The Second 90%



A few years back I read a long lost article for indy game devs, that you should:

"Take your worst case estimate, double it and use that as your best case estimate."

This puts it mildly, but humans are for the most part really bad at estimating effort.

The standard project management aproach to managing large tasks with a high degree of uncertainty is to break down the large task into smaller and smaller tasks. The idea is that each task can be reasoned about and thus estimated to a higher degree of certainty. This is true, but...

In many project management software, you can show single task progress in percent. Unfortunately, almost all tasks in progress are pegged at 90% and the software will helpfully show nice charts. But this is misleading, since a task that is not done, is not done. A task has exactly 3 relevant states:

If you have a kanban, any task that is not in the done column is in progress. The task may be in testing or acceptance, but unless it is actually approved, there is a chance that the developer did miss some edge case or misunderstood some requirement.

This is further aggravated, by the circumstances that without a doubt, we tend to gloss over the small but relevant details.

For example actually useful software needs:

This is what separates a production ready piece of software from a prototype.

Unfortunately, only a few of these tasks are thought about when doing the initial project plan and even seasoned project managers oversee at least half of these tasks. That is why many projects have months of QA before release.

This has prompted me to keep saying:

"Once you are done with the first 90%, you need to sit down and do the remaining 90%."

I don't recall where I picked it up (probably THE Wiki), but it is loosly based on the 90/10 rule by Tom Cargill of Bell Labs:

"The first 90 percent of the code accounts for the first 90 percent of the development time. The remaining 10 percent of the code accounts for the other 90 percent of the development time."

This has prompted me to see the end date on any larger project estimate as between half and two thirds of the actual delivery date and communicating the final delivery as such.