How do I get sufficient detail in planning and estimation when using TDD?

Question

When planning a 2-week iteration in the past I have taken a user story:

• Story: Rename a file

And broken it into tasks which were then estimated in hours:

• Story: Rename a file
  • Task: Create Rename command (2h)
  • Task: Maintain a list of selected files (3h)
  • Task: Hook up to F2 key (1h)
  • Task: Add context menu option (1h)

I would then pick a task to work on, and track the time spent on working on it. I would then repeat the process with another task. At the end of the iteration I could look at the time spent on each task, compare it to the estimation and use this information to improve future estimations.

When working entirely driven by tests, the only work that is clearly defined ahead of time are the acceptance tests that kicks off development and, on a user story that covers a large amount of work, the scope of an acceptance test can be too broad to give a good estimation.

So I can take a guess at the tasks that will end up being completed (as before), but the time spent on them is far more difficult to track as the tests make you work in tiny vertical slices, often working on a bit of each task at the same time.

Are there any techniques I could employ to give more detailed estimations and accurately track time when performing TDD? I am using TargetProcess which encourages splitting of user stories into tasks as outlined above, so keeping things in that format would be helpful.

Answer 1

In agile both tasks and estimates are fluid things that change all the time.

So you might start with (bear in mind that these are very loose examples):

• Story: Rename a file
  • Task: Investigate Problem and break down (0d/5d)

First developer/s pick up that task and break it down as they go:

• Story: Rename a file
  • Task: Investigate Problem and break down (4h/complete)
  • Task: 1st part (0d/2d)
  • Task: 2nd part (0d/3d)

Then as they progress these updates get more accurate. New tasks get added and split out as they emerge:

• Story: Rename a file
  • Task: Investigate Problem and break down (4h/complete)
  • Task: 1st part (4h/7h)
  • Task: 2nd part (1h/20h)
  • Task: new task realised while working on x (0h/5h)

It doesn't matter whether you are using Scrum, Crystal, XP, TDD or any other agile variant - they all rely on fluid estimations.

The fact is that you never know how long something is going to take - you just take your best guess and revise it every day. You'll never get a process where there are no surprises, but with agile you manage their impact.

For instance suppose something nasty comes up:

• Story: Rename a file
  • Task: Investigate Problem and break down (4h/complete)
  • Task: 1st part (10h/complete)
  • Task: 2nd part (10h/3h)
  • Task: new task realised while working on x (3h/1h)
  • Task: resolve messy issue found while working on y (0h/5d)

The story is now taking longer than expected, but everyone knows about it and knows why and you can handle it.

Your tasks and their estimates are constantly changing as the work gets done. A burndown chart is a good indicator of how much is left to do across the team. I wouldn't bother with velocity, but if you do it compares the 'amount done' between different iterations, giving you some idea of a project's momentum. Velocity only works when you have very consistent iteration lengths, team size and classification (size, difficulty, complexity etc) of stories, so I'd start with getting the burndown right each iteration and then move on to it.

Answer 2

We at TargetProcess use simpler tasks for stories:

Story: Rename a file

• Task: Specification (2h)
• Task: Development (14h)
• Task: Testing (6)
• Task: User Documentation update (2h)

If Development task takes more than 16 hrs, it is a sign to split it to several smaller tasks. In fact we don't usually create tasks with less than 2-3 h duration.