method at the same time. However, you can create multiple versions of a project using different sizing methods and compare the results.
Once you make your choices, the project wizard generates a Gantt chart with work breakdown structures, schedule, and effort estimates. The estimate can then be exported to Microsoft Project, or managed within KnowledgePlan, which has a lot of project-management features.
The Y2K project estimate was 52 months--unacceptable, since the finish date would have been March 2003. It was also estimated at 34 people and more than 38,000 days of work. The coding portion was estimated at 42.5 full-time employees working for 10,600 worker-days. That's about 16.5 lines of code per day (175,000 lines of the 700,000 were to be fixed)--one-third of a line per person each day.
I refined many of the estimation variables, trying to obtain a more reasonable schedule. For example,
I raised the level of developer experience with the system, language, and platform. This lowered the estimate of full-time employees to 25 but left duration unchanged. The coding portion showed 35 people working for 5,500 worker-days--almost 32 lines of code per day, still less than one line of code per person each day.
KnowledgePlan assigns too much duration to planning, though it estimates what seems like the right number of full-time employees (small numbers for any one planning task).
Obviously, the estimates could be further refined by hand, either within KnowledgePlan or by exporting the results into a project-management program. But this defeats the purpose of an estimating tool.
KnowledgePlan is a more sophisticated estimating tool than Cost Xpert, but the complexity seems to bloat the estimates it produces.
As soon as changes were made to task information in the Gantt chart, the numbers started changing in ways that didn't make sense. For example, the test plan started at five full-time employees for 21.48 worker-months, but after some changes to try to reduce the effort or duration, the original numbers wouldn't come back no matter what I tried.
There was no cost estimate in the results. On checking the tasks on the Gantt chart, it appeared that the knowledge base had no costs for any of them.
KnowledgePlan produces fairly clear and useful reports, but some puzzling things happened with the Task Category Overview report. For example, it showed start and end dates, but the Gantt chart didn't. A further examination showed some strange work overlap. In the planning phase, the test planning didn't start until almost five months after the end of the development planning, and quality-assurance planning didn't start until about four months after the start of test planning.
As long as you understand that this is a complex tool with some estimating quirks, and therefore that any estimates will need refining, KnowledgePlan is certainly a useful tool for estimating any size project. Its close integration with Microsoft Project makes it a natural for managers requiring an estimating aid.
Benefits Of Estimating Tools
Why should you use one of these estimating tools? They help speed and automate estimating while providing training in estimating along with a framework for understanding how to make estimates. It's unlikely you'll get estimates that fit your environment until you adjust the tools, but both allow you to do that.
Phillip Gordon is a system architect for a major West Coast financial services firm.
return to page 1, 2
Back to Labs
Send Us Your Feedback
Top of the Page
